[gentoo-user] Recommended CPU and cooling system

[gentoo-user] Recommended CPU and cooling system

[whiteman808]

Hello,

I'm saving money for building PC. I'm going to do a lot of compiling.
I'll use this PC as binary package server for multiple machines in my home network.
I'll also host many virtual machines for various purposes on this PC.

Target budget for only PC (not peripherals, additional devices) is 5000 EUR.

Do you think AMD Ryzen Threadripper Pro 5965WX is a good choice for
doing lots of compiling stuff on Gentoo?

Do you recommend to use all in one cooling or invest into custom
water cooling system? Is it true that heat cooling isn't a good idea in
my case?

whiteman808

Re: [gentoo-user] Recommended CPU and cooling system

[Alexandru N. Barloiu]

On Thu, 2025-07-24 at 02:38 +0200, whiteman808 wrote:
> Hello,
> 
> I'm saving money for building PC. I'm going to do a lot of compiling.
> I'll use this PC as binary package server for multiple machines in my
> home network.
> I'll also host many virtual machines for various purposes on this PC.
> 
> Target budget for only PC (not peripherals, additional devices) is
> 5000 EUR.
> 
> Do you think AMD Ryzen Threadripper Pro 5965WX is a good choice for
> doing lots of compiling stuff on Gentoo?
> 
> Do you recommend to use all in one cooling or invest into custom
> water cooling system? Is it true that heat cooling isn't a good idea
> in
> my case?
> 
> whiteman808


I would like to make 2 points. 

First about cooling. Don't believe the hype. They are both the same
thing. Both cooling solutions are a mixture of a liquid and air. Only
difference is the pump and the amount of liquid. But even water
solutions use a fan to cool down the liquid, and even normal coolers
have pipes with liquid in them that do exactly the same thing.
Personally I am partial to Noctua NH-D15. Its a monstrous cooler. And
its safe from a liquid perspective. That thing can cool anything. The
downside of it is that it doesn't fit in most cases. its 175 mm high. 

Second point. Number of cores is fine. Prolly dont need that many. What
will be a problem will be the memory. You need around 1GB per thread of
memory for normal C, and about 4GB of ram per thread for C++ stuff. So
for 48 threads multiplied by 4 = 192 GB of ram. For something like
webkit-gtk or chromium or spiderweb or rust or any other number of
packages. GCC too. Especially if you put LTO on as well.

I would invest in a CPU with less cores/threads (maybe 9950x3d = 36
threads) in favor of maxing RAM. In which case that is about 96GB. With
a 870x chipset board. Nice balance. And have to keep in mind that your
CPU alone is half the money you want to invest. But you will need a
bunch of RAM which is very expensive. A mobo. A cooler. A power supply.
Its a lot to start with a cpu that is already half your budget. Go
lower imho. Good luck and enjoy your machine. 

axl

Re: [gentoo-user] Recommended CPU and cooling system

[Dale]

Alexandru N. Barloiu wrote:
> On Thu, 2025-07-24 at 02:38 +0200, whiteman808 wrote:
>> Hello,
>>
>> I'm saving money for building PC. I'm going to do a lot of compiling.
>> I'll use this PC as binary package server for multiple machines in my
>> home network.
>> I'll also host many virtual machines for various purposes on this PC.
>>
>> Target budget for only PC (not peripherals, additional devices) is
>> 5000 EUR.
>>
>> Do you think AMD Ryzen Threadripper Pro 5965WX is a good choice for
>> doing lots of compiling stuff on Gentoo?
>>
>> Do you recommend to use all in one cooling or invest into custom
>> water cooling system? Is it true that heat cooling isn't a good idea
>> in
>> my case?
>>
>> whiteman808
>
> I would like to make 2 points. 
>
> First about cooling. Don't believe the hype. They are both the same
> thing. Both cooling solutions are a mixture of a liquid and air. Only
> difference is the pump and the amount of liquid. But even water
> solutions use a fan to cool down the liquid, and even normal coolers
> have pipes with liquid in them that do exactly the same thing.
> Personally I am partial to Noctua NH-D15. Its a monstrous cooler. And
> its safe from a liquid perspective. That thing can cool anything. The
> downside of it is that it doesn't fit in most cases. its 175 mm high. 
>
> Second point. Number of cores is fine. Prolly dont need that many. What
> will be a problem will be the memory. You need around 1GB per thread of
> memory for normal C, and about 4GB of ram per thread for C++ stuff. So
> for 48 threads multiplied by 4 = 192 GB of ram. For something like
> webkit-gtk or chromium or spiderweb or rust or any other number of
> packages. GCC too. Especially if you put LTO on as well.
>
> I would invest in a CPU with less cores/threads (maybe 9950x3d = 36
> threads) in favor of maxing RAM. In which case that is about 96GB. With
> a 870x chipset board. Nice balance. And have to keep in mind that your
> CPU alone is half the money you want to invest. But you will need a
> bunch of RAM which is very expensive. A mobo. A cooler. A power supply.
> Its a lot to start with a cpu that is already half your budget. Go
> lower imho. Good luck and enjoy your machine. 
>
> axl
>
>


OP, I tend to agree with Axl on the cooling part.  I've never used a
water cooling system.  I've always bought air cooling systems.  I went
with a Thermalright Peerless Assassin 120 SE and it cools pretty well
for my AMD AM4 Ryzen 7 5800X.  I won't say that the Noctua won't be
better, especially on a AM5 that is top tier.  It is a lot bigger and
would likely cool better.  My CPU runs at about 190F with the new way
they measure temps.  The old way of measuring it runs at about 115F.  As
best as I can tell, it runs cool enough that it isn't throttling any,
not real sure on how to check that tho.  That said, I wish the new way
of measuring was 15F cooler.  That Noctua may do that, may not.  If you
can fit that Noctua in your budget, and case, I'd buy that tho.  May
cost more but other than that, no downside to it.  Larger is almost
always better.  May have better fans too. 

As to cores, Axl makes a good point on threads and memory.  I suspect
that Libreoffice, that qtweb package, Firefox, Chrome and other large
packages will fail with larger numbers of threads.  Most all of them are
designed to compile in parallel pretty well but I bet there is a limit
somewhere.  Then you run into the memory limits as well.  I have 128GBs
here.  I was going to go with 64GBs for a while but when one stick went
bad, I had to have memory to use while doing the warranty swap.  So, I
bought another 64GB kit.  Once replacement sticks came in, I was at
128GBs.  If you go the Threadripper path, you will see limits kick in
somewhere.  Memory is likely a good bet unless you can have double or
more than what I have here.  I suspect you will find limits even on the
number of threads that can be compiled before failing too.  You might be
better with a really fast AM5 like Axl mentioned.  If you can afford a
mobo that supports 256GB of memory or more, it will be pricey but that
will raise the number of threads pretty high and also put you in a good
place for the system to be usable long term as far as power and memory
goes.  Even if I could afford it, I think for compiling, the
Threadripper might not be the best path.  Maybe some web searches will
turn up someone who has one and what they have seen it capable of
compiling wise.

Another good idea to look into.  The case.  My first case was a generic
type thing but still considered a full tower case.  I added fans
including cutting holes on the side.  I learned a lot about case cooling
with that old thing.  It had a AMD 2500+ CPU, the mid 2000's tech wise. 
My second rig has a Cooler Master HAF-932.  Pretty massive and it has a
large fan on the side.  I can tell you  that side fan improves cooling
for things like video cards and the larger controller chips on the mobo,
CPU to, a lot.  My current rig, because of the need for hard drive bays,
is a Fractal Design Define 7 XL.  It has two setups.  One that has less
hard drive bays and water cooling or lots of hard drive bays and air
cooling.  It is a very nice case.  I do wish it had a side fan tho. 
Again, that side fan works miracles.  Even if the case I have isn't for
you, you may find one of theirs to be nice or at least give you ideas on
something similar in another brand.  I sometimes wish I could have got a
more modern version of the Cooler Master for my new rig with lots of
hard drive bays and the side fan.  They just don't make those.  That
said, the case cools well and the mobo has lots of fan control.  My
current setup has like 6 or so fans.  If you can find a case you like
with a side fan, it is good to have that side fan. 

On the power supply, the best advice I can give, buy a very reputable
brand that has a low failure rate and if it fails, it fails safely. 
There used to be a thread on Overclockers forum where they tested power
supplies, usually to failure, and rated them.  They also took them apart
and looked at the quality of the components as well.  Most of the
cheaper things, failed and were not worth repairing.  Some of the really
expensive ones with a good reputation are pricey for a reason.  They are
built with high quality components and with a method of handling failure
that doesn't take your mobo and such with it.  Make sure what you get is
a really good one and has a reputation of failing in a safe way.  It
could save you a CPU, mobo etc later on. 

When I built my current rig, I was originally planning on a AM5 system. 
The lack of PICe slots moved me toward a AM4 based system.  You need to
look into that as well to make sure you can expand things if needed. 
AM5 systems are limited in that way.  AM4 isn't a lot better.  The days
of having 7 or more PCIe slots are pretty much gone.  Also, some PCIe
slots are shared with m.2 slots sometimes.  In other words, you can't
use both at the same time.  If a m.2 stick is installed, a PCIe slots is
disabled.  You already have less PCIe slots and lose more with m.2
sticks in use.  That may or may not affect you but it is worth
mentioning.  It's just something you need to keep in mind. 

This is a lengthy reply but I hope it will give you some ideas and some
things to look out for based on real world use.  A good case with good
cooling, good CPU cooler and a good power supply is the important parts
to get started.  I hope it will help you put together a nice system that
you can get good service out of long term.  I might add, I like my old
Cooler Master case well enough, if it had more hard drive bays, I would
have put my new system in it.  Again, that side fan is impressive on
cooling. 

Please, post back what you end up with.  I wouldn't mind pics sent to me
offlist.  If you have a place to upload them, share a link to pics for
us all to drool over.  ROFL 

Best of luck.  :-D 

Dale

:-)  :-) 

Re: [gentoo-user] Recommended CPU and cooling system

[Alexandru N. Barloiu]

On Wed, 2025-07-23 at 22:42 -0500, Dale wrote:
> Alexandru N. Barloiu wrote:
> 


btw. dont be surprised if AMD stuff doesn't work as well as Intel
stuff. 

