Unstable OC or just a good troll

[Midland Dog]

I've been playing with my CPU cache as i only had it at 100mhz above stock (devils canyon i5), so far im at 4.3ghz, prime95 on small FFT is stable for 30 mins, x264 is stable for 2 hours. But youtube at 8k24? yeah nah bsod/hardlock within 5 seconds, any ideas?

 

[hat]

Is there an x265 test you can run instead?

 

[GorbazTheDragon]

Which p95 version?

 

[John Naylor]

Unless you are trying to get on a website leader board, I don't see any value to testing OCs with P95 in post Sandy Bridge era since the unrealistic load limits the OC and 24 hour stable P95 won't necessarily get ya thru a multi-tasking stress test.

But as o your question ...

a) The general recommendation for CPU cache is 3 less the ya core multiplier. So if ya have 4.8 Core multiplier, then 4.5 for the cache.

b) I have tested this rather extensively on Haswell / DC and found that the only thing impacted were graphic programs. With 4.8 core, using 4.5 - 4.8 on the cache gave same results. When i dropped to 44 and below, there was a performance impact.

c) I could not determine any performance impact in any other type of stress test. Leaving cache at default had no impact on gaming.

d) Like core increases ..., raising the cache voltage is generally required to get to a higher cache multiplier and keep thing stable. To get stable at a higher cache and slight increase or core voltage may also be required.

I would recommend stress testing with RoG Real bench ... it's much better at picking up unstable OCs and faster to do it. Have had 24 hour P95 stable OCs fail in < 45 minutes. When you get thu the 8 minute multitasking benchmark, your voltages are very close to where they need to be. If you can get thru 4 hours with max memory, your system should survive anything. CPU temps will also be a lot lower.

http://dlcdnmkt.asus.com/rog/RealBench_v2.56.zip

 

[Kursah]

I always ran my cache at 4000MHz, with an 8X multi so it dropped to 800MHz. In testing, when at 4.8GHz core on my good 4790K, I noticed no measurable difference between 4000 and 4500 on cache, but my stability was much much better at 4000, so I went with the stable route. I was able to drop my CPU input voltage from 1.75 to 1.7 or 1.65 iirc. I'd have to check. Either way, I agree with John, use RoG Real Bench, I also use OCCT and Cinibench R15 as well.

 

[Midland Dog]

Which p95 version?

26.6 before avx melting, and ill see if i can find an x265 bench. also will try rog stress thingy. My cache is still 500mhz off core rn, i dropped it from 4.3 to 4.2 and my core clock is 4.7 all core

running real bench now, got a movie on in the background so if it hardlocks i notice

VLC is dropping frames like crazy right now

 

[EarthDog]

26.6 before avx melting,

Which makes Uncle John's warning a bit useless.

To test and see if it is the overclock/system being unstable and not a playback issue, go back to stock and test.

 

[Midland Dog]

yeah nah with cache at 39 its fine, 15 min preliminary test on realbench was stable max temps were 83c is that any good for no delid, 1.38 vcore and a 360mm rad? also is it worth spending time to properly tune memory timings, i target 144fps in most games and my core wont go further than 4.7

for reference im running 2133mhz ram 10-12-11-1t at the moment, it will go higher but stability is hard to find past 2400, all in all pretty damn good for a 1333 9-9-9-24-2t kit

Kid PC 1:
i7 4790K 4.6GHz 1.18v w/Gammaxx 400 LED, Asus Z87-Pro, 16GB DDR3 2400 CL13, GB GTX 970 WF3 4GB + Asus VN248 24" 1080P, 500GB SSD+1TB HDD, DVD Burner, EVGA 750G+, NZXT Source 210.
Kid PC 2:
i7 4790K 4.6GHz 1.21v w/212EVO LED, Asus Z87-Pro, 16GB DDR3 1866 CL9, MSI GTX 970 Gaming 4GB + Asus VN248 24" 1080P, 500GB SSD+1TB SSHD, DVD Burner, EVGA 750G+, Lian Li PC-K62.

mad respect for still using dvd readers/burners in your kids builds