mersenneforum.org  

Go Back   mersenneforum.org > New To GIMPS? Start Here! > Information & Answers

Reply
 
Thread Tools
Old 2013-11-07, 09:53   #1
Jayder
 
Jayder's Avatar
 
Dec 2012

2·139 Posts
Default Is there a program that performs like Prime95?

I look at a lot of benchmarks for CPUs, and a lot of the same programs are used for benchmarking. Blender, WinRAR, Handbrake. Unfortunately Prime95 is not one of them.

Is there a program out there commonly used for benchmarks that performs similarly to Prime95? Something that I can look at and say, "Oh. CPU X performs better than CPU Y in program Blank. CPU X would probably best CPU Y in Prime95, then." Even one that is only barely or only sometimes similar?

I realise programs run very differently and are optimised differently. I am not really expecting an answer, but am hoping to be surprised by one.

I also realise there are benchmarks both on mersenne.org and mersenne.ca, but neither site has everything.

Bonus question: Same question, but for mfaktc/o?
Jayder is offline   Reply With Quote
Old 2013-11-07, 22:14   #2
ewmayer
2ω=0
 
ewmayer's Avatar
 
Sep 2002
República de California

23·1,229 Posts
Default

Try Linpack.
ewmayer is offline   Reply With Quote
Old 2013-11-07, 22:19   #3
cheesehead
 
cheesehead's Avatar
 
"Richard B. Woods"
Aug 2002
Wisconsin USA

22×3×599 Posts
Default

Quote:
Originally Posted by Jayder View Post
Is there a program out there commonly used for benchmarks that performs similarly to Prime95?
No, because Prime95 was never intended for making fair comparisons between CPUs. It is designed to take particular advantage of Intel's decision to optimize floating-point calculations by using 80-bit floating-point arithmetic during the course of internal FP operations, between FP 64-bit register references.

Quote:
Something that I can look at and say, "Oh. CPU X performs better than CPU Y in program Blank. CPU X would probably best CPU Y in Prime95, then." Even one that is only barely or only sometimes similar?
Prime95 is "unfairly" designed to take advantage of Intel architecture, not intended for "fairly" comparing Intel performance to other CPUs in anything besides the specific prime95 operations.

(Yes, prime95 has some optimizations for a few other types of CPU, but AFAIK no non-Intel architecture has as much capability for optimizing FP arithmetic as Intel has, giving Intel a specific advantage in prime95 operations. Also, by "Intel" I mean "Intel and clones of Intel FP architecture" in these comments.)

Prime95 benchmarks are intended only for comparing CPU performance on prime95 operations, not for comparisons on any other non-prime95 computations.

Extrapolating prime95 performances to performance on any other type of task is not what prime95 benchmarks are designed for.

Linpack and other benchmarking software specifically does not take any "unfair" advantage of a particular manufacturer's CPU architecture. They just throw a particular set of tasks, resembling real-world applications commonly used by businesses, at the CPU to measure how it performs on those tasks. None of those tasks incorporate the intensive FP optimization that George Woltman programmed in prime95 for taking advantage of Intel FP architecture.

Last fiddled with by cheesehead on 2013-11-07 at 22:38
cheesehead is offline   Reply With Quote
Old 2013-11-08, 04:37   #4
TheMawn
 
TheMawn's Avatar
 
May 2013
East. Always East.

11·157 Posts
Default

Are you trying to decide on a CPU to purchase?
TheMawn is offline   Reply With Quote
Old 2013-11-08, 05:13   #5
Jayder
 
Jayder's Avatar
 
Dec 2012

2×139 Posts
Default

I think I understand you, Cheesehead. I know that Intel's AVX is such an advantage that it would be stupid to consider a non Intel AVX CPU for the purpose of running Prime95. Differences between architectures are also not insignificant.

I think I understand. I suppose I was just making sure.

Still keeping in mind that different programs run very differently, what if I were looking for a CPU that ggnfs runs particularly well on? How might I do such a thing? Very few benchmarks exist for ggnfs. Could I infer anything from looking at benchmarks for other programs? Or is it the same deal as with Prime95?

@TheMawn, I'm kind of fantasizing. But one day soon I hope to have the money to build a proper rig. Maybe when Broadwell comes out.

I guess what I'm trying to learn is how to identify a CPU that would run Prime95/ggnfs/gmp-ecm well. (Yes, I know each of those programs are extraordinarily different.)