I am not being a hater. Just being a realist. I was an Intel fan boy.
Over time started to hate Intel. Made my first AMD system... and I have
to say... not exactly loving it. 

Some simple examples. Until very recently, kernel 6.16... which is now
in RC7 stage, no cpu sensor. Kind of a problem. Intel cpus always had
coretemp. Not saying they are not coming up, one by one. But you spend
like 5000 or however much each of us spends... you expect to at least
get a cpu temperature sensor that works. 

Other example. Intel platform. Qemu. No problem. AMD. Weird ACPI table.
which some setups, if you dont use modern enough edk2 firmware, takes
45 seconds for the damn VM to start. 

Other example. No XMP memory profiles. Will be a pain to actually OC
memory. On intel platforms you just select the xmp profile and you are
done. Not on AMD. 

And finally another example. Not sure what the threadripper has. But my
9950x3d has 8 SMT high performance cores. and 8 SMT power saving cores.
but the kernel has no idea which is which. if you set your system on
powersaving or balanced or high performance... that means nothing to
the system. to be clear. 0->7 are high performance. 7->15 powersaving.
and 16->23 high performance siblings (smt - amd version of
hyperthreading), and 24->31 powersaving siblings. but again. the OS is
completely unaware of this. 

just some examples. am sure the amd folks will catch up with the
software. but just to be clear. lots of stuff dont work out of the box.
and have to dig really deep for each individual little thing. 

and some things just straight out dont work. and nobody tells you ahead
of the purchase that they dont work. like for instance. my asus 870X
creative mobo came very high recommended. but nobody told me you can't
install windows 10 on it. and nobody told me that wifi/bluetooth
doesn't work in linux. it will at some point. but it doesn't now. 

sorry for the rant. just things i found out when i got my first AMD
system. 

Re: [gentoo-user] Recommended CPU and cooling system

[Dale]

Alexandru N. Barloiu wrote:
> On Wed, 2025-07-23 at 22:42 -0500, Dale wrote:
>> Alexandru N. Barloiu wrote:
>>
>
> btw. dont be surprised if AMD stuff doesn't work as well as Intel
> stuff. 
>
> I am not being a hater. Just being a realist. I was an Intel fan boy.
> Over time started to hate Intel. Made my first AMD system... and I have
> to say... not exactly loving it. 
>
> Some simple examples. Until very recently, kernel 6.16... which is now
> in RC7 stage, no cpu sensor. Kind of a problem. Intel cpus always had
> coretemp. Not saying they are not coming up, one by one. But you spend
> like 5000 or however much each of us spends... you expect to at least
> get a cpu temperature sensor that works. 
>
> Other example. Intel platform. Qemu. No problem. AMD. Weird ACPI table.
> which some setups, if you dont use modern enough edk2 firmware, takes
> 45 seconds for the damn VM to start. 
>
> Other example. No XMP memory profiles. Will be a pain to actually OC
> memory. On intel platforms you just select the xmp profile and you are
> done. Not on AMD. 
>
> And finally another example. Not sure what the threadripper has. But my
> 9950x3d has 8 SMT high performance cores. and 8 SMT power saving cores.
> but the kernel has no idea which is which. if you set your system on
> powersaving or balanced or high performance... that means nothing to
> the system. to be clear. 0->7 are high performance. 7->15 powersaving.
> and 16->23 high performance siblings (smt - amd version of
> hyperthreading), and 24->31 powersaving siblings. but again. the OS is
> completely unaware of this. 
>
> just some examples. am sure the amd folks will catch up with the
> software. but just to be clear. lots of stuff dont work out of the box.
> and have to dig really deep for each individual little thing. 
>
> and some things just straight out dont work. and nobody tells you ahead
> of the purchase that they dont work. like for instance. my asus 870X
> creative mobo came very high recommended. but nobody told me you can't
> install windows 10 on it. and nobody told me that wifi/bluetooth
> doesn't work in linux. it will at some point. but it doesn't now. 
>
> sorry for the rant. just things i found out when i got my first AMD
> system. 
>
>


That's the reason why when I build, I don't build with the latest and
greatest CPU, video card and other hardware.  New stuff has always had
lagging support in Linux.  When something new comes out, Microsoft is in
on the details early, likely because they pay a huge sum of money for
the info.  So, when some new piece of hardware comes out, they get it
first.  In Linux, I've read that some things have to be reverse
engineered which is time consuming.  Think about document scanners and
printers.  Some scanners and printers are still not supported and a lot
only have limited support.  Some of those have been out for years. 

I'm not to surprised that you have ran into this.  Since I built a rig
that technology wise was already a year or two old, everything worked
out of the box for me.  CPU temps and all.  Mine seems to recognize my
cores properly, as best I can tell anyway.  Gkrellm isn't complaining
and neither is htop. 

The one thing that freaked me out, CPU temps.  They have a new and
improved sensor that reads from the silicone wafer itself, deep inside
it seems.  It responds faster and is likely more accurate but it is much
higher temps than the old temp that was of the die, right under where
the CPU cooler sits it seems.  When I first saw 190F temps, I kinda
freaked out a bit.  I thought I was trying to cook a egg with my CPU.  I
was checking that the fans were running and all that.  I think it was
Rich that posted a explanation.  As I said, the temp is likely more
accurate but still, I was not expecting that. 

When I build a new rig several years from now, I might build a AM5 that
is either available now or will be shortly.  Even my first rig was built
with things that had been released for close to a year.  It was still
plenty fast for the time tho.  I even named it Smoker.  My next rig was
named Fireball.  I couldn't think of a good name for this rig that would
line up with those two names so I went with Gentoo-1.  I guess next rig
will be Gentoo-2.  ;-)

It is sometimes hard to tell if everything is supported.  Linux is a
great community but someone has to be the first to buy something and see
if things work.  Once bought, you kinda stuck with it.  It takes a
while.  Then you have to figure out which forum or other website where a
person is describing what works and what doesn't.  Then how dated that
info is.  I find that usually after a year, it is likely supported as
far as CPU, mobo and such.  Other stuff like printers may take longer. 

It is a problem tho.  It's almost like a leap of faith when buying newly
released hardware.  Rant on.  This may help the OP too.  It may be a
good idea to mention that some things might not be well supported yet. 

Dale

:-)  :-) 

Re: [gentoo-user] Recommended CPU and cooling system

[Alexandru N. Barloiu]

On Thu, 2025-07-24 at 00:10 -0500, Dale wrote:
> Alexandru N. Barloiu wrote:
> > On Wed, 2025-07-23 at 22:42 -0500, Dale wrote:
> > > Alexandru N. Barloiu wrote:
> > > 
> > 
> > btw. dont be surprised if AMD stuff doesn't work as well as Intel
> > stuff. 
> > 
> > I am not being a hater. Just being a realist. I was an Intel fan
> > boy.
> > Over time started to hate Intel. Made my first AMD system... and I
> > have
> > to say... not exactly loving it. 
> > 
> > Some simple examples. Until very recently, kernel 6.16... which is
> > now
> > in RC7 stage, no cpu sensor. Kind of a problem. Intel cpus always
> > had
> > coretemp. Not saying they are not coming up, one by one. But you
> > spend
> > like 5000 or however much each of us spends... you expect to at
> > least
> > get a cpu temperature sensor that works. 
> > 
> > Other example. Intel platform. Qemu. No problem. AMD. Weird ACPI
> > table.
> > which some setups, if you dont use modern enough edk2 firmware,
> > takes
> > 45 seconds for the damn VM to start. 
> > 
> > Other example. No XMP memory profiles. Will be a pain to actually
> > OC
> > memory. On intel platforms you just select the xmp profile and you
> > are
> > done. Not on AMD. 
> > 
> > And finally another example. Not sure what the threadripper has.
> > But my
> > 9950x3d has 8 SMT high performance cores. and 8 SMT power saving
> > cores.
> > but the kernel has no idea which is which. if you set your system
> > on
> > powersaving or balanced or high performance... that means nothing
> > to
> > the system. to be clear. 0->7 are high performance. 7->15
> > powersaving.
> > and 16->23 high performance siblings (smt - amd version of
> > hyperthreading), and 24->31 powersaving siblings. but again. the OS
> > is
> > completely unaware of this. 
> > 
> > just some examples. am sure the amd folks will catch up with the
> > software. but just to be clear. lots of stuff dont work out of the
> > box.
> > and have to dig really deep for each individual little thing. 
> > 
> > and some things just straight out dont work. and nobody tells you
> > ahead
> > of the purchase that they dont work. like for instance. my asus
> > 870X
> > creative mobo came very high recommended. but nobody told me you
> > can't
> > install windows 10 on it. and nobody told me that wifi/bluetooth
> > doesn't work in linux. it will at some point. but it doesn't now. 
> > 
> > sorry for the rant. just things i found out when i got my first AMD
> > system. 
> > 
> > 
> 
> 
> That's the reason why when I build, I don't build with the latest and
> greatest CPU, video card and other hardware.  New stuff has always
> had
> lagging support in Linux.  When something new comes out, Microsoft is
> in
> on the details early, likely because they pay a huge sum of money for
> the info.  So, when some new piece of hardware comes out, they get it
> first.  In Linux, I've read that some things have to be reverse
> engineered which is time consuming.  Think about document scanners
> and
> printers.  Some scanners and printers are still not supported and a
> lot
> only have limited support.  Some of those have been out for years. 
> 
> I'm not to surprised that you have ran into this.  Since I built a
> rig
> that technology wise was already a year or two old, everything worked
> out of the box for me.  CPU temps and all.  Mine seems to recognize
> my
> cores properly, as best I can tell anyway.  Gkrellm isn't complaining
> and neither is htop. 
> 
> The one thing that freaked me out, CPU temps.  They have a new and
> improved sensor that reads from the silicone wafer itself, deep
> inside
> it seems.  It responds faster and is likely more accurate but it is
> much
> higher temps than the old temp that was of the die, right under where
> the CPU cooler sits it seems.  When I first saw 190F temps, I kinda
> freaked out a bit.  I thought I was trying to cook a egg with my
> CPU.  I
> was checking that the fans were running and all that.  I think it was
> Rich that posted a explanation.  As I said, the temp is likely more
> accurate but still, I was not expecting that. 
> 
> When I build a new rig several years from now, I might build a AM5
> that
> is either available now or will be shortly.  Even my first rig was
> built
> with things that had been released for close to a year.  It was still
> plenty fast for the time tho.  I even named it Smoker.  My next rig
> was
> named Fireball.  I couldn't think of a good name for this rig that
> would
> line up with those two names so I went with Gentoo-1.  I guess next
> rig
> will be Gentoo-2.  ;-)
> 
> It is sometimes hard to tell if everything is supported.  Linux is a
> great community but someone has to be the first to buy something and
> see
> if things work.  Once bought, you kinda stuck with it.  It takes a
> while.  Then you have to figure out which forum or other website
> where a
> person is describing what works and what doesn't.  Then how dated
> that
> info is.  I find that usually after a year, it is likely supported as
> far as CPU, mobo and such.  Other stuff like printers may take
> longer. 
> 
> It is a problem tho.  It's almost like a leap of faith when buying
> newly
> released hardware.  Rant on.  This may help the OP too.  It may be a
> good idea to mention that some things might not be well supported
> yet. 
> 
> Dale
> 
> :-)  :-) 



to be fair... the system i love most is a tired old x58 chipset. intel
950 cpu. 32GB of slow ass ram (think 1600). but has an intel 750 nvme.
and a modern nvidia 980. honestly... its just fine for most things
(HD). can't do 4k. i mean it can, but not well. but it does HD. 

