Extratasking during stress testing

[UltraTaco]

If you run prime or occt or whatever stress test and start doing something extra, like opening another stress tester or start playing video or audio, does it affect effectiveness of stress test?

Does it slow things down and in effect, hinders stress test? Or does it all pile up together and would throw error sooner, hence making it even better to find instability?

Reason I'm asking is, instead of just running prime small ffts, I fired up realbench heavy multi tasking and inage editing ticked on.

Another, probably similar question, running two different stress test programs together, does it make finding errors sooner due to diverse stress methods?

Thanx
Taco priming again..

--update---
I guess prime and realbench don't work right together something didn't go right as I've never seen realbench use that much ram while benchmarking (image edit+heavy multitasking)
View attachment 196141


interesting observation
I noticed temps drop like 5-7° Everytime I start a second stress tester like occt or realbench

 

[Blaylock]

Think of it from a logical point of view. If your CPU load is already at 100% and you open another 100% load will your new load become 200%? Of course not, it will just take longer to finish whatever it's doing.

As far as finding errors I think it would be completely circumstantial. While it might be quicker in one run it might be slower in another, again you are splitting the tasks and therefore slowing down both operations.

For your temp differential that one actually makes sense to me. The reason it is running at a lower temp is because P95 is designed to stress the temp to its upper limits via AVX. When you are running a non AVX instruction it will allow the CPU to cool just a bit.

This is just my interpretation of what's going on. I may be a little off or I may be completely off.

 

[UltraTaco]

Thank you, Blaylock, prime 26.6, no avx instructions and i7 920 doesn't have avx yet.

 

[Blaylock]

OK, must be something else then.

 

[Lochekey]

It probably has to do with the threads switching between loads and allowing cooling in between tasks. B it also has to do with the type of load. Small fft even without AVX will probably drool run hotter than Realbench.

 

[mackerel]

Talking of running multiple tests at the same time, that is what Realbench does by itself. It isn't one stress, but a selection of them going on.

Prime95 is a kinda special case. To correct one point, it's main optimisation is NOT to cause maximum heat, but to do the most work it in in the most clock cycle efficient way possible. That usually means it is also the most time efficient, if you have adequate cooling to avoid throttling. The bench option descriptions I find a bit ambiguous. Easy way to look at it is small FFT stresses the CPU cores and lower level local cache. Large FFT is less stress on the cores, but shifts more of it to the IMC and ram. Blend just does a mix of both, but it also invokes an option that means the data isn't held in one spot, and can spread throughout a ram range. You can also select how much ram it uses if you run custom.

Both P95 and Realbench have options in their stress tests to choose how much ram is used. Obviously don't pick values which would cause you to run out, as then you're only stressing your disk as it tries to cope with swapping stuff out. P95 default seems to be 2GB, Realbench 4GB, so that's 6GB already. OS would take a bit... and not too surprising you're running low. Personally I haven't seen any specific value in running P95 with really large amounts of ram unless you're testing RAM OC, so I wouldn't be too bothered about setting a large value there.

 

[Johan45]

I don't think it's a good idea to run 2 different CPU stress tests at the same time myself. It's like rying to drive in two directions at the same time. I do condone running CPU and GPU tesys simultaneously.

 

[UltraTaco]

Actually, realbench would never go that high on ram usage during benchmarks whenever I used it. Just seems like a freak case

When running stress test yes there are settings to use more ram, but benchmark doesn't use that much