cancel
Showing results for 
Search instead for 
Did you mean: 

Which Stress test is more representive of real applications?

DaGengster
Level 7
So Im trying to OC my CPU and based on 3 different stress testers, I have found 3 different configurations all with the same system that keeps the CPU stable and "safe" temps

My hardware setup is:
i5 4690k
midrange cooler (but my ambient temps are on the high end)
16 gb DDR3 1600 RAM
GTX 980ti
750W PSU

Stress tester 1:
Prime95 v2.87 with small FFT
Could only pull max 4.2ghz 1.04v before the temps went past what I call "safe"

Stress tester 2:
Prime95 v2.66 with blender
Highest clock I could pull was 4.5ghz 1.20v, temps were still under my "safe" temp

Stress tester 3:
Asus Realbench
Again, I could pull was 4.5ghz 1.20v. I tried 4.6ghz with 1.22v but the temps were slightly past my "safe" temp mark.

From what I understand, eventhough all the testers jack the CPU to 100%, they do it differently.
Tester 1 uses only small pieces of data therefore the CPU has almost no missed cache cycles and therefore has to run harder. Test 2 has a mix of small and large so there is missed cache cycles. Tester 3, idk what it is in terms of data size. Its rendering an image by testing how individual rays act so Im guessing its mixed?
So my question is, which stress test is actually more representative of real applications like games and CAD software?
The reason for me asking this is bc if normal programs dont actually use as much small cache data points as the FFT test, then the actual "use" of the data I got from the test is pretty useless. In which case, I would use the data I received from the other tests instead.
Info I found on the cache size thing if you want to read more about it:
http://superuser.com/questions/981466/in-prime95-why-do-small-ffts-generate-the-most-heat-despite-cp...
16,477 Views
26 REPLIES 26

gupsterg
Level 13
For OC in my sig but 4.3GHz cache I did 8hrs RB / 48 loops x264 v2.06 / 24hrs f@h and have had no issues in gaming / daily use.

When I upped the cache to 4.4GHz I did 4hrs RB / 48 loops x264 v2.06 / 24hrs f@H, this was several months ago.

With this 2nd chip I never used P95 (any version) TBH and don't think I'm missing a "stability test" angle.
Intel Defector :eek: AMD Rebel


R9 5900X - Custom WC - ASUS Crosshair VII Hero WiFi - Ballistix Sport LT 2x16GB 3800MHz C16 - RX 6800 XT - WD SN770 2TB - 2x 870 EVO 4TB


24/7 OC: i5 4690K @ 4.9GHz CPU@1.255v 4.4GHz Cache@1.10v - Archon SB-E X2 - Asus Maximus VII Ranger
Sapphire Fury X (1145/545 ~17.7K GS 3DM FS)

:eek: CPU Validation 5.198GHz@1.314v with 4.4GHz cache + RAM 2400MHz@1T :eek:
Da Music video

Arne_Saknussemm
Level 40
DaGengster wrote:
which stress test is actually more representative of real applications like games and CAD software?


The clue is in the name 😉 RealBench for sure...tests your system holistically and non destructively....

Stability at the end of the day is running what you need to run day to day day after day...only way to test is to do it...

However RealBench is a good start point for that real world testing 15 min 1 hour and then stop and get on with the real testing....

gupsterg
Level 13
I agree that RealBench stress mode is a really good test, if you compare it with say P95 / AiDA64 / XTU the GPU is not in use. So logically in my mind it's clear that RealBench is using more of the PC.

The only other thing that I do prefer over it is f@h, just because it's producing/contributing to a cause when being run. I set it to run on CPU & GPU, IIRC the loading can variate at times. So in my mind the yoyo effect of loading is also another type of stability test, which is similar to everyday use.
Intel Defector :eek: AMD Rebel


R9 5900X - Custom WC - ASUS Crosshair VII Hero WiFi - Ballistix Sport LT 2x16GB 3800MHz C16 - RX 6800 XT - WD SN770 2TB - 2x 870 EVO 4TB


24/7 OC: i5 4690K @ 4.9GHz CPU@1.255v 4.4GHz Cache@1.10v - Archon SB-E X2 - Asus Maximus VII Ranger
Sapphire Fury X (1145/545 ~17.7K GS 3DM FS)

:eek: CPU Validation 5.198GHz@1.314v with 4.4GHz cache + RAM 2400MHz@1T :eek:
Da Music video

kaufen
Level 7
If you are overclocking, the standard that most people go by is running Prime95 for 24 hours without crashes. The "Prime95 v2.66 with blender" is probably representative of real programs more than the small FFT version.

Retired
Not applicable
realbench+prime95+intelburntest+vampire diaries (season 2)

running them simultaneously (all at once) , a single stress testing software just isnt enough

WhitePaw wrote:
realbench+prime95+intelburntest+vampire diaries (season 2)


Really season 2! damn it I will have to run that all again! I thought I passed too easily...

55841

WhitePaw wrote:
realbench+prime95+intelburntest+vampire diaries (season 2)

running them simultaneously (all at once) , a single stress testing software just isnt enough


WhitePaw good to see you, how's life been treating you

Definitely RealBench and Aida 64 running at the same time. I'd skip Prime

WhitePaw wrote:
realbench+prime95+intelburntest+vampire diaries (season 2)

running them simultaneously (all at once) , a single stress testing software just isnt enough



Now this seems a bit crazy running all at once...


Conenubi701 wrote:
Definitely RealBench and Aida 64 running at the same time. I'd skip Prime


I know this is an OLD POST! 😉

But seriously, I'd like to know, is this really good to do, run multipile stress apps at the same time?

I also thought Aida isn't that good?

For now I've been only running 1 hour of RealBench, and I did come from the days before of 24 hrs of Prime, so not sure where all that stands now...

Been wondering what apps if any besides RealBench I should also stress/test with?

OCCT
Intel XTU
Prime
etc....

Hmm