if you push enough ram on your system, and a good ssd, rest dont really
matter. 

further more. its hard to tell. just saying. but its hard to tell. if
you have like intel 750 or 905 optane storage (a good ssd, these 2 are
legendary). if you have that. and if you have ram to the max that your
platform can support. 32GB. 64. 128. whatever you can. and you have a
decent gpu. 980 still decent today. u cant actually tell that intel 950
cpu, one of the first is bad. can't tell nvidia 980 is bad. can't tell
1600 ram is bad. 

and frankly, that machine is enough as a distcc server. funny thing
about distcc. :) it has a top bandwidth limit. somewhere between 8 and
16 threads you fill the entire gigabyte bandwidth. 

I will say this. I had a 9980xe before with 128GB of ram. its a
monster. its also old. 2018 if i am not mistaken. 18 cores and 36
threads. my new 9950x3d only has 16 cores and 32 threads and only 96GB
of ram. so its not that much of an upgrade. old boosted speed was 4GHZ
(on 9980xe). on 9950x3d is 5.76. i get better single core performance.
1.76 more. that's about all i get. in 7 years. 

I will also say this. I kept the intel 750 and intel 905p optane. Both
are god tier storage. But the optate is got tier squared. If you find
any of these bad boys... go do:

# smartctl -a /dev/nvme0 | grep "Available Spare Threshold"

and if that is below 100 buy them. i mean for the optane. 

there's the scoop. 

both intel 750 and 900(5) are way way way way faster than any modern
nvme drives. modern nvme drives are optimized for single thread
performance. u got one thread, single file, great. 7000 mbps. u dont
need 7000 coz nothing reads anything that fast. you cant feed it into
anything. but whatever. most things... when you increase the number of
threads, and maybe its not one single big file, but multiple small
files, and maybe you want to read and write at the same time....

i mean you are buying a 48 thread machine. right?

well. most modern nvmes can't handle stuff after 8. they were never
designed to do so. 

intel 750 was. it was created from the start with the idea in mind that
the controller had to be absolutely able to do hundreds of threads at
the same time. 

problem with intel 750... its a perishable drive. it has amazing
performance. it reads and writes with 2200/1600. regardless of threads
or files or anything else. it simply doesn't care. its performance per
core is equal to the performance of however many cores you put into it.

but then there's the 900. and the 905. optane. they kept the
controller, the one that doesn't care about how many threads you are
doing, and they changed the storage. and they invented this thing.
xpoint storage. in terms of ssd/nvme storage... that stuff doesn't die.
while normal drives count their max life in terabytes. tens. or
hundreds. or thousands on the high end. OPTANE counts its max life in
petabytes. 

believe me. if you find any of these drives, and they are not spent,
get them. your computer becomes something else when you have one of
these bad boys. they go directly into pciexpress. 4 lanes. but my god
they dont resemble anything that is out there. and a system with one of
them is so different from a system that doesn't. especially when you
push. the controller intel built into those things... what parallelism?
breaks the space time continuum. 


changed ram. changed cpu. changed mobo. changed gpu. changed power
supply. but kept the 905p. i got a 9100 pro from samsung. guess that's
the new standard ssd. my optane still kicks it ass in random seek
read/write with as many threads as the cpu has. 

Re: [gentoo-user] Recommended CPU and cooling system

[Michael]

[-- Attachment #1: Type: text/plain, Size: 5383 bytes --]

On Thursday, 24 July 2025 01:53:22 British Summer Time Alexandru N. Barloiu 
wrote:
> On Thu, 2025-07-24 at 02:38 +0200, whiteman808 wrote:
> > Hello,
> > 
> > I'm saving money for building PC. I'm going to do a lot of compiling.
> > I'll use this PC as binary package server for multiple machines in my
> > home network.
> > I'll also host many virtual machines for various purposes on this PC.
> > 
> > Target budget for only PC (not peripherals, additional devices) is
> > 5000 EUR.
> > 
> > Do you think AMD Ryzen Threadripper Pro 5965WX is a good choice for
> > doing lots of compiling stuff on Gentoo?
> > 
> > Do you recommend to use all in one cooling or invest into custom
> > water cooling system? Is it true that heat cooling isn't a good idea
> > in
> > my case?
> > 
> > whiteman808
> 
> I would like to make 2 points.
> 
> First about cooling. Don't believe the hype. They are both the same
> thing. Both cooling solutions are a mixture of a liquid and air. Only
> difference is the pump and the amount of liquid. But even water
> solutions use a fan to cool down the liquid, and even normal coolers
> have pipes with liquid in them that do exactly the same thing.
> Personally I am partial to Noctua NH-D15. Its a monstrous cooler. And
> its safe from a liquid perspective. That thing can cool anything. The
> downside of it is that it doesn't fit in most cases. its 175 mm high.

The pros & cons between air and water cooling are well covered and commented 
on all over the interwebs.  There also a lot of tests and measurements to back 
claims of one Vs the other;  e.g.:

https://www.relaxedtech.com/reviews/noctua/nh-d15-versus-closed-loop-liquid-coolers/1

I broadly agree with Axl's comment on cooling.  For normal everyday desktop 
workloads an AIO water cooler will not provide any significant benefit 
compared to a similarly priced air cooler, other than lower noise levels at 
maximum performance.  However, the topic is rather nuanced and there are 
notable differences between cooling components and CPU/GPU power outputs.

First let's differentiate between AIO and custom water cooling systems.  
Although AIOs and air coolers are comparable, with AIOs having an edge both in 
higher temperature dissipation and lower noise at maximum power output, an 
expensive custom water cooling system will perform better than any air cooling 
system today, in terms of temperature differentials at extreme performance 
levels.  A custom water cooling system can also water-cool your GPU, which you 
may need if you're an avid gamer, engage in crypto-mining, or undertake heavy 
media transcoding.

If you overclock, or you tune your AMD's PBO to maximise CPU power usability, 
a water cooler will deliver better performance during a continuous maximum CPU 
power draw.  When you are cranking up all cores on protracted monster emerges, 
like chromium/qtwebengine, then liquid cooling will have a measurable edge.

On the other hand, a water cooling system has the major disbenefit of a 
limited life span.  AIOs come with a 3, 5, or 6 years warranty for a good 
reason.  The components they use have a higher probability of failing soon 
after the warranty expires!  They may fail gradually, e.g. by their puny water 
pump becoming noisier, or they may fail suddenly and without warning.  They 
could fail and spring a leak all over your expensive MoBo, PCIe ports, M.2 SSD 
and graphics card.  An air cooler may also fail, but without collateral damage 
the cost of just replacing a cooling fan and carrying on where you left will 
be much much much lower.


> Second point. Number of cores is fine. Prolly dont need that many. What
> will be a problem will be the memory. You need around 1GB per thread of
> memory for normal C, and about 4GB of ram per thread for C++ stuff. So
> for 48 threads multiplied by 4 = 192 GB of ram. For something like
> webkit-gtk or chromium or spiderweb or rust or any other number of
> packages. GCC too. Especially if you put LTO on as well.
> 
> I would invest in a CPU with less cores/threads (maybe 9950x3d = 36
> threads) in favor of maxing RAM. In which case that is about 96GB. With
> a 870x chipset board. Nice balance. And have to keep in mind that your
> CPU alone is half the money you want to invest. But you will need a
> bunch of RAM which is very expensive. A mobo. A cooler. A power supply.
> Its a lot to start with a cpu that is already half your budget. Go
> lower imho. Good luck and enjoy your machine.
> 
> axl

I tend to go by the old muscle car saying, "... there's no substitute for 
cubic inches!"  On your PC the cubic inches translate to cores x frequency.  
More cores and higher frequency = faster emerge.  However, as Axl warns the 
bottleneck becomes the amount of RAM needed to support the parallelism of your 
CPU cores.

So what to do?  I tend to opt for the higher amount of cores commensurate with 
a proportionate increase in price, which means I won't buy the latest and 
greatest advertised CPU.  I also wait until the price drops during sales.  
Sometimes, later MoBo models increase the amount of RAM they can power so 
checking what the MoBo can feed may help overcoming the RAM limitation. If 
you're building a system to keep in the long term I suggest you buy as many 
cores as you can afford with enough RAM for now, with plans to increase it 
later on when RAM prices tend to fall.

