PFGW benchmarking

carpetpool · 2019-09-30, 00:29

I've been using PFGW 4.0.0 to find new random PRPs (and primes) of large sizes. Of the two machines I use to test numbers, one is an Intel Core i5-3230M CPU (Windows 10) and the other is an AMD FX-4300 CPU (Ubuntu).

Just now I was running one of my 160K digit PRP canditiates when I got times:

(757.9948s+3.4911s) with i5-3230M machine
(1079.1039s+10.0399s) with AMD FX-4300 CPU

What is strange about this is the i5 machine has a clock speed of 2.60 GHz and the AMD machine has a clock speed of 3.80 GHz. Not to mention the i5 only has 2 physical cores compared to the AMD with 4 physical cores.

So my question would be why the running time would be shorter for the Windows 10 machine compared to the Linux Machine? I think it has something to do with the CPU brand performance but I'm not sure.

Secondly, I'm assuming PFGW is NOT multithreaded. Weather it is or not, would there be time difference on two machines with the same operating system, processor type, and clock speed, but one machine has significantly more cores than the other (For example, a 3.5 Ghz @ 4 cores v.s. 3.5 Ghz @ 8 cores)?

Please let me know if you have any information that might be helpful. Thanks.

kuratkull · 2019-09-30, 04:34

PFGW isn't multithreaded and Intel has AVX special instructions. At least that is what seems to decide a large chunk of the performance for me.

Happy5214 · 2019-09-30, 06:19

According to Wikipedia, AMD FX-4300 doesn't have AVX2 support, but it does support AVX and FMA3.

mackerel · 2019-09-30, 08:02

Per-core/per-clock the Intel CPUs of that era is a LOT faster than AMD CPUs. Having instruction support isn't enough, there has to be adequate hardware implemented to back it up, which is lacking in the AMD CPU. I'm not too familiar with that era, is that the ones where there was one FPU shared between two cores?

At 160k FFT it would probably have poor scaling even if you were to run multi-threaded, compared to running one task per core. I don't recall if PFGW now has support for multithead, but it was talked about in the past. Also give LLR a try (which I know has multi-thread support via -t option). Sometimes one or the other might be faster.

If you start running multiple tasks on a CPU, be aware you might hit bottlenecks in ram bandwidth, and also check thermals especially on mobile devices if that's what the Intel CPU is in.

kuratkull · 2019-09-30, 20:06

My mistake. It seems at least the newest(v4) PFGW supports threads with the -T switch.

2019-09-30, 00:29	#1
carpetpool "Sam" Nov 2016 517₈ Posts	PFGW benchmarking I've been using PFGW 4.0.0 to find new random PRPs (and primes) of large sizes. Of the two machines I use to test numbers, one is an Intel Core i5-3230M CPU (Windows 10) and the other is an AMD FX-4300 CPU (Ubuntu). Just now I was running one of my 160K digit PRP canditiates when I got times: (757.9948s+3.4911s) with i5-3230M machine (1079.1039s+10.0399s) with AMD FX-4300 CPU What is strange about this is the i5 machine has a clock speed of 2.60 GHz and the AMD machine has a clock speed of 3.80 GHz. Not to mention the i5 only has 2 physical cores compared to the AMD with 4 physical cores. So my question would be why the running time would be shorter for the Windows 10 machine compared to the Linux Machine? I think it has something to do with the CPU brand performance but I'm not sure. Secondly, I'm assuming PFGW is NOT multithreaded. Weather it is or not, would there be time difference on two machines with the same operating system, processor type, and clock speed, but one machine has significantly more cores than the other (For example, a 3.5 Ghz @ 4 cores v.s. 3.5 Ghz @ 8 cores)? Please let me know if you have any information that might be helpful. Thanks.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Looking for benchmarking help with a Phenom or PhenomII X6	mrolle	Software	25	2012-03-14 14:15
GMP 5.0.1 vs GMP 4.1.4 benchmarking	unconnected	GMP-ECM	5	2011-04-03 16:16
PFGW 3.3.6 or PFGW 3.4.2 Please update now!	Joe O	Sierpinski/Riesel Base 5	5	2010-09-30 14:07
Benchmarking suite discussion	Mystwalker	GMP-ECM	7	2006-06-11 10:08
Benchmarking challenge!	Xyzzy	Software	17	2003-08-26 15:43

2019-09-30, 04:34	#2
kuratkull Mar 2007 Estonia 149 Posts	PFGW isn't multithreaded and Intel has AVX special instructions. At least that is what seems to decide a large chunk of the performance for me.

2019-09-30, 06:19	#3
Happy5214 "Alexander" Nov 2008 The Alamo City 991 Posts	According to Wikipedia, AMD FX-4300 doesn't have AVX2 support, but it does support AVX and FMA3.

2019-09-30, 08:02	#4
mackerel Feb 2016 UK 2⁶·7 Posts	Per-core/per-clock the Intel CPUs of that era is a LOT faster than AMD CPUs. Having instruction support isn't enough, there has to be adequate hardware implemented to back it up, which is lacking in the AMD CPU. I'm not too familiar with that era, is that the ones where there was one FPU shared between two cores? At 160k FFT it would probably have poor scaling even if you were to run multi-threaded, compared to running one task per core. I don't recall if PFGW now has support for multithead, but it was talked about in the past. Also give LLR a try (which I know has multi-thread support via -t option). Sometimes one or the other might be faster. If you start running multiple tasks on a CPU, be aware you might hit bottlenecks in ram bandwidth, and also check thermals especially on mobile devices if that's what the Intel CPU is in.

2019-09-30, 20:06	#5
kuratkull Mar 2007 Estonia 149 Posts	My mistake. It seems at least the newest(v4) PFGW supports threads with the -T switch.