Gentoo Archives: gentoo-user

From: R0b0t1 <r030t1@×××××.com>
To: gentoo-user@l.g.o
Subject: Re: [gentoo-user] processor speed
Date: Tue, 29 Aug 2017 02:58:02
Message-Id: CAAD4mYgACEVvXN4+wa7mz2-yzCFt3J83aHUjdX3O2sG-DivLNQ@mail.gmail.com
In Reply to: Re: [gentoo-user] processor speed by R0b0t1
1 On Mon, Aug 28, 2017 at 9:54 PM, R0b0t1 <r030t1@×××××.com> wrote:
2 > Hello, I apologize for the tangents.
3 >
4 > The only on-topic comments I can offer are that: yes, those parts seem
5 > to be usable with Gentoo, whereas similarly old parts a decade ago
6 > were not; and, I have been looking for a low power server setup and
7 > would appreciate if you could communicate your ultimate part
8 > selection.
9 >
10 > Also that a $350-$400 CPU seems to be more than sufficient. My
11 > i7-4770K is still very capable and that I look forward to some day
12 > using a multisocket system with very nice Xeons (or the AMD
13 > equivalent, if it becomes competitive).
14 >
15 > On Mon, Aug 28, 2017 at 7:48 PM, Rich Freeman <rich0@g.o> wrote:
16 >> On Mon, Aug 28, 2017 at 8:29 PM, wabe <wabenbau@×××××.com> wrote:
17 >>> Rich Freeman <rich0@g.o> wrote:
18 >>>
19 >>>> On Mon, Aug 28, 2017 at 5:22 PM, wabe <wabenbau@×××××.com> wrote:
20 >>>> >
21 >>>> > I'm using an AMD Phenom(tm) II X4 965 Processor. I bought it six or
22 >>>> > seven years ago when it was brand-new. It still works to my
23 >>>> > satisfaction. But of course recent CPUs (for example AMD Ryzen) are
24 >>>> > much faster. Therefore I wanna buy an AMD Threadripper next year.
25 >>>> > This should be an enormous speedup. :-)
26 >>>>
27 >>>> Having just upgraded one of those to a Ryzen 5 1600 I can tell you
28 >>>> that besides tripling your kernel build speeds, it will also sound
29 >>>> less like a hair dryer and make your room feel less like it has a
30 >>>> space heater inside.
31 >>>
32 >>>
33 >>> I'm not sure what TDP my Phenoms have (95W or 125W). The TDP of the
34 >>> 1950X is rated at 180W. But this is for all cores running at full load.
35 >>> So the effective heat output over time should be lower than with my old
36 >>> CPUs.
37 >>
38 >> Your old CPU has a TDP of 140W. I forget which model exactly I had
39 >> but I think its TDP was 195W.
40 >>
41 >> Sure, the 1950X is going to pull quite a bit of power, but my 1600
42 >> only pulls 65W when going full tilt. It is a very noticeable
43 >> difference. I suspect my old CPU probably used a good portion of that
44 >> at idle.
45 >>
46 >>>
47 >>> Because of the high price for the whole machine (board, ram, cpu...)
48 >>> I will replace my two PCs (one Windoze and one Gentoo) with a single
49 >>> machine. However I have some concerns regarding dualboot. I would
50 >>> prefer NVMe SSDs but I think it may be better to use eSATA disks. Then
51 >>> I easily can switch the disks and it should be impossible that one OS
52 >>> can compromise the other.
53 >>
54 >> Seems like eSATA is harder to find these days. USB3 seems to be the
55 >> way things are going. However, that works just fine.
56 >>
57 >
58 > For a small amount of time you could find combination eSATA/USB 3
59 > connectors. I lament their demise.
60 >
61 >
62 > Something to be aware of is that, in general, USB hubs will operate at
63 > the speed of the slowest device connected. This is problematic because
64 > a lot of motherboards and cases are such that a mouse and keyboard are
65 > on the same hub that you would use at the front of your case. Mice and
66 > keyboards are typically USB 1.1 devices.
67 >
68 > For USB 1.1 to USB 2, there is *supposed to be* one or more
69 > transaction translators that take USB 1.1 data and retransmit it at
70 > USB 2 speeds. Some hubs don't seem to implement this properly and
71 > connecting a USB 1.1 device slows the entire bus down to USB 1.1
72 > speeds. Even if a transaction translator is present, the bus will
73 > remain busy for the entire USB 1.1 communication time taken by the
74 > device, slowing everything down.
75 >
76 > For USB 2 to USB 3, there is no conversion performed. This leads to a
77 > situation contrary to what most people would expect - multiple USB 2
78 > devices can not take advantage of more than the default USB 2
79 > bandwidth. USB 2 connections to a USB 3 hub simply do not use the USB
80 > 3 data lines, which are necessary for the increased bandwidth.
81 >
82 > Additionally, some hubs will downgrade USB 3 links to USB 2 speeds if
83 > a USB 2 device is present for unknown reasons. This might be because
84 > of the issue in the second paragraph, e.g. the requirement to wait for
85 > USB 2 transmissions. Reading the specification as to whether this was
86 > allowed behavior didn't make clarify anything to me.
87 >
88 > Regardless, the result is that if you plug a USB 1.1 device into a USB
89 > 3 hub you might slow your file transfers down by an order of magnitude
90 > or more. This is exactly what I experienced that led me to researching
91 > this issue.
92 >
93 >> On my motherboard at least the PCI-based NVMe came at the cost of
94 >> disabling one of the x16 slots, and the SATA-based one came at the
95 >> cost of disabling one of the SATA ports. So, no PCI-based NVMe for me
96 >> as I have an 8x card in addition to my graphics card.
97 >>
98 >> They really need to make more flexible slots as I believe that the
99 >> slots themselves are electrically compatible - that is you can shove a
100 >> 16x card in a 1x slot as long as you eliminate the plastic that blocks
101 >> this from happening. Granted, I wouldn't want to put my LSI card in a
102 >> 1x slot - it would be nicer if they had a 2x or 4x slot in there, but
103 >> I realize that 1x and 16x seems to be where all the demand is.
104 >>
105 >
106 > This is true. Unless the OS on the graphics card is making assumptions
107 > it shouldn't be, it should be able to run with any number of lanes. A
108 > lot of PCIe bridges can only allocate lanes in multiples of 2, 4, 8,
109 > and 16, however.
110 >
111 > Based on some of my reading however the choices your motherboard
112 > manufacturer made were made because non-server Intel (and AMD?) parts
113 > have a very limited number of PCIe and other high speed interfaces
114 > available. I would need to double check, but the configurations you
115 > want might only be possible with server parts.
116 >
117 >>>
118 >>> Hopefully the price for RAM will drop before I buy the new rig. It's
119 >>> incredible high at the moment.
120 >>>
121 >>
122 >> Yeah, the best price I could find as $99 for 8GB of DDR4 ECC, and only
123 >> at 2400. Not much of a consumer market for ECC.
124 >>
125 >
126 > There was recently a price fixing class action settlement for DDR2
127 > RAM. I hope there is another for modern RAM, as you can plot the price
128 > against that natural disaster where the main manufacturing facilities
129 > were and see that it never went back down afterwards.
130 >
131 > But I suppose my greed is getting to me. Our betters have decided what
132 > the price should be, and I should be happy that I can afford RAM and a
133 > nice computer to use.
134 >
135 > R0b0t1.
136
137 I apologize for my terrible spelling and grammar.