Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Anandtech's Chess Benchmarks shows that Pentium is better than AMD

Author: Vincent Diepeveen

Date: 19:05:57 12/31/04

Go up one level in this thread


On December 31, 2004 at 18:53:15, Anthony Cozzie wrote:

This is the old review where everyone figured soon out how bad it was.

They compiled it wrong at the time simply. Forgot there is something called
'optimizations for k8' and forgot pgo too (no matter whether it favours P4 or k8
more, it SPEEDS UP THE APPLICATION). Look further in their review you might find
benchmarks like pi too, which actually use 0.1 second of the cpu time and the
rest is just i/o time :)

In fact they logged on remote to a p4 which was intel configured, versus a bad
configured A64 they had themselves at home.

In the newsgroups this beginner was later all corrected.

Anandtech has sincethen hopefully fired that tester and for sure taken the best
one. If i remember well Johan de Gelas is there now doing the cpu reviews. He's
very good and objective in testing. Attached to a belgium college faculty. He
does not get paid by manufacturers and is unbribable and keeps objective, simply
has hard proven that over the past years. Really one of the best testers i ever
met. Usual testers just are out for letting the latest cpu they test look good.
He's not like that. He's objective.

This entire review has been written with in mind letting the P4 look like good.
God knows where they found those benchmarks. Also the initial article was even
more horrible than this. Then it was posted how to compile using -march=k8 and
how to use pgo and the dudes still posted this crap. So it is 100% sure they
were there to let intel look fine. Even when pgo on paper is better for P4 than
for k8 i would still want pgo with gcc.

The current fx55 (2.4Ghz) is just kicking the butt of
the P4 3.8Ghz now bigtime. And that with slow RAM inside the fx55, see the slow
results they get for memory latencies of the fx55. I know others get <= 91 ns
there with 2-2-2 memory where they get 120 ns latency.

See a small comparision for example at www.sudhian.com with diep benchmarks at
the cpu's just released only a few weeks ago.

http://www.sudhian.com/showdocs.cfm?aid=635&pid=2403

The test was performed at sudhian with a net2003 compile using -O2 -G7

visual c++ does't have pgo features yet. in some months from now they promised
to release a professional edition with x86-64 support and pgo.

It will be interesting to see which cpu's profit most from that in the long run.

Vincent

>On December 31, 2004 at 18:33:01, Eran Karu wrote:
>
>>Please click below.
>>
>>http://www.anandtech.com/linux/showdoc.aspx?i=2213&p=4
>>
>>In the graph, you can see that Pentium runs more nodes per second than AMD does.
>>So, while many people say AMD is better than Pentium for chess, why does the
>>graph show the opposite?
>>
>>Eran Karu
>
>It appears you should buy a Pentium4 for all your TSCP analysis & games :)
>
>More seriously, the P4 runs very small progams like TSCP well because they fit
>entirely into its tiny caches.  The K8 is generally regarded as superior for
>most real engines.  Of course, this is a 10-20% difference for most programs,
>not exactly the end of the world if you want to buy a P4 for whatever other
>reason . . .
>
>anthony



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.