ok let me first start off with a pretty interesting question... take a look at this: http://www.pccasegear.com/index.php?main_page=product_i...
this little gizmo promises to keep your fans spinning after the pc is shut down. is this actually necessary? does the cpu, gpu and other hardware stay at a high temp after being closed down, because although it isnt actually working any more, there is still heat left, and if the fan is turned off, is this a risk? a long overlooked problem? hmmmmmmmm
and a question thats been bugging me for a while: when i run a blend test in orthos or prime, my cpu is sitting at 100%, while stressing my RAM as well! so why bother with small fft's if the blend test stresses the CPU as much as possible?
"Blend" in fact does not stress the CPU as much as "Small FFT's". The proof is in observing the thermal signatures, which are the result of workload.
Download SpeedFan, click on the "Charts" tab, select "temperatures" then check the boxes for CPU (Tcase) and the Cores (Tjunction). Start Prime95 and see for yourself.
"Blend" creates varying thermal patterns, which reflects intermittent workloads. Shown below is Prime95 Blend, then Idle. CPU 49c peak, Core 1 54c peak.
"Samll FFT's" instead provides a high steady state thermal baseline, which reflects maximum workloads. Shown below is Prime 95 Small FFT's, then Idle. CPU 51c steady, Core 1 56c steady.
This is why Small FFT's are used for stability testing and temperature benchmarking. While stability testing is recommended for 8 to 24 hours, a temperature benchmark only requires 10 minutes, since thermal saturation is typically reached within 7 to 8 minutes.
Funny thing, I have been stress testing with the blend test a lot lately, and while I agree that small ffts is the way to go for CPU testing...
I have noticed that the most "stressful" test appears to be test 2 of the blend test, the 8 k blend test.
It takes like 30 mins to complete the first battery of tests that make up 1024k test, but afterwards, your temps will soar! As a matter of fact, I seem to get 1c higher during that test!
So basically, its small fft 8 k test on crack, pumped up on 2 + gigs of RAM.
Next up would be the 8 k small fft test, and the 32k-38k small fft test!
(Like comp said, it appears that blend simulates workloads, and will cycle hot for a test, then down a bit, then hot again. And that alone makes it not good enough for just CPU testing alone. It isn't continuous at maximum power and drain.)