Threadripper 2000 series /thread

Well, i think the issue with that is that now we’re on the multi-chip package train, the previous silicon limits are somewhat relaxed because you aren’t actually ramming all that power through a single die any more.

well, unless you’re intel, and in the business of making the most compact and efficient heating elements known to mankind.

1 Like

Found a good DevOps 2990WX build, @wendell.

https://pcpartpicker.com/b/3rLJ7P

Has lstopo and IOMMU groupings.

Well yeah i still have some question marks with the way he,
did those tests.
Wendell exally had some great experiences with the Aorus Extreme.
And from a vrm point of view the board is one better boards out there.
But of course not every board overclocks the same, what i mean by this,
is that certain boards just need more fiddling and fine tuning on OC settings etc.
That a certain overclocking settings work well on board A, doesnt automaticlly mean that it would also work well on board B visa versa.
A decent airflow in the case is obviouslly very important.
But all boards were tested under the same conditions,
so seeing those differences in his video accros those boards,
is definitely interesting to do some deeper research on.

Hmm, 2990wx+MSI MEG is not liking 8x16G 3200C14 B-die.

I’ve had to drop to 3000C14 just to boot. SAT still found a single error in a 10m run through… Was working without issue at 4x16G (1/2 the same kit), but full-house is not happy.

We now know that the horrible game performance was due to the Nvidia driver. Newer Nvidia drivers has fixed this problem. AMD GPU’s was not affected.

Yeah, fully populating the DIMMs is only rated to 2933mhz. (AKA JEDEC spec for Threadripper 2)

Extremely common that populating all 8 slots will be too demanding for an XMP profile to run at full speed.

Yeah, understood on the spec. I bumped various voltages (SoC to 1.05, ram to 1.37) and and dropped to CR 2T even dropped timing to 15 then 16 hoping to bring it to life at 3200, but so far, will not even POST anywhere above 3000.

This kit pushes the Intel chip(s) it was designed for as well, but can usually tweak it eventually. It’s a lot to ask, but 4x worked so well that I had hope. I’ll have to explore some of 2nd/3rd level timing to see if I can find the show stopper(s).

The show stopper is the IMC has literally hit it’s limit. I had the same problem when I populated all DIMM slots on X79.

It actually behaves as if some artificial configuration (not electrical) limit is being imposed…

For an IMC to be able to do 4x3600C16-CR1 and not even boot 8x3133C16-CR2 is a little unusual to say the least.

It doesn’t even try - the hang on C5 boot-code happens immediately - requires cmos clear.

It’s a 2933mhz hard limit imposed by AMD likely with all DIMMs populated.

This chip is supposed to be unlocked. I’ve not yet found any reference to their having locked this setting. There are plenty of references to TR1’s 128G memory issues that required bios and ucode updates, but no suggestion (or reason I can think of they would bother) locking this when nothing else is locked.

If there is such a limit, I don’t believe it is hard-coded or really intentional, but may require ucode/bios to address.

I remember trying to POST a P55 system with a custom multiplier above DDR3-1600, and it just wouldn’t post at all. Sometimes a hard limit is reached, and if you’re certain the system can be pushed further, you need to experiment with AGESA versions then.

I suspect so. I have a bios with revB of this AGESA, but it bombs on PCH init (wondering if that’s my PCIE 2.0 10GbE card doing that?) Haven’t had time to experiment with that further.

The new AGESAs regressed on many things. I think @wendell knows more about this.

1 Like