Discussion in 'Intel x86 CPUs and chipsets' started by chainbolt, Nov 15, 2008.
Why isnt prime using very much ram anymore
Finally seem to have come up with a decent overclock for my i7 930. Have had lots of trouble with overclocking it much at all, I thought I might have had faulty ram even though they wouldn't show errors in memtests and such (sometimes a stick wouldn't show up on a cold boot). Updating the bios and cranking up the V-core up has finally given me a stable 4.2Ghz. I had been running it at 3.36 for months and finally today decided to give it a go again. Currently don't have hyperthreading enabled. Might see try that on next.
EDIT - Yep hyperthreading works
Bern Edit: Abuse = ban, see you in 24 hours.
it puts more strain on the processor significantly faster then prime95 does, it strains it in a way which is unique compared to other mathematical iteration based stress testing, and that is evident of the higher temperatures its capable of getting the processor to in that significantly shorter period of time.
its a more practical way of determining that the cpu is stable, and if you cant see that then its not my fucking problem. enjoy your slow cpu stress testing.
I always use Linx myself, I find it stresses my CPU a lot more than P95
P95 might be enough for some, but I do like my system to be perfectly stable hence my preference for Linx
He probably used linpack for the purpose of getting the max temperature.
That's just it. With LinX it is not perfectly stable. LinX does a shit job of testing your other compnents.
I too agree that LinX/IBT are not good choices for stress testing.
I use LinX as a preliminary test, to narrow voltages. If my settings do 25 passes without error, it's time to start a 24 hour Prime95 run.
The CPU I'm using now can do 25 passes of LinX with 1.15volts, but fails Prime95 within 7-8 hours. It needs 1.17v for 24+ hours of Prime95. (I've found it usually needs to be bumped 2-3 notches to be stable in Prime95, and usually 1-2 notches on the QPI)
I do a lot of video encoding and 3D rendering, and used to occasionally get crashes due to an unstable overclock. (Was testing stability with LinX only, because I thought Prime95 wasn't as good and took bloody ages)
After reading the merits of Prime95, I decided to try a run on my system. It failed after 7 hours.
From then on, I have always used 24 hours of Prime95 as the final stress test. Since I started this, I haven't had any crashes.
Basically, what I've found is this:
Completing x passes of LinX does not mean it will complete 24 hours of Prime95
Completing 24 hours of Prime95 means it will complete x passes of LinX
(Where x is any number > 0)
I will also post the test screenshot of my new cpu when I get home tonight.
This man speaks the truth and sounds like from experience. Unlike others who follow blindly like sheep.
And here is where I'm at currently:
I haven't tried over 4GHz yet, but it looks like it's got some headroom in it
That is a damn nice 920 you have there Radiactor.
I've had chips that can run prime 95 all day yet crash on LinX/IBT and vice versa.
I have never ever seen a chip that passes 8+ hours of prime, fail ANYTHING!
I've seen both happen. I use LinX to get a rough guide and then if I can be bothered, I run prime95. Been running Prime95 less and less though, heaviest real world load my PC sees is gaming, doing a LinX stress with all available RAM available seems to be enough for my needs.
Yeah basically the same with me. Run LinX when I'm working on an overclock then when I'm happy an think it's stable 12 hours or prime to be sure. They both serve there purpose. What shits me is when people use it as a benchmark comparing gflops. That's nonsense.
I wouldnt use it as a benchmark, theres far more appropriate things to use.. like a benchmark program.
but yes, it is the perfect tool to specifically test your cpu stability.
I use it for CPU and HyperPi 32M test for ram, to determine quickly that their stability is sufficient, when tweaking and overclocking, and I use Prime as a final overall stress (PSU/CPU/MOBO/RAM/etc).
P95 blend test is a great test for a broad test, but slow as shit for when your tweaking.
10 max memory Linpack passes (comparing residual norm) and a hyperPI 32M is more then enough to determine that your core specs are stable, only takes 1 - 3 hours to be very thorough instead of 8+ of blend stress testing.
Bern Edit: Abuse + trolling = ban, see you in 7 days.
Just curious about vtt (qpi dram in my case)
Keeping it as low as possible to minimise extra heat is the aim yeah?
Or overclocking to 4ghz+ simply required quite a bit?\
Reason i ask is that i overclocked my 920/R2E and while didnt do my full test ( max bclock etc) i tested to 4ghz just so i could get it going. Got it stable with 21x191 :
Vcore - 1.3 (R2E bloats it to 1.323 under load)
QPI DRAM - 1.2
IOH - 1.1
ICH - 1.1
Ram - 1.5
Ive read quite alot over the past week seeing if perhaps my methodology was wrong by starting off at lowest and increasing if/when required as pretty much every other x58 clocker seems to hit 1.25/1.275/1.3+ straight up and then start o/c.
Im guessing that once i start pushing the bclock 200+ it will start asking for a bit more juice. ( Thats my plan this w/e to chase my bclcock max and ram etc).
But is anyone else using minimum amounts of vtt ?
Right now Im on 1.43 with vtt/QPI. 1600mhz ram and 4.2 ghz testing. Meanwhile, does anyone know what the error code D1 means? Im guessing it is the QPI VTT or memory.