![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
I have tried various guides and presets and always end up with issues. The most common problem is getting a rounding error in Prime on blend. Today I got one with the fourth core after 2 hours and 48 minutes. The other three cores went for 4 hours until I came back to the computer and stopped it.
Yesterday, I was trying to get a stable dynamic overclock since the Tom's Hardware article said 3.2 GHz with turbo turned on offers the best mixture of power efficiency and performance. I had to raise the voltage up quite a lot higher than what Tom's used and still ended up with rounding errors when running one thread of Prime blend.
So today I tried going for a static overclock so I could get the voltage down and still have a decent clock rate. My current settings (the ones that gave me the error after 2 hours and 48 minutes) are:
Vcore: 1.2125
Vtt: 1.150
DRAM: 1.40 (shows as 1.42 in Speedfan)
QPI ratio: x32 (QPI link speed 5.12 GHz, uncore ratio 16x, uncore frequency 2560 MHz)
CPU multi: 20
BCLK: 160
PCI Express: 100 MHz
CPU Clock Drive: 800 mV
PCI Express Clock Drive: 900 mV
CPU Clock Skew: 0ps
All advanced core features on except Turbo Boost. Load line calibration is off and the Gigabyte board does have quite a bit of droop. I am not going to use load line, though, because from what I've read it's better to have a higher voltage than it is to use the anti-droop, in terms of safety for the processor.
The RAM I have is OCZ that is rated at 1600 MHz 9-9-9-24 with 1.35 volts. 2 sticks, 4 GB total. My BIOS does either 1.3 or 1.4. There is no 1.35 setting. Setting it to 1.3 gives me 1.32 or 1.33 and setting it to 1.4 gives me 1.42. I would like to have the voltage at spec, but I'm limited by the BIOS. I have tried using a 2T command rate and a 1T command rate. I have tried using 1333 MHz and 1600 MHz. MemTest never finds problems regardless of how I set things. I have all RAM timings set to AUTO except tRAS which I raised from 20 to 24.
Temps go up to 59 C with the current settings with an ambient temp of 27 or so. I had lower temps in terms of voltage applied yesterday when I had the intake fan for the H50 set to medium speed and had a big fan blowing onto the motherboard. But, the noise level was too high and I still ended up with the rounding errors. The highest temp yesterday was 63 C and the voltage was as high as 1.2875.
Currently, the BIOS says the system temp is 34 and the CPU is 32. This is just idling in the BIOS. The ambient temp is currently 26. My machine currently has a 140 mm intake (in the center of the front), two 120s (one at the top where the radiator is and one at the bottom where the hard drives are), another 120 behind the H50 radiator, a 120 exhaust, and a Big Boy (200 mm Antec) on top of the CoolerMaster HAF case (it's too large to fit inside).
I tried the suggestion of trying to find the BCLK limit. I set it to 200 and dropped the CPU multi and RAM multi. However, since I'm only going for 160 BCLK I doubt there is a big issue in this regard. However, I'm not sure how much voltage I need for VTT. Several guides said that for 3.2 GHz (160 BCLK) a Vtt of 1.10 is enough. I have it at 1.15 currently just to give it some extra but still got the rounding error in Prime.
The guides that suggest isolating this and that don't seem to notice that everything is interconnected. Dropping the RAM performance drops the amount of bandwidth the RAM is able to stress the other things with. Dropping the CPU performance drops the amount of stress in terms of Vtt/BCLK. Aside from fairly obvious stability issues (such as RAM that fails MemTest and a machine that reboots when all cores are heavily loaded), it seems like the only way to work toward real stability is to raise each thing individually one at a time, back and forth. It's very time-consuming and annoying though.
The power supply is a 600W Xigmatek unit but it only has 30 amps on the 12V rail. However, I doubt the power supply is the source of the problem since the GPU isn't even loaded for some of the testing where I get rounding errors. Yesterday, though, when I was trying to get a stable dynamic overclock of 3.2 GHz with turbo on I ran Heaven extreme in a window with one Prime blend thread and kept getting rounding errors within about ten minutes.
