Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hammer info. And som SMP musings.

Author: Eugene Nalimov

Date: 20:45:17 03/26/02

Go up one level in this thread


Of course cache will be NUMA; still, in the worst case it can be 3-5x faster
than main memory.

Let's take a look at the latencies of the not yet released CPU I am currently
working with (from memory, so I can be slightly wrong):
  L1                   1 clock cycle
  L2                   5 clock cycles
  L3                  12 clock cycles
  Main memory    180-225 clock cycles

Assuming that non-local L3 access is 5x slower than local one it'll be 60
cycles. Much better than main memory.

I agree that POWER4 is good processor, I just cannot believe it's so much better
than others.

Eugene

On March 26, 2002 at 23:18:21, Robert Hyatt wrote:

>On March 26, 2002 at 20:29:19, Eugene Nalimov wrote:
>
>>I would not trust POWER4 SPECint number too much. It was obtained at a multi-CPU
>>system with shared L3 cache when all CPUs but one were idle, so that one CPU
>>actually could use 16*8 == 128Mb of L3 cache. I doubt anybody ever will use
>>similar system for single-process calculations.
>>
>>Eugene
>>
>
>
>The cache _must_ be NUMA however.  There is no way to have a cache
>with that many ports without degrading cycle time horrendously.  I'll
>bet there is much more to this that just a bigger cache issue...
>
>
>>
>>On March 26, 2002 at 17:24:45, Tom Kerrigan wrote:
>>
>>>On March 26, 2002 at 14:29:28, Vincent Diepeveen wrote:
>>>
>>>>If a 64 bits processor isn't faster than a 32 bits processor the processor
>>>>is nothing more than a bad joke of course.
>>>
>>>Why is that? You might as well say any processor that's not the fastest in the
>>>world is a bad joke. I don't see why the datapath width matters. And you're
>>>saying "faster" now instead of "clocked higher." If you want to talk faster, the
>>>POWER4 posts higher SPECint numbers than all current 32-bit processors.
>>>
>>>>>No matter how hard you backpedal, you're not going to get out of your idiot
>>>>>statement that "not a single 64 bits processor is clocked *near* 32 bits
>>>>>processors."
>>>>
>>>>I'm very right here. fastest 32 bits processor which i can buy is
>>>>clocked at 2.4Ghz now. Fastest 64 bits processor (let's not even
>>>>mention its insane price) is the power4 or something 1.3Ghz if
>>>>i remember well?
>>>
>>>You said "32 bits processors," not the fastest 32-bit processor. The POWER4, at
>>>1.3GHz, is most certainly clocked *near* (your word) the 1.7GHz Athlon, which is
>>>a perfectly good 32-bit processor.
>>>
>>>-Tom



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.