best CPU stress-testing software?

V

venm11

2[H]4U
Joined
Oct 6, 2004
Messages
2,236
What's the best cpu stress-testing software?

I've read about superpi and others, but it seems to me that all of these stress the CPU differently. I'd really like one that stresses all execution units, burst modes (mmx/sse) memory, cache, speculative execution, everything at once, PLUS multiple threads for different cores. I want to see temperatures beyond anything I've seen.

Also... it seems to me that none of this software tests for corruption, we instead rely on observations that the system is locking up or getting unstable. Wouldn't it be good to have some workload that generates a data set, that can be compared against a reference for signs that there are cpu or memory errors? What about cache ecc statistics? The only way I've been able to do this previously is watching memtest86+ for individual memory errors, but that's only partially testing the cpu.
 
I think orthos is the one you want, My X2 hits 100% plus really heats it up more than anything else i've tried on it thus far. It will also do both cores with only one instance open.
 
I was wondering the exact same thing. Orthos is the gold standard, but AFAIK that's mostly a FPU stress test, which does compare generated #s to known good data. If Orthos (and Prime95) only stress the FPU, and whatever else may happen when accessing the FPU, is it really that effective in testing the entire CPU?

Also when I was using a C2D E6300, if I had one instance of Orthos running stressing both cores, I could open up an instance of Intel TAT and start that, and my load temps would shoot up another 5C and the system would become unresponsive - that seemed to me to be a real load test. If I have Orthos open and running by itself, my system is still responsive and I can work on it with Orthos in the background - application priority needs to be changed?
 
interesting... .thanks guys.

I guess that's true.. you could run multiple stress tests at once...
 
Any of you guys have an "official" link for orthos? I can only google up websites that I'm not familliar with.
 
Change the priority in Otho's, and your PC will not be very responsive.
I found another program called S&M that definitely raised my temps higher than Otho's and even started to force the PC P&C 510 P.S. to increase it's ripple voltage.

But in all reality, I believe Otho's is a real representation of just about any type of stress your CPU would experience.
 
I agree; Orthos simulates the kind of stress that might actually happen in real-world use, but it does it continuously for testing purposes, whereas under normal circumstances none of us would have our CPU usage pegged at 100% for more than an hour or so at a time, max.

If your concern is heat, and you really want to find out what your absolute temp cieling is with the CPU being stretched to the very limit, just use TAT. I too have found that running TAT's torture test gets me load temps ~4C higher than Orthos. I doubt you're going to find any single program that stresses a chip as effectively as TAT.
 
Crooked Paul said:
I agree; Orthos simulates the kind of stress that might actually happen in real-world use, but it does it continuously for testing purposes, whereas under normal circumstances none of us would have our CPU usage pegged at 100% for more than an hour or so at a time, max.
Click to expand...

I can have CPU usage at 100% for hours at a time when doing MPEG2 encoding or After Effects rendering. I may be slightly atypical but with the popularity of home video these days I bet a lot of people experience hours of full load as well.
 
In summary I was saying that ortho's simulates how the cpu would operate, be it at 100% or not. The program I mentioned above operates the cpu in a manner that most application do not. TAT I don't know how it uses the cpu, but I would guess it's close the S&M, but they have there place. TAT can show you good or how bad your setup will operate in say a ~100F temp under 100% stress.
 
Plz post a link to the Orthos website? Google didn't help for some reason
 
theelectic said:
I can have CPU usage at 100% for hours at a time when doing MPEG2 encoding or After Effects rendering. I may be slightly atypical but with the popularity of home video these days I bet a lot of people experience hours of full load as well.
Click to expand...

Good point, man. When I do encoding it doesn't peg the processor, but I'm generally making movies to go on my iPod so they're small and maybe less CPU-intensive. Anyway you're totally right, there are lots of times when you might be at full load for hours.
 
My point was that even if the CPU is pegged at 100%, that doesn't mean it's doing everything it could be doing, given that different instruction units may not be utilized. The CPU is able to make these units all go at once, we just need some code to make it happen. Most of these tools are mainly streams of FP calculations.

I remember reading an article about FEAR which described the unusual heat that was being generated due to it's level of optimization. A gx card manufacturer was quoted as saying they'd "never seen anything like it". Fear is also known to be "overclock-sensitive". This is what we want!

But... Orthos seems to be working pretty well :)

PS. will TAT work on AMD64?
 
