Originally Posted by ajc9988
Sorry about long delay on reply, the board IO got killed, taking IO chip in the 1950x CPU with it. Fortunately as it is it did not kill the expensive ram.
Tested against new MSI MEG x399 and with 2950x on that Asrock. Likely board voltage probe error as you suggest. The board is dead, chipset voltage shows invalid values, indicating a short or lack of initialization.
My VRM were cooled from the metal pads too with a thickness of Thermal Grizzly paste plus there was a cutout in the water block, so it touched the chips themselves. Chipset was cooled too.
My beef is that it allowed bus voltage to increase with no load to unsafe levels with flat LLC set to correct values. The offset was used because manual was too narrow for increased clocks.
Aren't those things supposed to be hard limited?
Worse, I even took the measurements from test points and they were fine, yet it blipped (in BIOS, not on load) to insane values taking the IO with it.
ID issue is irrelevant as both the CPU and the chipset were killed. That machine is dead, it's been replaced with 2950x on MSI MEG x399, which is perfectly fine, but does not like higher memory clocks too much, nor allows BCLK being increased, but instead allows better memory timings, making memory bandwidth close enough. (Goes CL16 2T with great secondaries on 3200. Asrock went higher clock but couldn't get as good secondaries.)
Silicon Lottery and Caseking (almost?) never roll Threadrippers, likely due to limited supply and demand. Haven't seen one on offer.
You're talking about BCLK "issue" (rather tuning pecularity) which has zero relevance. This is called being annoying knowitall and a putdown. Not a nice thing. If you know of a relevant "issue", please link to a description of it, rather than putting on airs. That was a final memory clock tuning phase and BCLK itself was frozen within single MHz to keep desired target memory clocks.
I have inspected the CPU visually, there's no apparent damage. Likewise on the chipset. It's inane to suggest that I wouldn't do so. Generally visually apparent damage would happen with overcurrent not overvoltage.
Power viruses were used for thermal stability test - 5 minutes thermal loading initial test, 20 minutes long run once to confirm. Memory performance was evaluated with AIDA64.
Power tests were NOT used in the final memory tuning phase, where it failed, either, only AIDA64 mixed with my compute load. Board actually failed in BIOS during setting. The bus voltage was being changed to stabilize memory overclock as suggested by others, as old Zen pins FCLK to BCLK and memory clock. (I do not remember where anymore.)
I'm not insane to let power viruses run for hours. If the machine passes viruses, then passes AIDA and memtest, it will pass any other load (with exclusion of GPUs) - but not the other way around.
The real stability is tested with actual load later.