Last fiddled with by Jayder on 2013-11-08 at 05:21
Jayder is offline   Reply With Quote
Old 2013-11-08, 05:32   #6
ewmayer
2ω=0
 
ewmayer's Avatar
 
Sep 2002
República de California

266816 Posts
Default

The Linpack testsuites include versions tuned for x86 SIMD - In fact the Haswell Preview thread in Hardware has data showing how the latest AVX+FMA optimized Linpack kernels are better thermal stressors of the Haswell CPUs than is Prime95, because linear algebra involves a better balance of add and mul instructions.

I did not make my above 2-word post out of ignorance, but rather because I felt more need not be said. Even if one were not acquainted with the aforementioned thread, a simple web search quickly turns up the answer to "is Linpack optimized for x86+SIMD?" Any hardware vendor offering high-performance math libraries will have expended significant effort tuning such core math libraries for their CPUs.

Also note that the x86 80-bit float register format is irrelevant to SIMD-optimized codes, including Prime95. [Except perhaps for high-precision one-time precomputations.]

Last fiddled with by ewmayer on 2013-11-08 at 05:35
ewmayer is offline   Reply With Quote
Old 2013-11-08, 05:40   #7
db597
 
db597's Avatar
 
Jan 2003

7×29 Posts
Default

What type of performance are you trying to measure? Programmes like P95 are built with a singular purpose in mind. They are very narrow and may not be a good indicator of real world performance.

If it's a distributed computing project you want to benchmark, other examples are dnetc (distributed.net), folding at home (cpu/graphics card), or for graphics card alone you can try gpugrid.net.
db597 is offline   Reply With Quote
Old 2013-11-08, 08:18   #8
Jayder
 
Jayder's Avatar
 
Dec 2012

1000101102 Posts
Default

My initial question wasn't very good at all. Sorry. A better question might be:

If no benchmarks are available, how can I tell what CPUs would be best suited for certain programs?

The program I'm interested in primarily is ggnfs, though I would like a machine that is also good for Prime95. Currently that means Intel.

I'm not looking for an answer like "Go with the best current-gen i7 you can afford". That's not very future proof. (Is it?) I would like to learn how to identify a good CPU for (say) ggnfs whenever a new one comes out.

I asked about other program benchmarks because I thought I could use them as a tool to figure out how a particular CPU performed. My line of thought was: "If I can find a program that runs similarly to ggnfs (or Prime95), I don't need benchmarks for ggnfs. I just need the benchmarks for that other program."

I hope that is not too convoluted.

Last fiddled with by Jayder on 2013-11-08 at 08:18
Jayder is offline   Reply With Quote
Old 2013-11-08, 08:53   #9
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

213448 Posts
Default

Q: Is there a program that performs like Prime95?

A: No, Prime 95 performs better than any program.

(hehe)
LaurV is offline   Reply With Quote
Old 2013-11-08, 10:46   #10
xilman
Bamboozled!
 
xilman's Avatar
 
"𒉺𒌌𒇷𒆷𒀭"
May 2003
Down not across

2·3·1,699 Posts
Default

Quote:
Originally Posted by Jayder View Post
Still keeping in mind that different programs run very differently, what if I were looking for a CPU that ggnfs runs particularly well on? How might I do such a thing?
Find someone with a cpu similar (ideally identical) to the one you are considering and see if they will run a benchmark for you.

It might not be possible within a specific time scale. However, it's all but certain to be impossible if you don't try.

Last fiddled with by xilman on 2013-11-08 at 10:46 Reason: Fix tag
xilman is offline   Reply With Quote
Old 2013-11-08, 19:29   #11
cheesehead
 
cheesehead's Avatar
 
"Richard B. Woods"
Aug 2002
Wisconsin USA

22×3×599 Posts
Default

Quote:
Originally Posted by ewmayer View Post
The Linpack testsuites include versions tuned for x86 SIMD -
Thank you.
cheesehead is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
GPU LLR program jasong GPU Computing 19 2011-08-23 03:32
about the program boooh Information & Answers 2 2010-03-22 15:22
Is there any program... mart_r Software 2 2009-11-15 20:06
Old Program moo Software 0 2006-06-27 00:19
program P-1 for K*2^n-1 jocelynl 15k Search 19 2004-01-11 17:24

All times are UTC. The time now is 12:11.

Fri Nov 27 12:11:49 UTC 2020 up 78 days, 9:22, 4 users, load averages: 0.99, 1.17, 1.20

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, 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.