Linux vs Windows performance issue

2010-05-19, 07:22

Hi,
I'm running mprime/prime95 on two identical computers, one is running Windows XP SP3 and the other runs Ubuntu 9.10. Both are now doing LL testing on large exponents (48 mil).
Linux mprime per-iteration time is roughly 0.127 sec.
Windows prime95 per-iteration time is roughly 0.055 sec.

What's going on here?? Shouldn't Linux have better performance than Windows?
Please help... thanks.

Uncwilly · 2010-05-19, 13:14

Quote:

Originally Posted by Freezer

What's going on here??

How many exponents are each of them working on? Is hyperthreading on for the Windows machine?

retina · 2010-05-19, 13:32

Quote:

Originally Posted by Freezer

Shouldn't Linux have better performance than Windows?

No. Why do you think that?

In my experience the Windows scheduler is very efficient. Whereas the Linux scheduler is very average.

rajula · 2010-05-19, 13:46

Are you sure you have installed 64-bit linux and using 64-bit mprime?

joblack · 2010-05-19, 14:20

Quote:

Originally Posted by Freezer

Hi,
I'm running mprime/prime95 on two identical computers, one is running Windows XP SP3 and the other runs Ubuntu 9.10. Both are now doing LL testing on large exponents (48 mil).
Linux mprime per-iteration time is roughly 0.127 sec.
Windows prime95 per-iteration time is roughly 0.055 sec.

What's going on here?? Shouldn't Linux have better performance than Windows?
Please help... thanks.

I've got the same problem (described in the hardware thread in more detail). It's a dual boot solution (means it's the same hardware) and the configuration is the same.

It seems that the mprime is slower in certain circumstances (on my main Pc the win and linux version performs the same). Haven't heart from Mr. Prime95 so far but it seems to be a problem.

PS: Retina - the Linux scheduler is very fast - even if it wouldn't be half of the performance is a little bit over the top ;).

retina · 2010-05-19, 16:13

Quote:

Originally Posted by joblack

PS: Retina - the Linux scheduler is very fast - even if it wouldn't be half of the performance is a little bit over the top ;).

I was just responding to the sentence I quoted. The thing with 1/2 performance is entirely a different matter.

__HRB__ · 2010-05-19, 17:26

Quote:

Originally Posted by Freezer

Please help... thanks.

I doubt it's a Linux-issue. You could try ruling out a debian/*buntu-issue by picking some other 64-bit live-CD from http://distrowatch.com/

jrk · 2010-05-19, 17:55

Quote:

Originally Posted by Freezer

Hi,
I'm running mprime/prime95 on two identical computers, one is running Windows XP SP3 and the other runs Ubuntu 9.10. Both are now doing LL testing on large exponents (48 mil).
Linux mprime per-iteration time is roughly 0.127 sec.
Windows prime95 per-iteration time is roughly 0.055 sec.

What's going on here?? Shouldn't Linux have better performance than Windows?
Please help... thanks.

Check whether CPU frequency scaling is being used.

joblack · 2010-05-19, 21:05

Quote:

Originally Posted by jrk

Check whether CPU frequency scaling is being used.

I can speak for my configuration. I have the cpufreq configuration set up to 'performance' and it didn't change the bad performance.

By the way I got the same kernel version from the same source (debian amd64 2.6.32-x) and I also tried the 2.6.34 (self-compiled - nothing changed).

__HRB__ · 2010-05-20, 01:05

Hm...what happens when you start another instance of mprime? If the performance/process is unchanged on linux but drops by ~50% on windows it's obvious that we should check the configuration for multithreading (presumably you have a dual core processor).

joblack · 2010-05-20, 16:27

Quote:

Originally Posted by __HRB__

Hm...what happens when you start another instance of mprime? If the performance/process is unchanged on linux but drops by ~50% on windows it's obvious that we should check the configuration for multithreading (presumably you have a dual core processor).

The point is that `top` shows a fairly good processor use (around 190 %) so both cores are occupied.

Perhaps mprime does the wrong cpu allocation ...

2010-05-19, 07:22	#1
Freezer 2×7²×41 Posts	Linux vs Windows performance issue Hi, I'm running mprime/prime95 on two identical computers, one is running Windows XP SP3 and the other runs Ubuntu 9.10. Both are now doing LL testing on large exponents (48 mil). Linux mprime per-iteration time is roughly 0.127 sec. Windows prime95 per-iteration time is roughly 0.055 sec. What's going on here?? Shouldn't Linux have better performance than Windows? Please help... thanks.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
GMP-ECM Linux ARM32 performance	VictordeHolland	GMP-ECM	7	2018-02-21 17:19
DC performance issue	cuBerBruce	Information & Answers	2	2017-07-10 13:43
Windows 7 Windows Update & Prime95 issue!!!	Unregistered	Information & Answers	14	2010-04-10 21:47
Performance tweak advice for Linux	joblack	Software	2	2010-03-21 15:56
Performance of Prime95 on Windows XP x64	Schamschi	Software	5	2009-09-20 20:45

2010-05-19, 13:46	#4
rajula "Tapio Rajala" Feb 2010 Finland 315₁₀ Posts	Are you sure you have installed 64-bit linux and using 64-bit mprime?

2010-05-20, 01:05	#10
__HRB__ Dec 2008 Boycotting the Soapbox 2⁴·3²·5 Posts	Hm...what happens when you start another instance of mprime? If the performance/process is unchanged on linux but drops by ~50% on windows it's obvious that we should check the configuration for multithreading (presumably you have a dual core processor).