Aorus Master x570 'Auto' voltage is not enough

[smonkie]

I'm running a 3950X in an Aorus Master x570. When voltage is set to 'auto', system is not stable -OCCT gives errors within the first minute (small ffts, Avx2). Other tests like LinpackXtreme or Prime95 fail too. It seems like an offset voltage of +0.05V manages to pass the tests just right.

This is the first time I recall an auto voltage setting fails at providing enough voltage for a CPU. I'm running F11 BIOS, but have also tried F12e beta with the same luck. Could Gigabyte please provide some feedback about this weird behaviour?

[Fred155]

Have you performed a reset to defaults within the bios, and are any settings changed from default?

What memory are you running, and is it at SPD or XMP or manually dialed in timings and speeds?
Have you run a memory test program like - https://www.memtest86.com/

You've definitely got an issue that needs some diagnosis, but from what you've described, I'm not yet sure if it's a cpu, motherboard, ram, or uefi / agesa issue.

A direct support ticket with Gigabyte may also help, but either way, it's likely to go back and forth until it can be determined which component is at fault.

Any chance you might have other components you might have access to, to narrow down the culprit?

I'd start with memtest.

[shadowsports]

Fred155's recommendation of a CLR_CMOS is good.

He's also correct about one or more of your components contributing to the problem. 

What BIOS did your board ship with?
What BIOS rev did this behavior appear in?

The 3950x is supported from BIOS F4.  There are 3 non-beta, non-review releases between F4 and F10 which may not exhibit the behavior at all.

What issue or need did F11 address that wasn't being met in the BIOS rev you started with?

[smonkie]

Have you performed a reset to defaults within the bios, and are any settings changed from default?

What memory are you running, and is it at SPD or XMP or manually dialed in timings and speeds?
Have you run a memory test program like - https://www.memtest86.com/

You've definitely got an issue that needs some diagnosis, but from what you've described, I'm not yet sure if it's a cpu, motherboard, ram, or uefi / agesa issue.

A direct support ticket with Gigabyte may also help, but either way, it's likely to go back and forth until it can be determined which component is at fault.

Any chance you might have other components you might have access to, to narrow down the culprit?

I'd start with memtest.

Fred155's recommendation of a CLR_CMOS is good.

He's also correct about one or more of your components contributing to the problem.

What BIOS did your board ship with?
What BIOS rev did this behavior appear in?

The 3950x is supported from BIOS F4.  There are 3 non-beta, non-review releases between F4 and F10 which may not exhibit the behavior at all.

What issue or need did F11 address that wasn't being met in the BIOS rev you started with?

Thanks for the answers to both of you. I run everything at stock, including memory. I don't think it's a RAM problem because I'm able to pass memtest86 4 passes (about 6 hours) without erros. It's not a BIOS related issue neither because I've tried last four BIOS versions including the lastest one -beta- and each one of them shows the same behaviour.

I've found another user with the exact same problem but he's using an Aorus Pro with his 3950X. Our problem is pretty much identical: unable to stress test without errors at stock defaults, managing to solve it either by setting LLC at low, enabling PBO or setting a +0.05V offset voltage. It leads us to believe it's motherboard default settings related, probably not capable of managing voltage droop correctly.   

[Fred155]

I've not heard of this on x570, but I have heard of a similar case on trx40:
https://forum.level1techs.com/t/solved-amd-threadripper-3970x-under-heavy-avx2-load-defective-by-design-nope-but-there-was-a-bug/153883
https://forum.level1techs.com/t/solved-3970x-prime95-stability/153206

Could the above links possibly sound like they might be related - even if the platform is different, there could be something in common? (I know it's a stretch, but with multiple bios revisions tested, and stock settings, I'm running out of ideas...)

I think you might need to open a support ticket with Gigabyte. With everything at stock, basic avx workloads should not error - I only expect increased wattage and reduced cpu frequency due to heat / power draw. not errors.

Beyond that, there is a small chance that the cpu has a fault. But that seems less likely to me than a motherboard or bios fault.

What's your PSU model?