mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware

Reply
 
Thread Tools
Old 2005-12-02, 12:17   #1
lpmurray
 
lpmurray's Avatar
 
Sep 2002

1318 Posts
Default mflops and new processors

We all have seen the standard 32.98 MFLOP for a P90 but I was curious about what the MFLOP rating for 3.2-3.8GHz P4's and more important what are the MFLOP's for the new high end duel core processors... If anyone has these numbers I would be very greatful if you could post them here. Also with all the talk of the xbox 360 and the new sony running in the terra flop range..... I understand they are single precision but if they ever change these to duel at these speeds we could actually start working on 100 million digit numbers (can't wait to see the poster for the first 1 of those found )
lpmurray is offline   Reply With Quote
Old 2005-12-02, 18:28   #2
ColdFury
 
ColdFury's Avatar
 
Aug 2002

26·5 Posts
Default

MFLOPs as a performance measure are pretty useless.
ColdFury is offline   Reply With Quote
Old 2005-12-03, 00:27   #3
dsouza123
 
dsouza123's Avatar
 
Sep 2002

2×331 Posts
Default

On Top500.org they rate a 2.0 Ghz Opteron (64 Bit) at 4 GFlops.

An Athlon 64 at the same speed should be about the same.
dsouza123 is offline   Reply With Quote
Old 2005-12-03, 00:31   #4
dsouza123
 
dsouza123's Avatar
 
Sep 2002

66210 Posts
Default

Intel Xeon EM64T 3.6 Ghz at 7.2 GFlops.

A Pentium 4 with EM64T (64 bit extensions)
at the same speed should be about the same.
dsouza123 is offline   Reply With Quote
Old 2005-12-03, 01:07   #5
alpertron
 
alpertron's Avatar
 
Aug 2002
Buenos Aires, Argentina

22×3×113 Posts
Default

It appears that the Pentium 4 has very high latencies compared against Opterons and 64-bit Athlons. Please read the GMP benchmarks for details.
alpertron is offline   Reply With Quote
Old 2005-12-05, 15:49   #6
Peter Nelson
 
Peter Nelson's Avatar
 
Oct 2004

232 Posts
Default

Quote:
Originally Posted by alpertron
It appears that the Pentium 4 has very high latencies compared against Opterons and 64-bit Athlons. Please read the GMP benchmarks for details.
The GMP bench now is not a fair comparison because the AMD 64 bit code has not been optimised in assembler (they would like some support to complete this), whereas the intel /32 bit code has been.

I think you meant their paper on comparing instruction timings:

http://swox.com/doc/x86-timing.pdf

I also wonder whether Intel might have improved it slightly in 8xx or particularly 9xx processors over the 6xx implementation of 64 bit instructions (particularly the math ones we are interested in).

Last fiddled with by Peter Nelson on 2005-12-05 at 15:52
Peter Nelson is offline   Reply With Quote
Old 2005-12-05, 20:18   #7
ixfd64
Bemusing Prompter
 
ixfd64's Avatar
 
"Danny"
Dec 2002
California

2·3·397 Posts
Default

I've heard that a Pentium 4 at 3.06 GHz runs at 12 GFLOPS. I'm not sure what model, though.
ixfd64 is offline   Reply With Quote
Old 2005-12-06, 00:38   #8
Peter Nelson
 
Peter Nelson's Avatar
 
Oct 2004

232 Posts
Default

Quote:
Originally Posted by ixfd64
I've heard that a Pentium 4 at 3.06 GHz runs at 12 GFLOPS. I'm not sure what model, though.
When WE talk about FLOPS we generally refer to full 64 bit ("double") floating point calculations rather than single precision (32 bit) operations.

Since the bigger calcs take longer they are not comparable.

Benchmarks like Linpack use doubles for their "FLOPS" measurements.

Also, the instructions per cycle very much depend on the particular instruction on a particular chip. Thus one which beats another on adding may be slower on multiply but the same speed on divide. Or any such permutations. Thus a "floating point operation" (FLOP) could be measured using some mixture of different operations to give a performance metric, but the real speed will depend on what your application uses. eg BLAS library, Linpack, Prime95 use different mixture of instructions.

The big "Gigaflops" ratings attributed to Cell (in Playstation3) or graphics cards, are NOT comparable with generally used double precision cpu results, because the Cell and gpus can't handle double precision, only single. To compare them is comparing apples and oranges. Similarly comparing single-precision Pentium benchmarks (when double could be used) with something else's double precision gigaflops is silly as the single calcs are doing less work.
Peter Nelson is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
AVX on AMD processors. pepi37 Software 16 2013-01-10 00:37
HT processors paulunderwood 3*2^n-1 Search 7 2007-02-15 15:47
New 65nm Processors moo Hardware 8 2005-12-22 05:46
New processors and chipsets Peter Nelson Hardware 4 2005-11-28 20:09
64 bit processors brandon_2003 Hardware 52 2005-03-27 11:19

All times are UTC. The time now is 08:16.

Wed May 12 08:16:24 UTC 2021 up 34 days, 2:57, 0 users, load averages: 1.82, 1.81, 1.86

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.