Post by MachineLearning on Apr 30, 2022 5:32:24 GMT -5
Hi all,
I had an idea the other day, on how to essentially get 1.5 samples out of an unlockable PhII X4. What if I use core unlocker to activate all six cores, and then bin them down to the best four? Surely, the first four cores can't be the best, if all 6 work.
Here is my first result, using the ASUS M4N98TD Evo, with core unlock disabled. On this board, it's all-or-nothing - I may not select which cores I wish to unlock, unfortunately, although I do get per core NVCC controls.
hwbot.org/submission/4985740_
4.678 FPS, as you can see it's only called the Phenom II X4 960T with four cores enabled. The max I could get out of these first four cores was ~ 4350MHz, core three kept giving out above that.
And here is the final result, for now...
hwbot.org/submission/4986354_
4.847 FPS, a lot better as I'm now able to bring it to ~ 4550MHz. Since core unlock is technically enabled, CPU-Z still identifies it as the X6 1600T, but clearly only four cores are enabled. I found cores three and six to be the worst in my sample (produced week 40/2011). I used the M4A89GTD Pro/USB3 for this second result, as it provides granular core activation control except for the first core, which must always be enabled.
All testing was done using the Arctic LF II 280 and MX-4 thermal paste, so I did not improve the cooling between overclocks.
The way that I went about eliminating the worst cores was Prime95 blend. Either a) moving up big leaps in frequency or b) high-stress FFTs would crash the system far more quickly than P95 blend, which of course makes sense, but is also useless information for core binning. Whichever process dies, is the next core to be disabled, until the best remain...
This was a lot of fun to test. It looks like I really can't bring this any farther on ambient! I looked at the other scores there, and it seems that nobody else has tried the same thing yet for the 960T 4c category. I hope someone on the team reading this can get a wicked good sample and blow everything out of the water.
I had an idea the other day, on how to essentially get 1.5 samples out of an unlockable PhII X4. What if I use core unlocker to activate all six cores, and then bin them down to the best four? Surely, the first four cores can't be the best, if all 6 work.
Here is my first result, using the ASUS M4N98TD Evo, with core unlock disabled. On this board, it's all-or-nothing - I may not select which cores I wish to unlock, unfortunately, although I do get per core NVCC controls.
hwbot.org/submission/4985740_
4.678 FPS, as you can see it's only called the Phenom II X4 960T with four cores enabled. The max I could get out of these first four cores was ~ 4350MHz, core three kept giving out above that.
And here is the final result, for now...
hwbot.org/submission/4986354_
4.847 FPS, a lot better as I'm now able to bring it to ~ 4550MHz. Since core unlock is technically enabled, CPU-Z still identifies it as the X6 1600T, but clearly only four cores are enabled. I found cores three and six to be the worst in my sample (produced week 40/2011). I used the M4A89GTD Pro/USB3 for this second result, as it provides granular core activation control except for the first core, which must always be enabled.
All testing was done using the Arctic LF II 280 and MX-4 thermal paste, so I did not improve the cooling between overclocks.
The way that I went about eliminating the worst cores was Prime95 blend. Either a) moving up big leaps in frequency or b) high-stress FFTs would crash the system far more quickly than P95 blend, which of course makes sense, but is also useless information for core binning. Whichever process dies, is the next core to be disabled, until the best remain...
This was a lot of fun to test. It looks like I really can't bring this any farther on ambient! I looked at the other scores there, and it seems that nobody else has tried the same thing yet for the 960T 4c category. I hope someone on the team reading this can get a wicked good sample and blow everything out of the water.