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [gentoo-user] Recommended CPU and cooling system

[Dale]

Alexandru N. Barloiu wrote:
> On Thu, 2025-07-24 at 00:10 -0500, Dale wrote:
>> Alexandru N. Barloiu wrote:
>>> On Wed, 2025-07-23 at 22:42 -0500, Dale wrote:
>>>> Alexandru N. Barloiu wrote:
>>>>
>>> btw. dont be surprised if AMD stuff doesn't work as well as Intel
>>> stuff. 
>>>
>>> I am not being a hater. Just being a realist. I was an Intel fan
>>> boy.
>>> Over time started to hate Intel. Made my first AMD system... and I
>>> have
>>> to say... not exactly loving it. 
>>>
>>> Some simple examples. Until very recently, kernel 6.16... which is
>>> now
>>> in RC7 stage, no cpu sensor. Kind of a problem. Intel cpus always
>>> had
>>> coretemp. Not saying they are not coming up, one by one. But you
>>> spend
>>> like 5000 or however much each of us spends... you expect to at
>>> least
>>> get a cpu temperature sensor that works. 
>>>
>>> Other example. Intel platform. Qemu. No problem. AMD. Weird ACPI
>>> table.
>>> which some setups, if you dont use modern enough edk2 firmware,
>>> takes
>>> 45 seconds for the damn VM to start. 
>>>
>>> Other example. No XMP memory profiles. Will be a pain to actually
>>> OC
>>> memory. On intel platforms you just select the xmp profile and you
>>> are
>>> done. Not on AMD. 
>>>
>>> And finally another example. Not sure what the threadripper has.
>>> But my
>>> 9950x3d has 8 SMT high performance cores. and 8 SMT power saving
>>> cores.
>>> but the kernel has no idea which is which. if you set your system
>>> on
>>> powersaving or balanced or high performance... that means nothing
>>> to
>>> the system. to be clear. 0->7 are high performance. 7->15
>>> powersaving.
>>> and 16->23 high performance siblings (smt - amd version of
>>> hyperthreading), and 24->31 powersaving siblings. but again. the OS
>>> is
>>> completely unaware of this. 
>>>
>>> just some examples. am sure the amd folks will catch up with the
>>> software. but just to be clear. lots of stuff dont work out of the
>>> box.
>>> and have to dig really deep for each individual little thing. 
>>>
>>> and some things just straight out dont work. and nobody tells you
>>> ahead
>>> of the purchase that they dont work. like for instance. my asus
>>> 870X
>>> creative mobo came very high recommended. but nobody told me you
>>> can't
>>> install windows 10 on it. and nobody told me that wifi/bluetooth
>>> doesn't work in linux. it will at some point. but it doesn't now. 
>>>
>>> sorry for the rant. just things i found out when i got my first AMD
>>> system. 
>>>
>>>
>>
>> That's the reason why when I build, I don't build with the latest and
>> greatest CPU, video card and other hardware.  New stuff has always
>> had
>> lagging support in Linux.  When something new comes out, Microsoft is
>> in
>> on the details early, likely because they pay a huge sum of money for
>> the info.  So, when some new piece of hardware comes out, they get it
>> first.  In Linux, I've read that some things have to be reverse
>> engineered which is time consuming.  Think about document scanners
>> and
>> printers.  Some scanners and printers are still not supported and a
>> lot
>> only have limited support.  Some of those have been out for years. 
>>
>> I'm not to surprised that you have ran into this.  Since I built a
>> rig
>> that technology wise was already a year or two old, everything worked
>> out of the box for me.  CPU temps and all.  Mine seems to recognize
>> my
>> cores properly, as best I can tell anyway.  Gkrellm isn't complaining
>> and neither is htop. 
>>
>> The one thing that freaked me out, CPU temps.  They have a new and
>> improved sensor that reads from the silicone wafer itself, deep
>> inside
>> it seems.  It responds faster and is likely more accurate but it is
>> much
>> higher temps than the old temp that was of the die, right under where
>> the CPU cooler sits it seems.  When I first saw 190F temps, I kinda
>> freaked out a bit.  I thought I was trying to cook a egg with my
>> CPU.  I
>> was checking that the fans were running and all that.  I think it was
>> Rich that posted a explanation.  As I said, the temp is likely more
>> accurate but still, I was not expecting that. 
>>
>> When I build a new rig several years from now, I might build a AM5
>> that
>> is either available now or will be shortly.  Even my first rig was
>> built
>> with things that had been released for close to a year.  It was still
>> plenty fast for the time tho.  I even named it Smoker.  My next rig
>> was
>> named Fireball.  I couldn't think of a good name for this rig that
>> would
>> line up with those two names so I went with Gentoo-1.  I guess next
>> rig
>> will be Gentoo-2.  ;-)
>>
>> It is sometimes hard to tell if everything is supported.  Linux is a
>> great community but someone has to be the first to buy something and
>> see
>> if things work.  Once bought, you kinda stuck with it.  It takes a
>> while.  Then you have to figure out which forum or other website
>> where a
>> person is describing what works and what doesn't.  Then how dated
>> that
>> info is.  I find that usually after a year, it is likely supported as
>> far as CPU, mobo and such.  Other stuff like printers may take
>> longer. 
>>
>> It is a problem tho.  It's almost like a leap of faith when buying
>> newly
>> released hardware.  Rant on.  This may help the OP too.  It may be a
>> good idea to mention that some things might not be well supported
>> yet. 
>>
>> Dale
>>
>> :-)  :-) 
>
>
> to be fair... the system i love most is a tired old x58 chipset. intel
> 950 cpu. 32GB of slow ass ram (think 1600). but has an intel 750 nvme.
> and a modern nvidia 980. honestly... its just fine for most things
> (HD). can't do 4k. i mean it can, but not well. but it does HD. 
>
> if you push enough ram on your system, and a good ssd, rest dont really
> matter. 
>
> further more. its hard to tell. just saying. but its hard to tell. if
> you have like intel 750 or 905 optane storage (a good ssd, these 2 are
> legendary). if you have that. and if you have ram to the max that your
> platform can support. 32GB. 64. 128. whatever you can. and you have a
> decent gpu. 980 still decent today. u cant actually tell that intel 950
> cpu, one of the first is bad. can't tell nvidia 980 is bad. can't tell
> 1600 ram is bad. 
>
> and frankly, that machine is enough as a distcc server. funny thing
> about distcc. :) it has a top bandwidth limit. somewhere between 8 and
> 16 threads you fill the entire gigabyte bandwidth. 
>
> I will say this. I had a 9980xe before with 128GB of ram. its a
> monster. its also old. 2018 if i am not mistaken. 18 cores and 36
> threads. my new 9950x3d only has 16 cores and 32 threads and only 96GB
> of ram. so its not that much of an upgrade. old boosted speed was 4GHZ
> (on 9980xe). on 9950x3d is 5.76. i get better single core performance.
> 1.76 more. that's about all i get. in 7 years. 
>
> I will also say this. I kept the intel 750 and intel 905p optane. Both
> are god tier storage. But the optate is got tier squared. If you find
> any of these bad boys... go do:
>
> # smartctl -a /dev/nvme0 | grep "Available Spare Threshold"
>
> and if that is below 100 buy them. i mean for the optane. 
>
> there's the scoop. 
>
> both intel 750 and 900(5) are way way way way faster than any modern
> nvme drives. modern nvme drives are optimized for single thread
> performance. u got one thread, single file, great. 7000 mbps. u dont
> need 7000 coz nothing reads anything that fast. you cant feed it into
> anything. but whatever. most things... when you increase the number of
> threads, and maybe its not one single big file, but multiple small
> files, and maybe you want to read and write at the same time....
>
> i mean you are buying a 48 thread machine. right?
>
> well. most modern nvmes can't handle stuff after 8. they were never
> designed to do so. 
>
> intel 750 was. it was created from the start with the idea in mind that
> the controller had to be absolutely able to do hundreds of threads at
> the same time. 
>
> problem with intel 750... its a perishable drive. it has amazing
> performance. it reads and writes with 2200/1600. regardless of threads
> or files or anything else. it simply doesn't care. its performance per
> core is equal to the performance of however many cores you put into it.
>
> but then there's the 900. and the 905. optane. they kept the
> controller, the one that doesn't care about how many threads you are
> doing, and they changed the storage. and they invented this thing.
> xpoint storage. in terms of ssd/nvme storage... that stuff doesn't die.
> while normal drives count their max life in terabytes. tens. or
> hundreds. or thousands on the high end. OPTANE counts its max life in
> petabytes. 
>
> believe me. if you find any of these drives, and they are not spent,
> get them. your computer becomes something else when you have one of
> these bad boys. they go directly into pciexpress. 4 lanes. but my god
> they dont resemble anything that is out there. and a system with one of
> them is so different from a system that doesn't. especially when you
> push. the controller intel built into those things... what parallelism?
> breaks the space time continuum. 
>
>
> changed ram. changed cpu. changed mobo. changed gpu. changed power
> supply. but kept the 905p. i got a 9100 pro from samsung. guess that's
> the new standard ssd. my optane still kicks it ass in random seek
> read/write with as many threads as the cpu has. 
>


This is true in a lot of ways.  On my old system, the speed of the
spinning rust drives was almost always a bottleneck.  On my new system,
same thing.  No matter how fast my CPU is or how much memory I have, it
can max out the speed of the hard drives and that slows everything
else.  I mostly built this new rig because the old rig was aging and I
was concerned about something failing and me being without a puter.  I
watch TV through my puter.  My TV is on at all times I'm home.  Radio is
OK but can't see much on it.  LOL 

If I had built a threadripper based system, hard drives would still be a
bottleneck. Sure, I have portage's work directory on tmpfs but still, at
some point, that data hits a hard drive.  There's that bottleneck
again.  It's still going to slow things down.  I do a lot of file
copying on my system.  I grab video files, copy them to their new
location.  Then I process them as to naming and such, plus testing them
to be sure they play correctly, and are in English for me. 

On one hand, puters have come a long ways in the past decade or so.  On
the other hand, some things are still the bottleneck that makes the
other fast stuff useless.  Some older machines can run fine even today. 
Heck, I still boot and update my old rig once or twice a month.  I
usually do that when there is a new set of KDE packages to update. 

