Testing the Tests – CPU Stability
The System
The system that I will be using for this review will be:
- I7-930 D0
- Gigabyte GA-X58A-UD5
- 3x2GB OCZ DDR3 – 1066 @ 7-7-7-16 timings
- EVGA GTX 260c216
- Raid 0 on 2x150gb Raptors
The Test
I will be testing 3 programs that stress the CPU to maximum loads in order to test for stability. I will be using various monitoring programs to output results. The idea is that I’m trying to prove if any one program is “better” than the others. If stress program A runs for 10 hours, but program B fails after 10 minutes, it may be beneficial to use program B to quickly test for failures.
I initially ran into many problems created by the stock i7 cooler provided by Intel. Just as I had thought all hope was lost, UPS decided to stop by and give me some motivation. Meet the Swiftech Apogee XT:
This should prove to be useful for the massive amount of heat generated by the i7 930.
Meet the Programs
Prime95 stresses CPUs by attempting to solve for prime numbers. You can read up on the theory behind the program via the above link. This program does support testing each thread with only one client.
For Prime95, if a CPU passes 24 hours without failure, it’s assumed to be completely stable.
I don’t personally know too much about OCCT. In fact, I’ve never actually used it. It has the best graphic interface of all the programs listed here along with built-in monitoring and auto termination temperatures. OCCT is said to be much more sensitive to failures.
OCCT will be used in auto and custom mode.
Intel Burn Test
Intel Burn Test is my personal favorite. I’ve never tested it on an AMD CPU, but it states compatibility. The test uses Intel’s own Linpack testing. It states that it can provide load temperatures up to 22C higher than Prime95. The reason why I love this test is because it also provides a benchmark along with a stress test. Each run is reported and given the time that it took to complete. This provides me with knowledge if my overclock was helpful or harmful (throttling).
Intel Burn Test claims that 8 minutes of testing is roughly equal to 40 hours of Prime95.
Um…. Your testing of OCCT was not very thorough… You should have tested on both small and large data sets. Large is easiest to pass while small is not.
Think I might have to pull out IBT and see how it runs on my 955. Can’t say I ever ran it, but between Prime95 and OCCT it took a rather long time for each test before moving on.
Fully agree on reliable and expected. If I turn on my computer overclocked; never have an issue but one test some where fails causing a BSOD; I just move on, since if its playing all the games I play with no issue’s and doesn’t randomly crash then it working well
@ Lord XeB
I realize your comment on the OCCT test not being very thorough. I agree that OCCT could probably be as sensitive as IBT was, but OCCT was very unclear on which test to use. A good addition to OCCT would be a clear definition of each test. Test Mode Small Data Set nets you “The test wont touch the RAM.” You can conclude two things from this:
1) Best test for CPU since it focuses only on this
2) Is as effective on the CPU as the other two, but doesn’t test the RAM
All 3 tests did eventually fail – but as my conclusion suggests, IBT did it the fastest. In fact, the amount of time it took for both OCCT and Prime95 to fail was greater than the time it took me to run 1 loop of IBT at 191-194. It doesn’t make the program better, but it does make it the fastest.
@ Micheal Turner
I completely agree. I am a (in)famous OCer over at OCN (mostly because I have habit of destroying graphics cards D:), but I am VERY thorough with my stress testing. If it an overclock does not pass 24 hours of prime (to be deemed 24/7 stable) and 50 runs of LinX (AKA a better GUI’ed version of LinX) with max memory (only works with 64-bit systems) I do not deem it stable. <.< Little ruthless on my poor Q6600 but I do not stand for system instability either.
Anywho, very nice guide BTW. And I do agree with you on OCCT. It does need to be clearer, but for reference, P95 during default stress testing is set at small data set (or small fft or whatever they are called).