Not sure what's going on exactly. My LinX is running slower now. The GFlops reading has decreased and instead of 136 GFlops I now get 128 GFlops and the time to complete has gone up from 404 seconds to 428 seconds. The core voltage will occasionally drop to 1.224 while LinX is running and it isn't happening in between runs but instead during the runs. I have the same exact settings except for on VCCSA Load-line Calibration and VCCSA Current Capability. The only other thing I've changed since then has been disabling C3, C6, and C7 states as I got a blue screen with error code 124 at idle while browsing the web last night.
Could disabling the C-states lower my benchmark scores by nearly 8 GFlops? That seems like an awful big loss and I don't know why. Ambient temperatures are the same as they were when I was seeing 136 GFlops. Anyone know what could be going on here?
According to this thread changing the C-state settings from the default could have an adverse effect on performance: http://www.evga.com/forums/tm.aspx?m=928984
. 8 GFlops difference seems like too much, but is there any way around this? Is it possible to fix the 0x0124 blue screen error code when overclocking without sacrificing a whole 8 GFlops of performance or without disabling C-states? I could try raising the VCore or changing the LLC setting but I don't want to keep pushing more and more VCore. My temperatures reach the mid 80s under full load and I would like to keep them below 90 °C at full load. I'd also like to keep VCore under 1.45 for 24/7 usage and I've seen it go up to 1.448 at my current settings already.
I raised my CPU VCore Offset Voltage to +0.020 and I'm just about back to where I was at before. +0.010 wasn't stable, +0.015 was iffy. CPU VCCSA Offset Voltage is set to Auto. Idle voltage is 1.392~1.400 V and load voltage is up to 1.448~1.456 V as measured by CPU-Z, but I even saw it jump to 1.464 briefly while stress testing. This is a little higher than the 1.45 VCore max that I wanted for 24/7 use but this is also a 50% overclock on a hexa-core processor. I think these voltages should still be safe as my temperatures are still staying in the mid 80s under load at maximum.
Update: Flashed to 0901 BIOS. Now running the following settings:
- CPU Clock Gen Filter: 20UF
- Memory Frequency: DDR3-2133MHz
- BCLK Skew: -2
- PCIE Skew: -20
- CPU Load-line Calibration: Ultra High
- CPU Current Capability: 180%
- CPU VCore Offset Voltage: +0.040
- VTT CPU Voltage: Auto
- 2nd VTTCPU Voltage: Auto
- CPU VCCSA Manual Voltage: Auto
- DRAM Voltage (CHA, CHB, CHC, CHD): 1.650
- CPU PLL Voltage: Auto
- CPU & PCIE Spread Spectrum: Both Disabled
- CPU C3, C6, & C7 Report: All Disabled
CPU-Z is reporting 1.456~1.464 V load and 1.400~1.408 V idle. I won't worry about any differences in GFlops with LinX from here out as I am running a newer/different version of BIOS. I won't be able to notice any real world difference and if I really wanted higher scores in LinX I would just disable hyper-threading. For 24/7 usage I think I may run 4.5 GHz with 1.4 V max VCore under load as my temperatures stay about 10 °C cooler. To get 4.5 GHz stable is a breeze. I simply set CPU Core Offset Voltage to -0.005 and don't really have to change much else. I could just change the multiplier to 45 and leave everything else at the default setting to get 4.5 GHz stable with this board. The difference measured between 4.8 GHz and 4.5 GHz is about 2 GFlops with hyper-threading turned off under LinX but the difference in voltage is 0.050 volts and the difference in temperature is approximately 10 °C.
4.5 GHz Stable under LinX (Not final settings, may try reducing VCore more):
4.5 GHz Stable under Prime 95 (Not final settings, may try reducing VCore more):Edited by Improvidus - 12/15/11 at 9:05am