LLC Regular P95 Blend stable:
46x100 +0.030V (1.288-1.304V)
48x100 +0.090V (1.344-1.360V)
LLC High P95 Blend stable:
46x100 -0.040 (1.288-1.304V)
48x100 +0.010 (1.344-1.368V)
LLC Regular vs. High
46x100 LLC Regular +0.030V / 0.976-1.136V idle / 1.288-1.304V load / min 8.924W in idle (voltage from CPU-Z, W from HWiNFO32)
46x100 LLC High -0.040V / 0.920-1.064V idle / 1.288-1.304V load / min 9.013W in idle
Manual vs. offset voltage
46x100 LLC Ultra High 1.290V / 1.280-1.296V idle / 1.264-1.280V load / min 9.343W in idle - 65-71°C after 10min P95 Blend
46x100 LLC High -0.040V / 0.920-1.064V idle / 1.288-1.304V load / min 9.013W in idle - 68-75°C after 10min P95 Blend
Conclusion, Load = better on manual (less voltage needed, lower temp.), Idle = better on offset, higher the LLC, the better
I'll stick to offset because it runs 90-95% of the day in idle/light load.
These are still on default:
VRM Frequency
Phase Control
Duty Control
CPU Current Capability, didn't need to raise it from 100%, completely stable in idle
By my experience, Prime95 Blend (I use 4 threads) should let you know where you are with idle stability, it BSOD's 0x124 instantly even on +0.01V more what was IntelBurnTest max 20x stable.to test the "light" load stability,test linx 20 pass on 1 thread.
in my tests i could do 25 pass linx avx on 4/8 threads but was failing linx 1 thread on the second pass so i had to up offset to pass linx 25 runs on 1 thread.
You can do it with prime too.
Should be almost fully stable with 30min Blend. For example, I got BSOD immediately after boot with -0.050V, was 8 hours stable with -0.040V.
Currently on 46x100 LLC High -0.040V, will see how stable idle is. On -0.050V it wasn't stable overnight, so 24 hours in idle should be enough to call it stable.
All in all I'm pretty happy with last 2 days progress, at least those WTF's with random BSOD's are gone.
My main concern remain temperatures with 48x100, cannot even test Prime for long time, will try to remount in next days.
Reply With Quote
Bookmarks