Still, as users of Gentoo, faster CPUs do help with compiling.  For the
rest tho, the puter just sits there waiting on us.  Like always.  LOL 

Oh, the command above for my little m.2 stick.  It's a Samsung SSD 990
EVO 1TB, from smartctl -i. 


smartctl -a /dev/nvme0n1 | grep "Available Spare Threshold"
Available Spare Threshold:          10%
root@Gentoo-1 / #


Dale

:-)  :-) 

Re: [gentoo-user] Recommended CPU and cooling system

[Michael]

[-- Attachment #1: Type: text/plain, Size: 3944 bytes --]

On Thursday, 24 July 2025 04:59:22 British Summer Time Alexandru N. Barloiu 
wrote:
> On Wed, 2025-07-23 at 22:42 -0500, Dale wrote:
> > Alexandru N. Barloiu wrote:
> btw. dont be surprised if AMD stuff doesn't work as well as Intel
> stuff.
> 
> I am not being a hater. Just being a realist. I was an Intel fan boy.
> Over time started to hate Intel. Made my first AMD system... and I have
> to say... not exactly loving it.
> 
> Some simple examples. Until very recently, kernel 6.16... which is now
> in RC7 stage, no cpu sensor. Kind of a problem. Intel cpus always had
> coretemp. Not saying they are not coming up, one by one. But you spend
> like 5000 or however much each of us spends... you expect to at least
> get a cpu temperature sensor that works.

Perhaps because I wait a year or at least a few months after a new CPU 
generation comes out I have not experienced the same problem with AMD CPUs.  
Instead, I have experienced the same problem with missing sensors many years 
ago when I bought the 1st gen i7 CPU.

Meanwhile, the way Intel threw its users under the bus refusing to provide 
microcode for older CPUs when the CPU vulnerabilities were revealed and now 
news on Intel 13/14th Gen CPU degradation problems, convinced me to 
permanently turn my back on Intel.  ;-)


> Other example. Intel platform. Qemu. No problem. AMD. Weird ACPI table.
> which some setups, if you dont use modern enough edk2 firmware, takes
> 45 seconds for the damn VM to start.

I must admit I have not come across this problem, perhaps because I do use 
edk2.


> Other example. No XMP memory profiles. Will be a pain to actually OC
> memory. On intel platforms you just select the xmp profile and you are
> done. Not on AMD.

The Intel XMP (Extreme Memory Profile) and/or AMD's D.O.C.P. (Direct Over 
Clock Profile) are available on my MoBos.  Depending on the MoBo capabilities 
you can tweak the memory clock manually if you really want to, or just switch 
on DOCP and it will achieve its overclocked performance level.  Is DOCP 
missing on your MoBo?  It may be worth checking if the latest EFI firmware is 
needed to make it available.


> And finally another example. Not sure what the threadripper has. But my
> 9950x3d has 8 SMT high performance cores. and 8 SMT power saving cores.
> but the kernel has no idea which is which. if you set your system on
> powersaving or balanced or high performance... that means nothing to
> the system. to be clear. 0->7 are high performance. 7->15 powersaving.
> and 16->23 high performance siblings (smt - amd version of
> hyperthreading), and 24->31 powersaving siblings. but again. the OS is
> completely unaware of this.

I read there is a new AMD specific V-cache driver made available in the 6.13 
kernel.  Perhaps this will allow switching processes to the chiplet with the 
3D cache?  I wonder how much faster this will make compiling code ... O_O

Please post how long it takes you to emerge qtwebengine, rust, libreoffice.


> just some examples. am sure the amd folks will catch up with the
> software. but just to be clear. lots of stuff dont work out of the box.
> and have to dig really deep for each individual little thing.
> 
> and some things just straight out dont work. and nobody tells you ahead
> of the purchase that they dont work. like for instance. my asus 870X
> creative mobo came very high recommended. but nobody told me you can't
> install windows 10 on it. and nobody told me that wifi/bluetooth
> doesn't work in linux. it will at some point. but it doesn't now.
> 
> sorry for the rant. just things i found out when i got my first AMD
> system.

In my experience, new hardware takes a bit longer to bed in kernel-wise on 
Linux than on MSWindows.  This is why I wait for a while before I jump in with 
my purchases.  However, sometimes the jump in technology is so big and 
enticing it is hard to wait when your old system has been dragging its heels 
for years. 

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [gentoo-user] Recommended CPU and cooling system

[Alexandru N. Barloiu]

On Thu, 2025-07-24 at 11:48 +0100, Michael wrote:
> Please post how long it takes you to emerge qtwebengine, rust,
> libreoffice.

sorry, i dont use any of that. best i can do is:




[root@trandafira:~]# genlop -t sys-devel/gcc
 * sys-devel/gcc

     Fri Jun  6 00:24:16 2025 >>> sys-devel/gcc-14.3.0
       merge time: 10 minutes and 50 seconds.

     Fri Jun  6 00:57:02 2025 >>> sys-devel/gcc-15.1.0
       merge time: 15 minutes and 24 seconds.




which is odd. and wrong. i can brag about it. but the truth is, it
should be around 20 minutes. 


     Fri Jun  6 03:08:48 2025 >>> sys-devel/gcc-15.1.0
       merge time: 22 minutes and 41 seconds.

     Sat Jun  7 12:08:11 2025 >>> sys-devel/gcc-14.3.0
       merge time: 21 minutes and 43 seconds.

     Sat Jun  7 21:46:39 2025 >>> sys-devel/gcc-15.1.0
       merge time: 22 minutes and 24 seconds.

     Sat Jun  7 22:55:58 2025 >>> sys-devel/gcc-14.3.0
       merge time: 21 minutes and 25 seconds.

     Wed Jul  9 19:01:53 2025 >>> sys-devel/gcc-15.1.1_p20250705-r1
       merge time: 22 minutes and 36 seconds.


with LTO on. should say. ignore the 10/15 minutes one. prolly a bug.
but gcc with lto never lies because its not about distcc. its about
your own system. those are the times. no BS times. 

Re: [gentoo-user] Recommended CPU and cooling system

[Dale]

Alexandru N. Barloiu wrote:
> On Thu, 2025-07-24 at 11:48 +0100, Michael wrote:
>> Please post how long it takes you to emerge qtwebengine, rust,
>> libreoffice.
> sorry, i dont use any of that. best i can do is:
>
>
>
>
> [root@trandafira:~]# genlop -t sys-devel/gcc
>  * sys-devel/gcc
>
>      Fri Jun  6 00:24:16 2025 >>> sys-devel/gcc-14.3.0
>        merge time: 10 minutes and 50 seconds.
>
>      Fri Jun  6 00:57:02 2025 >>> sys-devel/gcc-15.1.0
>        merge time: 15 minutes and 24 seconds.
>
>
>
>
> which is odd. and wrong. i can brag about it. but the truth is, it
> should be around 20 minutes. 
>
>
>      Fri Jun  6 03:08:48 2025 >>> sys-devel/gcc-15.1.0
>        merge time: 22 minutes and 41 seconds.
>
>      Sat Jun  7 12:08:11 2025 >>> sys-devel/gcc-14.3.0
>        merge time: 21 minutes and 43 seconds.
>
>      Sat Jun  7 21:46:39 2025 >>> sys-devel/gcc-15.1.0
>        merge time: 22 minutes and 24 seconds.
>
>      Sat Jun  7 22:55:58 2025 >>> sys-devel/gcc-14.3.0
>        merge time: 21 minutes and 25 seconds.
>
>      Wed Jul  9 19:01:53 2025 >>> sys-devel/gcc-15.1.1_p20250705-r1
>        merge time: 22 minutes and 36 seconds.
>
>
> with LTO on. should say. ignore the 10/15 minutes one. prolly a bug.
> but gcc with lto never lies because its not about distcc. its about
> your own system. those are the times. no BS times. 
>
>


Just for comparison.  My CPU is a AMD Ryzen 7 5800X 8-Core Processor
with 8 additional threads.  Compile times for my gcc with binary
installs removed. 


root@Gentoo-1 / # genlop -t gcc
 * sys-devel/gcc

     Fri Jul 19 03:27:59 2024 >>> sys-devel/gcc-13.3.1_p20240614
       merge time: 21 minutes and 51 seconds.

     Tue Sep  3 15:13:38 2024 >>> sys-devel/gcc-13.3.1_p20240614
       merge time: 22 minutes and 13 seconds.

     Tue Sep  3 20:19:44 2024 >>> sys-devel/gcc-13.3.1_p20240614
       merge time: 19 minutes and 26 seconds.

     Wed Sep  4 07:45:44 2024 >>> sys-devel/gcc-13.3.1_p20240614
       merge time: 20 minutes and 7 seconds.

     Sat Nov  2 01:05:39 2024 >>> sys-devel/gcc-13.3.1_p20241025
       merge time: 21 minutes and 4 seconds.


root@Gentoo-1 / #


There's not a lot of difference.  One thing, there is likely other
packages compiling at the same time tho.  I have mine set to do several
at a time. 

Isn't there a website that has this info too?  I just remembered there
used to be one ages ago. 

Dale

:-)  :-) 

Re: [gentoo-user] Recommended CPU and cooling system

[Javier Martinez]

[-- Attachment #1.1.1: Type: text/plain, Size: 3919 bytes --]


Both are in serious troubles, more intel than amd by two reasons Spectre 
Meldown vulnerabilities. Solution to one of them: Disabling 
hyperthreading xD

Intel make things great, they made one backdoor called VPRO et all and 
they assure thereselfs the hability to get in remotely including 
critical vulnerabilities in it xD

I would not defend neither of two, but intel proof thereself to be a 
pain in the ass to the enduser. At least AMD has possibility to crypt 
RAM contents

I'm looking for ARM instead amd64 arch.

You hadn't to play with winmodems in your life usually is it???

To whiteman808:

I would do somethings:

If your place is cold (you live in norway, iceland or places like that), 
compile your system in winter, you will save a lot of money in heating 
your home, also you can use it to cook meat, I have recipts of "steak at 
threadripper" with the meat cutted in thin strips made over the cpu. If 
you are living in a mid temperature place, when your neighbours get 
angry to your door house criticizing the excessive hot that reaches them 
(because their dog got fried) from your house just tell them:

"It's the climate change, not me"

I look for systems with as much RAM as possible to protect harddisks. 
Gentoo is very high disk intensive using and makes a lot of writtings in 
/var/tmp/portage or /usr/portage for example. So, I would have a system 
with enough RAM memory to have serveral zram partitions mounted everywhere

Performance will be kicked with mitigations of spectre vulnerabilities 
as they appear.




