Prime95 Stress Test - 4 instances for Quad Core? + How to use for me?

Max Dread · 2015-02-25, 11:29

Hi all

I'm running some testing on my PC. I have read in several places that it is best to run four instances of Prime95 if you have a Quad Core. However, I am wondering if this advice is a thing of the past and whether the current P95 automatically tests all 4 cores in a single instance of the app. Could anyone confirm?

Also, while I'm here... I am not an overclocker. I just want to make sure that my almost-silent PC is stable, solid and not too hot. So I was planning on just running it overnight. Would that be enough?

Also, it seems to me that there are benefits to all three of the different tests (Small FFT, Large FFT, Blend) so I thought I could run each one over different nights separately. I can then observes temps, and if all stays under my maximum (and there are no errors reported) move on. Does that sound like a plan?

Cheers

Max

Dubslow · 2015-02-25, 16:59

I don't have any idea where you could have read that multiple instances is best, because it's utterly not -- Prime95 is one of the most capable multi-threaded programs out there. It should automatically use all your cores, even just for stress testing.

24 hours is the usual number recommended for *rock solid* stability, though it is more superstitious wisdom than actually supported by empirical data. Three over night runs of each the the test types should be more than plenty to ensure your system is among the most stable in the world.

Max Dread · 2015-02-25, 18:41

Hi

Thanks you very much for the reply and re-assurance.

I encountered the idea of running more than one instance in a lot of places. I can back track and see if I can find them again if that woud be of any interest?

Ta

TheMawn · 2015-02-25, 18:52

We don't need to see who said that multiple instances is best. It's just not true. The only conceivable reason is to run different kinds of stress tests at the same time but that isn't necessary.

I would be more than confident with a blend test. Alternatively, you can run small FFT's for an hour or so just to make sure the temperatures are reasonable and then run a blend for 12 - 24 hours and you should be good.

If your machine is not overclocked and you're more oriented for silence, you should be more concerned about temperature than stability, so 12 - 24 hours is plenty.

Max Dread · 2015-02-25, 19:13

Great stuff... I'll forge ahead with the tests then.

Thanks for the clarifications.

TimSorbet · 2015-02-25, 21:44

Quote:

Originally Posted by Dubslow

I don't have any idea where you could have read that multiple instances is best, because it's utterly not -- Prime95 is one of the most capable multi-threaded programs out there. It should automatically use all your cores, even just for stress testing.

Quote:

Originally Posted by TheMawn

We don't need to see who said that multiple instances is best. It's just not true. The only conceivable reason is to run different kinds of stress tests at the same time but that isn't necessary.

Back when dual- and quad-core CPUs were new, Prime95 (version 24) only supported one thread per instance. Most likely, he was just reading some old advice. With recent versions of Prime95 (25 was released in testing in 2007 and a final release in 2009), you're right: only one instance is needed.

LaurV · 2015-02-26, 11:52

He may also have read about the 32 core limitations, so if you have a 80 cores machine, you would need 3 instances of p95.
Well...
Back to mortal stuff...

chalsall · 2015-02-26, 19:15

So, a few days ago my main workstation (not server) started having Firefox and Chrome crash randomly. Bloody hell...!

Run all the updates, reboot. Still random crashings of the browsers. Check my recent mprime logs... One "01000100" error code out of the last dozen runs; that run was confirmed good as the residual matched.

Open up the case, blow it out (many dead mosquitoes -- damn those little bastards!!!).

Boot. Open up a shell, run mprime stress tests -- 1 (small FFTs) runs fine; 3 (blend) dies within seconds. So, my CPU and MB is good.

Several reboots and RAM module swaps later, determine that my memory controller is fine, but I have one bad 4GB module. Down to 12 now. Will have to buy another -- my supplier's warranty on RAM is three months.

Thanks George! This would have taken much longer to figure out if I didn't have this tool!

TheMawn · 2015-02-27, 02:41

The ASUS fanboys at overclock.net absolutely swear by ASUS's own ROG RealBench stress test (the "Real" makes it real, see?). It's funny, because I had a 2 hour stable RB run that crashed immediately with Prime95. Prime95 continues to be my most trusted and most useful stress test.

I helped someone diagnose their issues with the simplest of instructions.

"Run a Prime95 blend test"

"It crashed"

"Try Large FFT's"

"It crashed"

"Try Small FFT's"

"24 hours no problems"

"Replace memory"

"That fixed it thanks"

I still think the whole small-FFT story is funny. It was the end of the world because Intel would have CPU's add a tenth of a volt or so by default when running AVX2 instructions and peoples' aggressive 1.5V overclocks would turn into "zomg wtf" 1.6V overclocks and draw 500W through the socket and, uh, cause some problems.

People go and say "Naw Prime95 overheats" yet say that their CPU's are 100% stable anyway.