I'm working on a version of Toast for the Core 2 Duo that should interest you. I've written plenty of assembly language and I know how to make these things work.

Basically the idea is to get the CPU doing as much as possible, and on the Core 2 Duo there is a 66% increase in Peak IPC over the Athlon 64.

I've got a version right now that is peaking at 4.555 IPC, using macro-op fusion on missed braches (there is no penalty), and the goal is to get as close to 5.0 ICP as possible. Its also streaming 16-bytes of data from the L1 cache into SSE registers on each of the 16 clock cycles out of 18 total; the L2 bandwidth is nearly pegged at 45 GB/sec.

All of the "stress testing" software out there; Prime, Orthos, and others, may peg the CPU at 100% "workload" but never utilize the chipss maximum efficiency; most software is probably running at around 2.0 IPC overall, and not stressing the cache so much as it is stressing memory, which causes the CPU to sleep (and cool down) while waiting for data...

Anyway, with this current version I'm working on, I've already crashed my CPU within seconds about 4 times, any more more after just a few minutes... and this computer has otherwise been completely stable at anything I've thrown at it; I've been running at 3.4GHz for days now and Toast is the only thing that can bring it down. Idle temperature is about is 28 degrees C and under load its 58 degrees C. It takes about 3 seconds to get that hot.

I'm working on it every day and should be ready soon; please bookmark this site and check back regularly:
http://www.nanerbeet.com/toast


Maybe by the weekend?
 
If C2D gets around 4.55 IPC, what is it on Athlon 64 X2? I tried two instances of the Toast program and it didn't make any difference in my maximum load temps compared to Orthos. Everything seemed stable (no crashes) in the 2+ hours I ran it as well.
 
The version of Toast that is up right now is optimized for the AMD pipeline and achieves 3.0 IPC; the maximum possible, and doesn't run so well on the Core 2 Duo; it actually only achieves 2.4 IPC; less than half the target of 5. As soon as the update is ready I'll post it on the web-site. The biggest thing right now is merging the two code paths so that Toast can detect whether you have an AMD processor or Core 2 Duo and select the appropriate instruction sequence.


So, just to be clear, the current version on my web site doesn't do any good on Core 2 Duo. The version coming at the end of the week will :)
 
Should have mentioned I'm using AM2 X2 3800+ :) Is it supposed to get hotter than Orthos w/AMD?
 
Yes; or at least Orthos or Prime will never get any hotter than Toast; depends on how aggressive your cooling is and how high the voltage is. Put simply, the trick to getting it "hotter" is to exceed the cooling capacity of the heatsink.

Next time you leave the computer for a while, try running both instances with animation turned off and priority set to maximum.
 
old skool, I am very impressed with your coding kung fu. How do you indemnify yourself against lawsuits / flame wars after people d/l your software and fry their chips? :)
 
Jasonxxx said:
Change the priority in Otho's, and your PC will not be very responsive.
I found another program called S&M that definitely raised my temps higher than Otho's and even started to force the PC P&C 510 P.S. to increase it's ripple voltage.

But in all reality, I believe Otho's is a real representation of just about any type of stress your CPU would experience.
Click to expand...

I've just finished subjecting my rig to its last round of S&M, and I have to say this is the toughest stress test I've found yet. It tests CPU + RAM + PSU; it definitely heats up the processor to up to 5 degs C more than any other stability tester, too.

I highly recommend it!
 
OCCT is a new stress test & the best tool I have found yet for quickly finding instabilities, far better IMHO than Orthos which is strictly an FPU test. OCCT runs an automatic test that runs in 30 minutes and finds instabilities that Orthos would not find after 8 hours. OCCT heats up the cores a lot more too almost the only way I have ever been able to pass 50C is running OCCT. OCCT does a lot of memcpy operations too not just FPU tests.

Frankly I am not all that impressed by Orthos it is a very narrowly focused stress test. The small fft test that everyone seems to favor is pretty ridiculous. Maybe if you are just using the PC for protein folding or something that is all you care about. It is like trying to benchmark a system by only looking a dhrystone and whetstone measurements alone.
 
ive had my machine specs in sig, running orthos for around 8 hours now and it hasnt had any errors (i havnt overclocked it yet) and it has an increase in temps of about 10 degress @ idle at 30 and full load at 41 or so.
 
You must log in or register to reply here.
Back
Top