El 24/7/25 a las 5:59, Alexandru N. Barloiu escribió:
> On Wed, 2025-07-23 at 22:42 -0500, Dale wrote:
>> Alexandru N. Barloiu wrote:
>>
> 
> 
> btw. dont be surprised if AMD stuff doesn't work as well as Intel
> stuff.
> 
> I am not being a hater. Just being a realist. I was an Intel fan boy.
> Over time started to hate Intel. Made my first AMD system... and I have
> to say... not exactly loving it.
> 
> Some simple examples. Until very recently, kernel 6.16... which is now
> in RC7 stage, no cpu sensor. Kind of a problem. Intel cpus always had
> coretemp. Not saying they are not coming up, one by one. But you spend
> like 5000 or however much each of us spends... you expect to at least
> get a cpu temperature sensor that works.
> 
> Other example. Intel platform. Qemu. No problem. AMD. Weird ACPI table.
> which some setups, if you dont use modern enough edk2 firmware, takes
> 45 seconds for the damn VM to start.
> 
> Other example. No XMP memory profiles. Will be a pain to actually OC
> memory. On intel platforms you just select the xmp profile and you are
> done. Not on AMD.
> 
> And finally another example. Not sure what the threadripper has. But my
> 9950x3d has 8 SMT high performance cores. and 8 SMT power saving cores.
> but the kernel has no idea which is which. if you set your system on
> powersaving or balanced or high performance... that means nothing to
> the system. to be clear. 0->7 are high performance. 7->15 powersaving.
> and 16->23 high performance siblings (smt - amd version of
> hyperthreading), and 24->31 powersaving siblings. but again. the OS is
> completely unaware of this.
> 
> just some examples. am sure the amd folks will catch up with the
> software. but just to be clear. lots of stuff dont work out of the box.
> and have to dig really deep for each individual little thing.
> 
> and some things just straight out dont work. and nobody tells you ahead
> of the purchase that they dont work. like for instance. my asus 870X
> creative mobo came very high recommended. but nobody told me you can't
> install windows 10 on it. and nobody told me that wifi/bluetooth
> doesn't work in linux. it will at some point. but it doesn't now.
> 
> sorry for the rant. just things i found out when i got my first AMD
> system.
> 


[-- Attachment #1.1.2: OpenPGP public key --]
[-- Type: application/pgp-keys, Size: 3145 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]

Re: [gentoo-user] Recommended CPU and cooling system

[Michael]

[-- Attachment #1: Type: text/plain, Size: 2061 bytes --]

On Thursday, 24 July 2025 12:16:21 British Summer Time Alexandru N. Barloiu 
wrote:
> On Thu, 2025-07-24 at 11:48 +0100, Michael wrote:
> > Please post how long it takes you to emerge qtwebengine, rust,
> > libreoffice.
> 
> sorry, i dont use any of that. best i can do is:
> 
> 
> 
> 
> [root@trandafira:~]# genlop -t sys-devel/gcc
>  * sys-devel/gcc
> 
>      Fri Jun  6 00:24:16 2025 >>> sys-devel/gcc-14.3.0
>        merge time: 10 minutes and 50 seconds.
> 
>      Fri Jun  6 00:57:02 2025 >>> sys-devel/gcc-15.1.0
>        merge time: 15 minutes and 24 seconds.
> 
> 
> 
> 
> which is odd. and wrong. i can brag about it. but the truth is, it
> should be around 20 minutes.
> 
> 
>      Fri Jun  6 03:08:48 2025 >>> sys-devel/gcc-15.1.0
>        merge time: 22 minutes and 41 seconds.
> 
>      Sat Jun  7 12:08:11 2025 >>> sys-devel/gcc-14.3.0
>        merge time: 21 minutes and 43 seconds.
> 
>      Sat Jun  7 21:46:39 2025 >>> sys-devel/gcc-15.1.0
>        merge time: 22 minutes and 24 seconds.
> 
>      Sat Jun  7 22:55:58 2025 >>> sys-devel/gcc-14.3.0
>        merge time: 21 minutes and 25 seconds.
> 
>      Wed Jul  9 19:01:53 2025 >>> sys-devel/gcc-15.1.1_p20250705-r1
>        merge time: 22 minutes and 36 seconds.
> 
> 
> with LTO on. should say. ignore the 10/15 minutes one. prolly a bug.
> but gcc with lto never lies because its not about distcc. its about
> your own system. those are the times. no BS times.

Thank you for this.  I don't use distcc.  On an AM4 MoBo with DDR4 I get 29% 
slower times if I enable LTO when emerging gcc, which is to be expected.

     Thu Jun  5 11:40:53 2025 >>> sys-devel/gcc-14.3.0
       merge time: 17 minutes and 22 seconds.

     Thu Jul 24 12:55:07 2025 >>> sys-devel/gcc-14.3.0
       merge time: 24 minutes and 14 seconds.

Compared to your LTO times my PC is ~12% slower, which is surprising.  I was 
expecting yours would show a ~25% improvement.  I wonder what performance 
boost you'd get with the new AMD V-cache driver in the kernel and if this 
would be realisable by a compiler.  o_O

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [gentoo-user] Recommended CPU and cooling system

[Javier Martinez]

[-- Attachment #1.1.1: Type: text/plain, Size: 2430 bytes --]

To be more realistics you have to compite in equal conditions, compiling 
in ram for example. Maybe your harddisk is faster that his one.

8  dd if=/dev/urandom to ram would be fine IMO


El 24/7/25 a las 14:50, Michael escribió:
> On Thursday, 24 July 2025 12:16:21 British Summer Time Alexandru N. Barloiu
> wrote:
>> On Thu, 2025-07-24 at 11:48 +0100, Michael wrote:
>>> Please post how long it takes you to emerge qtwebengine, rust,
>>> libreoffice.
>>
>> sorry, i dont use any of that. best i can do is:
>>
>>
>>
>>
>> [root@trandafira:~]# genlop -t sys-devel/gcc
>>   * sys-devel/gcc
>>
>>       Fri Jun  6 00:24:16 2025 >>> sys-devel/gcc-14.3.0
>>         merge time: 10 minutes and 50 seconds.
>>
>>       Fri Jun  6 00:57:02 2025 >>> sys-devel/gcc-15.1.0
>>         merge time: 15 minutes and 24 seconds.
>>
>>
>>
>>
>> which is odd. and wrong. i can brag about it. but the truth is, it
>> should be around 20 minutes.
>>
>>
>>       Fri Jun  6 03:08:48 2025 >>> sys-devel/gcc-15.1.0
>>         merge time: 22 minutes and 41 seconds.
>>
>>       Sat Jun  7 12:08:11 2025 >>> sys-devel/gcc-14.3.0
>>         merge time: 21 minutes and 43 seconds.
>>
>>       Sat Jun  7 21:46:39 2025 >>> sys-devel/gcc-15.1.0
>>         merge time: 22 minutes and 24 seconds.
>>
>>       Sat Jun  7 22:55:58 2025 >>> sys-devel/gcc-14.3.0
>>         merge time: 21 minutes and 25 seconds.
>>
>>       Wed Jul  9 19:01:53 2025 >>> sys-devel/gcc-15.1.1_p20250705-r1
>>         merge time: 22 minutes and 36 seconds.
>>
>>
>> with LTO on. should say. ignore the 10/15 minutes one. prolly a bug.
>> but gcc with lto never lies because its not about distcc. its about
>> your own system. those are the times. no BS times.
> 
> Thank you for this.  I don't use distcc.  On an AM4 MoBo with DDR4 I get 29%
> slower times if I enable LTO when emerging gcc, which is to be expected.
> 
>       Thu Jun  5 11:40:53 2025 >>> sys-devel/gcc-14.3.0
>         merge time: 17 minutes and 22 seconds.
> 
>       Thu Jul 24 12:55:07 2025 >>> sys-devel/gcc-14.3.0
>         merge time: 24 minutes and 14 seconds.
> 
> Compared to your LTO times my PC is ~12% slower, which is surprising.  I was
> expecting yours would show a ~25% improvement.  I wonder what performance
> boost you'd get with the new AMD V-cache driver in the kernel and if this
> would be realisable by a compiler.  o_O


[-- Attachment #1.1.2: OpenPGP public key --]
[-- Type: application/pgp-keys, Size: 3145 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]

Re: [gentoo-user] Recommended CPU and cooling system

[Michael]

[-- Attachment #1: Type: text/plain, Size: 590 bytes --]

On Thursday, 24 July 2025 13:56:05 British Summer Time Javier Martinez wrote:
> To be more realistics you have to compite in equal conditions, compiling 
> in ram for example. Maybe your harddisk is faster that his one.

My OS is on an M.2 SSD and I also use a RAM tmpfs on this PC, but the way 
packages keep growing larger and larger I wonder how long before I need more 
RAM.  ;-)


> 8  dd if=/dev/urandom to ram would be fine IMO


However, "... building packages in tmpfs is unlikely to provide any benefit on 
a modern system."

https://wiki.gentoo.org/wiki/Portage_TMPDIR_on_tmpfs


[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [gentoo-user] Recommended CPU and cooling system

[Javier Martinez]

[-- Attachment #1.1.1: Type: text/plain, Size: 724 bytes --]

El 24/7/25 a las 15:05, Michael escribió:
> On Thursday, 24 July 2025 13:56:05 British Summer Time Javier Martinez wrote:
>> To be more realistics you have to compite in equal conditions, compiling
>> in ram for example. Maybe your harddisk is faster that his one.
> 
> My OS is on an M.2 SSD and I also use a RAM tmpfs on this PC, but the way
> packages keep growing larger and larger I wonder how long before I need more
> RAM.  ;-)
> 
> 
>> 8  dd if=/dev/urandom to ram would be fine IMO
> 
> 
> However, "... building packages in tmpfs is unlikely to provide any benefit on
> a modern system."
> 
> https://wiki.gentoo.org/wiki/Portage_TMPDIR_on_tmpfs
> 

It avoids to made your ultra fast SSD burn

[-- Attachment #1.1.2: OpenPGP public key --]
[-- Type: application/pgp-keys, Size: 3145 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]

Re: [gentoo-user] Recommended CPU and cooling system

[Dale]

Javier Martinez wrote:
> El 24/7/25 a las 15:05, Michael escribió:
>> On Thursday, 24 July 2025 13:56:05 British Summer Time Javier
>> Martinez wrote:
>>> To be more realistics you have to compite in equal conditions,
>>> compiling
>>> in ram for example. Maybe your harddisk is faster that his one.
>>
>> My OS is on an M.2 SSD and I also use a RAM tmpfs on this PC, but the
>> way
>> packages keep growing larger and larger I wonder how long before I
>> need more
>> RAM.  ;-)
>>
>>
>>> 8  dd if=/dev/urandom to ram would be fine IMO
>>
>>
>> However, "... building packages in tmpfs is unlikely to provide any
>> benefit on
>> a modern system."
>>
>> https://wiki.gentoo.org/wiki/Portage_TMPDIR_on_tmpfs
>>
>
> It avoids to made your ultra fast SSD burn


That's the biggest reason I have portage's work directory on tmpfs.  If
I start having to do it on a disk because of a lack of memory, I'll do
it on spinning rust to save my m.2 stick. 

