Re: [gentoo-user] Re: Overclocking CPU causes segmentation fault

[Michael Mol]

On Wed, Jan 23, 2013 at 12:16 PM, Volker Armin Hemmann
<volkerarmin@××××××××××.com> wrote:
> Am 23.01.2013 16:35, schrieb Nikos Chantziaras:
>> On 23/01/13 17:09, Nilesh Govindrajan wrote:
>>> On Wednesday 23 January 2013 07:52:03 PM IST, Nikos Chantziaras wrote:
>>>> [...]
>>>> In my experience, most of the time you can overclock.  The issue is
>>>> with the user not knowing exactly how to do it.  You need to
>>>> understand a few things and how they affect each other.  It's not just
>>>> a knob you can turn.
>>>
>>> That pretty much applies to me. I don't know much about hardware stuff.
>>> Regarding your 1 Ghz overclock, you probably have good components in
>>> terms of RAM & SMPS.
>>> When I bought this rig in 2008, I knew nothing about good components,
>>> blindly trusted local vendor... also internet shopping wasn't advanced
>>> here.
>>> So pretty much substandard components.
>>
>> The part that's really important is the mainboard.  RAM doesn't
>> matter.  In my case, I had pretty basic 800MHz DDR2 RAM.  Raising the
>> FSB would bring it above that, so I changed the DRAM ratio to 1:1, and
>> the RAM then ran at only 600Mhz.
>>
>> That was the starting point to rule out RAM problems.  After that, I
>> raised FSB but kept the VCore constant until I hit the first
>> instabilities.  When that happened, I raised VCore a bit.  Rinse and
>> repeat, until the VCore was still below the maximum recommendation by
>> Intel.  That happened at 3.4GHz (378MHz FSB * 9 CPU multiplier =
>> 3402MHz CPU clock.)  The E6600 CPU I got was an average sample.
>> Others were running it at 3.6GHz (or even higher with water cooling.)
>>
>> This was a process that took about 3 days to complete (needs a lot of
>> stability testing.)  The good thing about those older CPUs was that
>> the performance boost I got by OCing wasn't just scaling linearly with
>> the CPU frequency.  It was scaling *better* than that, because raising
>> the FSB also made the mainboard itself perform better and with lower
>> latencies.
>>
> and here we are - the point where the suspension of disbelief ends.
>
> All you may have gained you threw away with the slower ram - and you are
> trying to tell us that your rig was faster?
>
> You do know that with today's CPUs the CPU is not the bottleneck - the
> slow as molasses, no speed bump for 10 years ram is.
>
> (just look at the internal clock rate of dram chips - and you realize
> that ddr1-3 are pretty much the same crap).
>

Volker, in applications speficially tuned to keep their hot data small
enough to stay in CPU cache (so, anything with a "frames per second"
measurement), overclocking the CPU would still see performance
improvements. Cache misses are always painful.


--
:wq