Setting Prime95 to use same calculation method over a period of time?

Monberg75 · 2017-03-10, 06:44

Hi.

I'm a hardware reviewer, and have noticed under the torture test, that the load on the CPU is different at each test.

That might be correct, I guess so, as i can see it on the numbers.

But I would like to have the same load on the cpu all the time, to get some coherent results.

So, let's say I would like to run M7471105 for like a million iterations.

Would that be possible? And would that be harmfull to my CPU?

The goal is to make the exactly same amount of heat**load on the CPU in a constant period of time.

I havent found any program that can do that, it seems like the load/heat is changing all the time?

Is there something I'm missing?

Hope anyone can help me.

Best regards: Monberg75

mackerel · 2017-03-10, 08:37

When I check power usage I do the following:

In the torture test window, select Custom.
Number of threads to run = same number as physical cores, so it runs one task per core. You can double that for HT/SMT.
Min FFT size, Max FFT size, set both to the same value so it doesn't change. This sets the working size of the data, and results will vary depending on the value. I'd suggest two values done separately, and use 64 and 1024 myself. The value impacts the amount of storage needed. At 64k FFT, 512k of ram is needed per thread, which easily fits in on-processor caches and the CPU execution is effectively unrestricted. At 1024k, it uses 8MB per thread, and running multiple threads it wont fit in processor cache other than those few with L4 e.g. i5-5675C. Where the work doesn't fit in processor cache, it will be impacted by ram performance. e.g. a system with fast ram configuration can run hotter as it is less restricted by ram performance.

I normally leave Run FFTs in place set for power measurements, which uses ram normally. If unchecked, the data is moved around more to stress different parts of ram. Not really looked into impact of doing this.

In many cases Prime95 will run a CPU hotter than other workloads, but it is safe providing the system has reasonable cooling and isn't overclocked to insane levels. For weaker cooled systems, keep an eye on CPU temps and possible throttling.

Prime95 · 2017-03-10, 09:34

The problem with getting a constant load using torture test is that at the end of each test / beginning of the next test there is a drop in the load as prime95 reinitializes the FFTs for the next test.

You could do Advanced/Test and enter a prime number near 7471105. If you've used Test/Worker Windows to have prime95 use all cores you will get a constant load for a good long time. If Test/Worker Windows is configured to run one core per test, then you'll need to do one Advanced/Test for each core.

Hope that helps.

2017-03-10, 06:44	#1
Monberg75 Mar 2017 1₁₆ Posts	Setting Prime95 to use same calculation method over a period of time? Hi. I'm a hardware reviewer, and have noticed under the torture test, that the load on the CPU is different at each test. That might be correct, I guess so, as i can see it on the numbers. But I would like to have the same load on the cpu all the time, to get some coherent results. So, let's say I would like to run M7471105 for like a million iterations. Would that be possible? And would that be harmfull to my CPU? The goal is to make the exactly same amount of heat**load on the CPU in a constant period of time. I havent found any program that can do that, it seems like the load/heat is changing all the time? Is there something I'm missing? Hope anyone can help me. Best regards: Monberg75

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
deletion disabled during the edit period	S485122	Forum Feedback	18	2014-09-11 06:55
Prime95 26.x credit calculation	ixfd64	Software	1	2011-02-25 00:04
Find the Period	davar55	Puzzles	22	2008-09-26 08:12
Calculation of P90 time for each exponent	bayanne	PrimeNet	5	2006-01-07 15:54
calculation time not added to account-stats	ulikilian	Software	1	2004-10-06 23:10

2017-03-10, 08:37	#2
mackerel Feb 2016 UK 2³·5·11 Posts	When I check power usage I do the following: In the torture test window, select Custom. Number of threads to run = same number as physical cores, so it runs one task per core. You can double that for HT/SMT. Min FFT size, Max FFT size, set both to the same value so it doesn't change. This sets the working size of the data, and results will vary depending on the value. I'd suggest two values done separately, and use 64 and 1024 myself. The value impacts the amount of storage needed. At 64k FFT, 512k of ram is needed per thread, which easily fits in on-processor caches and the CPU execution is effectively unrestricted. At 1024k, it uses 8MB per thread, and running multiple threads it wont fit in processor cache other than those few with L4 e.g. i5-5675C. Where the work doesn't fit in processor cache, it will be impacted by ram performance. e.g. a system with fast ram configuration can run hotter as it is less restricted by ram performance. I normally leave Run FFTs in place set for power measurements, which uses ram normally. If unchecked, the data is moved around more to stress different parts of ram. Not really looked into impact of doing this. In many cases Prime95 will run a CPU hotter than other workloads, but it is safe providing the system has reasonable cooling and isn't overclocked to insane levels. For weaker cooled systems, keep an eye on CPU temps and possible throttling.

2017-03-10, 09:34	#3
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 7³×23 Posts	The problem with getting a constant load using torture test is that at the end of each test / beginning of the next test there is a drop in the load as prime95 reinitializes the FFTs for the next test. You could do Advanced/Test and enter a prime number near 7471105. If you've used Test/Worker Windows to have prime95 use all cores you will get a constant load for a good long time. If Test/Worker Windows is configured to run one core per test, then you'll need to do one Advanced/Test for each core. Hope that helps.