LaurV · 2015-02-27, 06:07

Quote:

Originally Posted by TheMawn

I helped someone diagnose their issues with the simplest of instructions.
<snip>
"Replace memory"

2015-02-25, 11:29	#1
Max Dread Feb 2015 3 Posts	Prime95 Stress Test - 4 instances for Quad Core? + How to use for me? Hi all I'm running some testing on my PC. I have read in several places that it is best to run four instances of Prime95 if you have a Quad Core. However, I am wondering if this advice is a thing of the past and whether the current P95 automatically tests all 4 cores in a single instance of the app. Could anyone confirm? Also, while I'm here... I am not an overclocker. I just want to make sure that my almost-silent PC is stable, solid and not too hot. So I was planning on just running it overnight. Would that be enough? Also, it seems to me that there are benefits to all three of the different tests (Small FFT, Large FFT, Blend) so I thought I could run each one over different nights separately. I can then observes temps, and if all stays under my maximum (and there are no errors reported) move on. Does that sound like a plan? Cheers Max

2015-02-26, 19:15	#8
chalsall If I May "Chris Halsall" Sep 2002 Barbados 2×7²×113 Posts	Yay Prime95/mprime stress testing! So, a few days ago my main workstation (not server) started having Firefox and Chrome crash randomly. Bloody hell...! Run all the updates, reboot. Still random crashings of the browsers. Check my recent mprime logs... One "01000100" error code out of the last dozen runs; that run was confirmed good as the residual matched. Open up the case, blow it out (many dead mosquitoes -- damn those little bastards!!!). Boot. Open up a shell, run mprime stress tests -- 1 (small FFTs) runs fine; 3 (blend) dies within seconds. So, my CPU and MB is good. Several reboots and RAM module swaps later, determine that my memory controller is fine, but I have one bad 4GB module. Down to 12 now. Will have to buy another -- my supplier's warranty on RAM is three months. Thanks George! This would have taken much longer to figure out if I didn't have this tool!

2015-02-27, 02:41	#9
TheMawn May 2013 East. Always East. 11×157 Posts	The ASUS fanboys at overclock.net absolutely swear by ASUS's own ROG RealBench stress test (the "Real" makes it real, see?). It's funny, because I had a 2 hour stable RB run that crashed immediately with Prime95. Prime95 continues to be my most trusted and most useful stress test. I helped someone diagnose their issues with the simplest of instructions. "Run a Prime95 blend test" "It crashed" "Try Large FFT's" "It crashed" "Try Small FFT's" "24 hours no problems" "Replace memory" "That fixed it thanks" I still think the whole small-FFT story is funny. It was the end of the world because Intel would have CPU's add a tenth of a volt or so by default when running AVX2 instructions and peoples' aggressive 1.5V overclocks would turn into "zomg wtf" 1.6V overclocks and draw 500W through the socket and, uh, cause some problems. People go and say "Naw Prime95 overheats" yet say that their CPU's are 100% stable anyway. Last fiddled with by TheMawn on 2015-02-27 at 02:41

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Legal to implant prime95 as a stress-test in another software	SvenBent	Software	1	2014-10-07 17:36
Dual Core to Quad Core Upgrade	Rodrigo	Hardware	6	2010-11-29 18:48
exclude single core from quad core cpu for gimps	jippie	Information & Answers	7	2009-12-14 22:04
Quad Core	R.D. Silverman	Hardware	76	2007-11-19 21:57
Optimising work for Intel Core 2 Duo or Quad Core	S485122	Software	0	2007-05-13 09:15

2015-02-25, 16:59	#2
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 1C35₁₆ Posts	I don't have any idea where you could have read that multiple instances is best, because it's utterly not -- Prime95 is one of the most capable multi-threaded programs out there. It should automatically use all your cores, even just for stress testing. 24 hours is the usual number recommended for rock solid stability, though it is more superstitious wisdom than actually supported by empirical data. Three over night runs of each the the test types should be more than plenty to ensure your system is among the most stable in the world.

2015-02-25, 18:41	#3
Max Dread Feb 2015 3 Posts	Hi Thanks you very much for the reply and re-assurance. I encountered the idea of running more than one instance in a lot of places. I can back track and see if I can find them again if that woud be of any interest? Ta

2015-02-25, 18:52	#4
TheMawn May 2013 East. Always East. 11×157 Posts	We don't need to see who said that multiple instances is best. It's just not true. The only conceivable reason is to run different kinds of stress tests at the same time but that isn't necessary. I would be more than confident with a blend test. Alternatively, you can run small FFT's for an hour or so just to make sure the temperatures are reasonable and then run a blend for 12 - 24 hours and you should be good. If your machine is not overclocked and you're more oriented for silence, you should be more concerned about temperature than stability, so 12 - 24 hours is plenty.

2015-02-25, 19:13	#5
Max Dread Feb 2015 3 Posts	Great stuff... I'll forge ahead with the tests then. Thanks for the clarifications.

2015-02-26, 11:52	#7
LaurV Romulan Interpreter "name field" Jun 2011 Thailand 2827₁₆ Posts	He may also have read about the 32 core limitations, so if you have a 80 cores machine, you would need 3 instances of p95. Well... Back to mortal stuff...