Dale

:-)  :-) 

P. S.  What is LTO and should I enable something?  I already need to
reboot soon for a newly rebuilt kernel.  May as well do both. 

Re: [gentoo-user] Recommended CPU and cooling system

[Michael]

[-- Attachment #1: Type: text/plain, Size: 1520 bytes --]

On Thursday, 24 July 2025 15:36:57 British Summer Time Dale wrote:
> Javier Martinez wrote:
> > El 24/7/25 a las 15:05, Michael escribió:
> >> On Thursday, 24 July 2025 13:56:05 British Summer Time Javier
> >> 
> >> Martinez wrote:
> >>> To be more realistics you have to compite in equal conditions,
> >>> compiling
> >>> in ram for example. Maybe your harddisk is faster that his one.
> >> 
> >> My OS is on an M.2 SSD and I also use a RAM tmpfs on this PC, but the
> >> way
> >> packages keep growing larger and larger I wonder how long before I
> >> need more
> >> RAM.  ;-)
> >> 
> >>> 8  dd if=/dev/urandom to ram would be fine IMO
> >> 
> >> However, "... building packages in tmpfs is unlikely to provide any
> >> benefit on
> >> a modern system."
> >> 
> >> https://wiki.gentoo.org/wiki/Portage_TMPDIR_on_tmpfs
> > 
> > It avoids to made your ultra fast SSD burn
> 
> That's the biggest reason I have portage's work directory on tmpfs.  If
> I start having to do it on a disk because of a lack of memory, I'll do
> it on spinning rust to save my m.2 stick. 
> 
> Dale
> 
> :-)  :-) 
> 
> P. S.  What is LTO and should I enable something?  I already need to
> reboot soon for a newly rebuilt kernel.  May as well do both. 

Come on Dale, you know how to search the Gentoo wiki.  ;-)

It has some benefits, especially if you are building binaries to run on a 
resource constrained system, but it takes longer to build them and not all 
packages benefit from it.  YMMV ...

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [gentoo-user] Recommended CPU and cooling system

[Rahul Sandhu]

Hi Dale,

> That's the biggest reason I have portage's work directory on tmpfs.  If
> I start having to do it on a disk because of a lack of memory, I'll do
> it on spinning rust to save my m.2 stick.

This really isn't all that true these days, please take a look at the
Gentoo wiki article for Portage TMPDIR on tmpfs, which states[1]:

> Users are strongly cautioned against buying additional RAM to use as tmpfs.

[1] https://wiki.gentoo.org/wiki/Portage_TMPDIR_on_tmpfs

Regards,
Rahul

Re: [gentoo-user] Recommended CPU and cooling system

[Javier Martinez]

[-- Attachment #1.1.1: Type: text/plain, Size: 1429 bytes --]

El 24/7/25 a las 16:43, Rahul Sandhu escribió:
> Hi Dale,
> 
>> That's the biggest reason I have portage's work directory on tmpfs.  If
>> I start having to do it on a disk because of a lack of memory, I'll do
>> it on spinning rust to save my m.2 stick.
> 
> This really isn't all that true these days, please take a look at the
> Gentoo wiki article for Portage TMPDIR on tmpfs, which states[1]:
> 
>> Users are strongly cautioned against buying additional RAM to use as tmpfs.
> 
> [1] https://wiki.gentoo.org/wiki/Portage_TMPDIR_on_tmpfs
> 
> Regards,
> Rahul
> 
Its not question of thuth if not opinions instead, and neither to buy 
more RAM modules but to take a computer with enough RAM from the 
beginning (I will try the next one having 64 GB).

I don't have swap on disk. And got a system with 32 gb of RAM. I 
dedicate 16 of them to /var/tmp/portage. My PC could die and the 
harddisk still alive to be used in another system (such as rockpi4c+)

If you don't use RAM as tmpfs maybe your harddisk will live 5 years, if 
it's almost full don't so many years according of the TBW of each hardisk.

So, is not buy more RAM, is question to get from the beginning enough 
RAM to be able to protect your SSD disk. You can also put your distfiles 
dir in tmpfs use --jobs 1 and got it to remove it after emerging.

Less writtings more lifespan and gentoo does so many writes when emerging.

[-- Attachment #1.1.2: OpenPGP public key --]
[-- Type: application/pgp-keys, Size: 3145 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]

[gentoo-user] Re: Recommended CPU and cooling system

[Nuno Silva]

On 2025-07-24, Javier Martinez wrote:

> El 24/7/25 a las 16:43, Rahul Sandhu escribió:
>> Hi Dale,
>>
>>> That's the biggest reason I have portage's work directory on tmpfs.  If
>>> I start having to do it on a disk because of a lack of memory, I'll do
>>> it on spinning rust to save my m.2 stick.
>>
>> This really isn't all that true these days, please take a look at the
>> Gentoo wiki article for Portage TMPDIR on tmpfs, which states[1]:
>>
>>> Users are strongly cautioned against buying additional RAM to use as tmpfs.
>>
>> [1] https://wiki.gentoo.org/wiki/Portage_TMPDIR_on_tmpfs
>>
>> Regards,
>> Rahul
>>
> Its not question of thuth if not opinions instead, and neither to buy
> more RAM modules but to take a computer with enough RAM from the
> beginning (I will try the next one having 64 GB).
>
> I don't have swap on disk. And got a system with 32 gb of RAM. I
> dedicate 16 of them to /var/tmp/portage. My PC could die and the
> harddisk still alive to be used in another system (such as rockpi4c+)
>
> If you don't use RAM as tmpfs maybe your harddisk will live 5 years,
> if it's almost full don't so many years according of the TBW of each
> hardisk.

I think hard drives tend to usually last way longer than that, including
those used on Gentoo systems and which house /var/tmp/portage.

> So, is not buy more RAM, is question to get from the beginning enough
> RAM to be able to protect your SSD disk. You can also put your
> distfiles dir in tmpfs use --jobs 1 and got it to remove it after
> emerging.
>
> Less writtings more lifespan and gentoo does so many writes when
> emerging.

Maybe a separate on-disk filesystem with very lazy writeback is a more
appropriate solution for this?

-- 
Nuno Silva

Re: [gentoo-user] Re: Recommended CPU and cooling system

[Javier Martinez]

[-- Attachment #1.1.1: Type: text/plain, Size: 2414 bytes --]

El 24/7/25 a las 18:07, Nuno Silva escribió:
> On 2025-07-24, Javier Martinez wrote:
> 
>> El 24/7/25 a las 16:43, Rahul Sandhu escribió:
>>> Hi Dale,
>>>
>>>> That's the biggest reason I have portage's work directory on tmpfs.  If
>>>> I start having to do it on a disk because of a lack of memory, I'll do
>>>> it on spinning rust to save my m.2 stick.
>>>
>>> This really isn't all that true these days, please take a look at the
>>> Gentoo wiki article for Portage TMPDIR on tmpfs, which states[1]:
>>>
>>>> Users are strongly cautioned against buying additional RAM to use as tmpfs.
>>>
>>> [1] https://wiki.gentoo.org/wiki/Portage_TMPDIR_on_tmpfs
>>>
>>> Regards,
>>> Rahul
>>>
>> Its not question of thuth if not opinions instead, and neither to buy
>> more RAM modules but to take a computer with enough RAM from the
>> beginning (I will try the next one having 64 GB).
>>
>> I don't have swap on disk. And got a system with 32 gb of RAM. I
>> dedicate 16 of them to /var/tmp/portage. My PC could die and the
>> harddisk still alive to be used in another system (such as rockpi4c+)
>>
>> If you don't use RAM as tmpfs maybe your harddisk will live 5 years,
>> if it's almost full don't so many years according of the TBW of each
>> hardisk.
> 
> I think hard drives tend to usually last way longer than that, including
> those used on Gentoo systems and which house /var/tmp/portage.
> 
>> So, is not buy more RAM, is question to get from the beginning enough
>> RAM to be able to protect your SSD disk. You can also put your
>> distfiles dir in tmpfs use --jobs 1 and got it to remove it after
>> emerging.
>>
>> Less writtings more lifespan and gentoo does so many writes when
>> emerging.
> 
> Maybe a separate on-disk filesystem with very lazy writeback is a more
> appropriate solution for this?
> 
How many RAM modules have you broken in your life by excessive use? I 
did not break any yet, hard disks a lot

Why are you assign /tmp or /run a tmpfs directory? is a no sense for me.

Data that would not be keeped between reboots should not be written to a 
disk which dies slowly with each byte-written

If you have /tmp in RAM is a no sense to defend /var/tmp/portage in 
disk. I agree that you SHALL NOT buy RAM expansions to your PC to do it 
as it can be cheaper one m2 hard disk. But if you have RAM use RAM and 
protect your disks.

