Jump to content
Sign in to follow this  

Prime: Small FFTs or Blend, to check 100% stability ?

Recommended Posts

The different tests in Prime95 reveal different things about your system. Whiich you use depends on what you are trying to learn about your rig. When I get a new chip, I test as follows.


Initial testing - Vcore at stock (as determined by BIOS). Increase HTT with memory at very conservative settings (loose timings, divider, 2T). Test each jump with SuperPi 1M. Continue raising clockspeed until it won't pass 1M. Back off ~100MHz, run Prime95 Small FFT's, adjust HTT until I can pass minimum 30 mins. Rarely have the patience to go more than an hour - I'm just looking for an idea of what I could get at stock at this point.


Second phase - Raise Vcore to a little below my comfortable max. Continue upping HTT, checking with SPi 1M until I hit a wall. Then change multiplier and HTT (accordingly) to see if different combinations at roughly the same frequency can get me any further. Once best combination of multiplier/HTT are found, I bump the Vcore that up that last bit and see if I gain a significant amount. If not, I bump it down to see how low I can go and still maintain this highest 1M clock.


Third phase - CPU stability. Back clock down ~100MHz. At this point, I change the divider to where I plan to run it and go to 1T. (My logic here is that this is where I want my mem anyways, so may as well get a sense of the Mem Controller. Still keep loose timings.) Now the slow phase. One run of SuperPi 32M as a quick mem check. If no go, I revert to previous RAM settings. Then Prime Small FFT, fail, reboot, adjust, rerun, fail, adjust... until I find a speed where I am stable 10 hours. Though I ultimately shoot for 8 hours, this lets me know my CPU frequency will not be the source of errors in future testing.


Fourth phase - memory optimization. I now set very aggressive timings in BIOS. Run SuperPi 32M to see if I have a chance. If fail, loosen up. Once OK, I test further with Prime in place Large FFT's. Once I get a couple hours to pass, I know my memory is ready for more intensive testing.


Last phase - system stability. Time for Blend test. Failure during a small FFT test (like 12K) often means temps got too high, so I adjust air flow, lower clock, or lower Vcore. Failure during a Large FFT test (like 768K) indicates mem problems, so it's back to BIOS. I frequently encounter errors in Blend that the previous tests did not detect.



Anyways, my point is that all the flavors of Prime95 tests have their own uses. Each tells a different tale, and each is useful for diagnosing differnt things. A quick note about MemTest86+: for me, all it's good for is revealing instability, it tells me nothing about stability. Can run my mem at 318MHz 2.5-3-3 in MT86+, but it has been a MAJOR struggle to get just 2.5-4-3 at 297MHz in Windows! This mistrust holds for MemTest for Windows too (although it is much better). Last night I had over 3000% coverage in WMT, 0 errors. Stable? Nope. Couldn't pass Prime Blend for more than 50 minutes. Kept failing during a Large FFT run, suggesting I still had a memory problem. Loosened a couple things, then ran Prime Custom (500K to 4096K to isolate RAM). Finally got it to pass for several hours. Went back to regular Blend, and all was well. 8+ hours Blend stable.


Gamers, yes, looping Futuremark (or whatever) stresses well. Further, it seems like a highly appropriate stability test, given it is exaxtly how you use your rig. If your games don't crash, what else do you need? Just watch out for data corruption - your mem timings may be stable enough to run your games, but still manage to ruiin your MP3's!


Benchers, well, hell, even passing SuperPi 32M is probably overkill for y'all! But again, it makes sense for how you use your system. There was a funny comment at XS the other day - someone complainded that a bencher's rig wasn't prime stable; bencher essentially said "no freakin way I'm sittin there droppin fresh dry ice pellets into my mousepot for 8 freakin hours". Prime in this sense is pointless.



Last note - in the early phases I often boot into Windows with unstable RAM. This inevitably corrupts my Windows installation. Acronis True Image comes in VERY handy at this point. I keep a clean one saved (my initial install at stock speed), and when things get screwed up, I just restore a fresh one. Very cool - disposable Windows. And yes, Acronis works with RAID (whereas Norton Ghost does not).


Don't know if any of this is useful, but there it is.


Peace to all.

Share this post

Link to post
Share on other sites

Depends on your definition of stability. Some say 8 hours, some say 24. Some say certain programs, some say others. My opinion is that if I can fold and run prime torture test for 8hrs, and pass the 3dmarks, and most important, no crashing, then it's stable. If you're running a bunch of benchmark's then a few extra hours running both won't hurt.

Share this post

Link to post
Share on other sites
Guest Crusader

ixtapalapaquetl, thanks for the tip about that Acronis True Image software. Will check it out and I'm sure it will be very useful to me next time I corrupt Windows due to my OC efforts. :

Share this post

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this