I am about ready to just sell the motherboard and CPU and move on. I've been trying on and off to overclock this thing for years and it's never stable, even with tiny overclocks like 3.2.
Yesterday, I was trying to get a stable dynamic overclock since the Tom's Hardware article said 3.2 GHz with turbo turned on offers the best mixture of power efficiency and performance. I had to raise the voltage up quite a lot higher than what Tom's used and still ended up with rounding errors when running one thread of Prime blend.
So today I tried going for a static overclock so I could get the voltage down and still have a decent clock rate. My current settings (the ones that gave me the error after 2 hours and 48 minutes) are:
Vcore: 1.2125
Vtt: 1.150
DRAM: 1.40 (shows as 1.42 in Speedfan)
QPI ratio: x32 (QPI link speed 5.12 GHz, uncore ratio 16x, uncore frequency 2560 MHz)
CPU multi: 20
BCLK: 160
PCI Express: 100 MHz
CPU Clock Drive: 800 mV
PCI Express Clock Drive: 900 mV
CPU Clock Skew: 0ps
All advanced core features on except Turbo Boost. Load line calibration is off and the Gigabyte board does have quite a bit of droop. I am not going to use load line, though, because from what I've read it's better to have a higher voltage than it is to use the anti-droop, in terms of safety for the processor.
The RAM I have is OCZ that is rated at 1600 MHz 9-9-9-24 with 1.35 volts. 2 sticks, 4 GB total. My BIOS does either 1.3 or 1.4. There is no 1.35 setting. Setting it to 1.3 gives me 1.32 or 1.33 and setting it to 1.4 gives me 1.42. I would like to have the voltage at spec, but I'm limited by the BIOS. I have tried using a 2T command rate and a 1T command rate. I have tried using 1333 MHz and 1600 MHz. MemTest never finds problems regardless of how I set things. I have all RAM timings set to AUTO except tRAS which I raised from 20 to 24.
Temps go up to 59 C with the current settings with an ambient temp of 27 or so. I had lower temps in terms of voltage applied yesterday when I had the intake fan for the H50 set to medium speed and had a big fan blowing onto the motherboard. But, the noise level was too high and I still ended up with the rounding errors. The highest temp yesterday was 63 C and the voltage was as high as 1.2875.
Currently, the BIOS says the system temp is 34 and the CPU is 32. This is just idling in the BIOS. The ambient temp is currently 26. My machine currently has a 140 mm intake (in the center of the front), two 120s (one at the top where the radiator is and one at the bottom where the hard drives are), another 120 behind the H50 radiator, a 120 exhaust, and a Big Boy (200 mm Antec) on top of the CoolerMaster HAF case (it's too large to fit inside).
I tried the suggestion of trying to find the BCLK limit. I set it to 200 and dropped the CPU multi and RAM multi. However, since I'm only going for 160 BCLK I doubt there is a big issue in this regard. However, I'm not sure how much voltage I need for VTT. Several guides said that for 3.2 GHz (160 BCLK) a Vtt of 1.10 is enough. I have it at 1.15 currently just to give it some extra but still got the rounding error in Prime.
The guides that suggest isolating this and that don't seem to notice that everything is interconnected. Dropping the RAM performance drops the amount of bandwidth the RAM is able to stress the other things with. Dropping the CPU performance drops the amount of stress in terms of Vtt/BCLK. Aside from fairly obvious stability issues (such as RAM that fails MemTest and a machine that reboots when all cores are heavily loaded), it seems like the only way to work toward real stability is to raise each thing individually one at a time, back and forth. It's very time-consuming and annoying though.
The power supply is a 600W Xigmatek unit but it only has 30 amps on the 12V rail. However, I doubt the power supply is the source of the problem since the GPU isn't even loaded for some of the testing where I get rounding errors. Yesterday, though, when I was trying to get a stable dynamic overclock of 3.2 GHz with turbo on I ran Heaven extreme in a window with one Prime blend thread and kept getting rounding errors within about ten minutes.
I am about ready to just sell the motherboard and CPU and move on. I've been trying on and off to overclock this thing for years and it's never stable, even with tiny overclocks like 3.2.