[-- Attachment #1.1.2: OpenPGP public key --]
[-- Type: application/pgp-keys, Size: 3145 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]

[gentoo-user] Re: Recommended CPU and cooling system

[Nuno Silva]

On 2025-07-24, Javier Martinez wrote:

> El 24/7/25 a las 18:07, Nuno Silva escribió:
>> On 2025-07-24, Javier Martinez wrote:
>>
>>> El 24/7/25 a las 16:43, Rahul Sandhu escribió:
>>>> Hi Dale,
>>>>
>>>>> That's the biggest reason I have portage's work directory on tmpfs.  If
>>>>> I start having to do it on a disk because of a lack of memory, I'll do
>>>>> it on spinning rust to save my m.2 stick.
>>>>
>>>> This really isn't all that true these days, please take a look at the
>>>> Gentoo wiki article for Portage TMPDIR on tmpfs, which states[1]:
>>>>
>>>>> Users are strongly cautioned against buying additional RAM to use as tmpfs.
>>>>
>>>> [1] https://wiki.gentoo.org/wiki/Portage_TMPDIR_on_tmpfs
>>>>
>>>> Regards,
>>>> Rahul
>>>>
>>> Its not question of thuth if not opinions instead, and neither to buy
>>> more RAM modules but to take a computer with enough RAM from the
>>> beginning (I will try the next one having 64 GB).
>>>
>>> I don't have swap on disk. And got a system with 32 gb of RAM. I
>>> dedicate 16 of them to /var/tmp/portage. My PC could die and the
>>> harddisk still alive to be used in another system (such as rockpi4c+)
>>>
>>> If you don't use RAM as tmpfs maybe your harddisk will live 5 years,
>>> if it's almost full don't so many years according of the TBW of each
>>> hardisk.
>>
>> I think hard drives tend to usually last way longer than that, including
>> those used on Gentoo systems and which house /var/tmp/portage.
>>
>>> So, is not buy more RAM, is question to get from the beginning enough
>>> RAM to be able to protect your SSD disk. You can also put your
>>> distfiles dir in tmpfs use --jobs 1 and got it to remove it after
>>> emerging.
>>>
>>> Less writtings more lifespan and gentoo does so many writes when
>>> emerging.
>>
>> Maybe a separate on-disk filesystem with very lazy writeback is a more
>> appropriate solution for this?
>>
> How many RAM modules have you broken in your life by excessive use? I
> did not break any yet, hard disks a lot
>
> Why are you assign /tmp or /run a tmpfs directory? is a no sense for me.
>
> Data that would not be keeped between reboots should not be written to
> a disk which dies slowly with each byte-written
>
> If you have /tmp in RAM is a no sense to defend /var/tmp/portage in
> disk. I agree that you SHALL NOT buy RAM expansions to your PC to do
> it as it can be cheaper one m2 hard disk. But if you have RAM use RAM
> and protect your disks.
>

There have certainly been times when buying RAM wasn't really
financially worth it compared to just buying more hard disk space. That
*may* now change if RAM is sufficiently cheaper, and for systems where
RAM modules are easily available, given that several hard disk
manufacturers are replacing the more affordable hard disk tiers with
SMR-only offerings, meaning the price tag for "CMR" disks ends up being
higher.

There have also been times when the motherboard simply could not take
more RAM. I have one amd64 system where I can't use more than
3-something GiB - chipset limitation on the address space puts the
maximum at 4 GiB, and on top of that it still has to include the PCI
hole. (Thanks, Intel!)



(You do understand that if /var/tmp/portage is in disk, then it is
*also* in RAM?)

-- 
Nuno Silva

Re: [gentoo-user] Recommended CPU and cooling system

[Peter Humphrey]

On Thursday, 24 July 2025 15:41:45 British Summer Time Michael wrote:
> On Thursday, 24 July 2025 15:36:57 British Summer Time Dale wrote:
[...]
> > P. S.  What is LTO and should I enable something?  I already need to
> > reboot soon for a newly rebuilt kernel.  May as well do both.
> 
> Come on Dale, you know how to search the Gentoo wiki.  ;-)

I once saw jargon defined as a secret language to denote membership of an 
exclusive club.

Please don't use it like that.

-- 
Regards,
Peter.

Re: [gentoo-user] Recommended CPU and cooling system

[Michael]

[-- Attachment #1: Type: text/plain, Size: 1376 bytes --]

On Friday, 25 July 2025 12:00:39 British Summer Time Peter Humphrey wrote:
> On Thursday, 24 July 2025 15:41:45 British Summer Time Michael wrote:
> > On Thursday, 24 July 2025 15:36:57 British Summer Time Dale wrote:
> [...]
> 
> > > P. S.  What is LTO and should I enable something?  I already need to
> > > reboot soon for a newly rebuilt kernel.  May as well do both.
> > 
> > Come on Dale, you know how to search the Gentoo wiki.  ;-)
> 
> I once saw jargon defined as a secret language to denote membership of an
> exclusive club.
> 
> Please don't use it like that.

Of course, my apologies.  LTO in this case stands for Link Time Optimisation, 
as opposed to Whole Time Optimisation.  It refers to an inter-procedural 
optimisation performed by the compiler at link time.  I understand this to 
mean linking different compilation units together into a single module, which 
/should/ make the final executable faster to execute.  All this analysis and 
optimisation takes time making the compilation process longer, but the final 
executable swifter (hopefully).  Some software benefits more than others, e.g. 
firefox.

This is the article on the wiki:

https://wiki.gentoo.org/wiki/LTO

The section "Experiments" in this article presents a couple of real life 
comparisons:

https://johnnysswlab.com/link-time-optimizations-new-way-to-do-compiler-optimizations/

HTH.

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [gentoo-user] Recommended CPU and cooling system

[Dale]

Peter Humphrey wrote:
> On Thursday, 24 July 2025 15:41:45 British Summer Time Michael wrote:
>> On Thursday, 24 July 2025 15:36:57 British Summer Time Dale wrote:
> [...]
>>> P. S.  What is LTO and should I enable something?  I already need to
>>> reboot soon for a newly rebuilt kernel.  May as well do both.
>> Come on Dale, you know how to search the Gentoo wiki.  ;-)
> I once saw jargon defined as a secret language to denote membership of an 
> exclusive club.
>
> Please don't use it like that.
>


I looked at the wiki.  I'm still not sure what LTO is other than
something about linking.  However, the reply gave me a good hint.  It is
more about resource restrained systems.  I got a fairly powerful CPU,
lot of memory and plenty of OS drive space.  I'm not restrained so it
seems like that option won't help me any.  While the wiki didn't do me
any good, the reply did.  If it might would help, I wanted to adjust my
kernel if that was how this thing works.  I still haven't rebooted yet.
Seems it is a USE flag tho. 

I might add, I rarely use euse to see what a USE flag is.  The
description it gives tends to be meaningless.  I've seen a lot of
descriptions that are like this:  This enables <USE flag here> to
install <certain type> of files or install USE flag options.  I've seen
some where I'm like, 'what does this do again'????  I get why tho.  They
don't have the space for the info that is really needed.  They just
doing the best they can given the limits they have to work with.  Thing
is, sometimes even a google search doesn't yield much better. 

This is why sometimes I ask.  In a very short period of time I can get
the info I need and that really isn't available elsewhere.  I still
don't really know what LTO is, it just sounds like something I don't
need.  ;-) 

Dale

:-)  :-) 

Re: [gentoo-user] Recommended CPU and cooling system

[Frank Steinmetzger]

[-- Attachment #1: Type: text/plain, Size: 3810 bytes --]

A bit late to the party, but I am on the internet and have an opinion. :D

Am Thu, Jul 24, 2025 at 06:59:22AM +0300 schrieb Alexandru N. Barloiu:

> btw. dont be surprised if AMD stuff doesn't work as well as Intel
> stuff. 
> 
> I am not being a hater. Just being a realist. I was an Intel fan boy.
> Over time started to hate Intel. Made my first AMD system... and I have
> to say... not exactly loving it. 
> 
> Other example. Intel platform. Qemu. No problem. AMD. Weird ACPI table.
> which some setups, if you dont use modern enough edk2 firmware, takes
> 45 seconds for the damn VM to start. 

I have an 8700G and qemu works just fine. ¯\(ツ)/¯

> Other example. No XMP memory profiles. Will be a pain to actually OC
> memory. On intel platforms you just select the xmp profile and you are
> done. Not on AMD. 

Intel has XMP, AMD has Expo. Same thing, different name. BTW using RAM 
overclocking increases power consumption considerably because the CPU/ 
controller/whatnot cannot enter powersave states anymore.

> And finally another example. Not sure what the threadripper has. But my
> 9950x3d has 8 SMT high performance cores. and 8 SMT power saving cores.

No, it has 16 cores, all of the same type. And those 16 cores have 16 SMT 
siblings, amounting to 32 threads in total. Perhaps what you mean is that 
only one of the two 8-core chiplets is connected to the humongous 3D cache.

However, AMD *does* offer smaller “powersave” cores in the form of Zen 4/4c 
and Zen 5/5c. But those smaller cores are still full Zen cores architecture- 
wise with the entire instruction set, they are just built smaller, have less 
cache and don’t clock as high.

https://en.wikipedia.org/wiki/Zen_4#Zen_4c
> “Even though the Zen 4c core has a smaller footprint, it is still able to maintain the same IPC as the larger Zen 4 core.“

This is in contrast to Intel, where the small efficiency cores in a Core 
processor are actually a different architecture coming from the Atom lineup.

> but the kernel has no idea which is which. if you set your system on
> powersaving or balanced or high performance... that means nothing to
> the system. to be clear. 0->7 are high performance. 7->15 powersaving.

One of the biggest benefitors of the big cache is gaming. So what *is* 
important is to tell the OS to run games (and ofc. other applications that 
benefit most) on those “better” cores.

> just some examples. am sure the amd folks will catch up with the
> software. but just to be clear. lots of stuff dont work out of the box.
> and have to dig really deep for each individual little thing. 
> 
> and some things just straight out dont work.

Sorry, that last part sounds a little like FUD, you could say the same thing 
about Intel-based systems. I’ve been running on an Intel i5-4590 since 2014, 
and last year built my first AMD rig with the 8700G. Apart from switching 
the graphics driver, I’ve had no real pains from the switchover (using Arch, 
though).

> and nobody tells you ahead of the purchase that they dont work. like for 
> instance. my asus 870X creative mobo came very high recommended. but 
> nobody told me you can't install windows 10 on it.

Huh?

> and nobody told me that wifi/bluetooth doesn't work in linux. it will at 
> some point. but it doesn't now. 

That’s always the pain point when it comes to running Linux on bleeding-edge 
hardware.

> sorry for the rant. just things i found out when i got my first AMD
> system. 

Which could just as well have happened on the newest Intel stuff.

-- 
Grüße | Greetings | Salut | Qapla’
Please do not share anything from, with or about me on any social network.

“Selfies are electronic masturbation.” — Karl Lagerfeld

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]