SkylakeX teasers (aka prime95 29.5)

[simon389]

Originally Posted by Mysticial

I didn't realize there were multiple of these machines!

I guess the obvious questions/clarifications are:

• Are all the systems identical? (same CPU, mobo, memory, etc...)
• Are all the systems unstable?
• If they are unstable, are they all unstable in the same way? (AIDA64 AVX512 + Prime95 PRP?)

Slightly different mobo, slightly different RAM. Otherwise all 4 machines are the same.

2/4 have EVGA x299 Micro (131-SX-E295) - latest BIOS
2/4 have EVGA x299 Micro 2 (121-SX-E296) - latest BIOS
2/4 have G.SKILL Sniper X Series 64GB (4 x 16GB) 288-Pin DDR4 SDRAM DDR4 3600 (PC4 28800) Desktop Memory Model F4-3600C19D-32GSXKB
2/4 have G.SKILL Ripjaws V Series 64GB (4 x 16GB) 288-Pin DDR4 SDRAM DDR4 3600 (PC4 28800) Desktop Memory Model F4-3600C19D-32GVRB

4/4 have:
Intel 9800X Skylake-X CPU (lapped)
Noctua NH-D15 coolers with NT-H1 paste (all temps under 80C at load)
All are using XMP settings for 3600Mhz RAM @ cas 19
400W Seasonic platinum PSU (actually one has a 850W to see if that was the problem)
Headless with junk 2nd hand graphics cards
120GB SSD
Win10 64 bit
Cheap microatx Cases
All are running default BIOS settings except for XMP and the one I’m underclocking to reach stability
All are plugged into a KillOWatt to measure consumption.
All are stable under 29.4 (no AVX512) and unstable @ 29.5b9 (AVX512)

[chalsall]

Originally Posted by simon389

Slightly different mobo, slightly different RAM. Otherwise all 4 machines are the same.

Please forgive me for this, but your MB count adds up to eight. May I presume you meant you have four different MBs, from two different manufacturers? That's a good sign.

The only important commonality I see is the Intel 9800X Skylake-X CPU (lapped).

If you have the time and the coin, swapping one of those out and testing deeply might be worth your time.

Intel have learnt the hard way that selling bad kit is not good for public relations, nor their stock price....

Edit: Just to be clear, you "lapped" your kit!?!?!? And you're wondering why it's not working reliably?

[preda]

Originally Posted by simon389

I’m dropping all the speeds. I keep an eye on CPU Z and even with all speeds @ 3.8 (incl AVX and AVX512 @ 3.8) I cant get a stable system. Failed at 7.5 hours. Trying now with 3.7 ::shrug::

IMHO, it would be better to keep the failing system untouched and failing, and run mprime95 again on it, maybe even the same exponent again, to hopefully fail in the same way -- in order to find out why PRP didn't detect the error. That's more important than getting the one system stable.

The core problem is not why the system failed, but why the PRP didn't catch it.

[Mysticial]

Originally Posted by simon389

Slightly different mobo, slightly different RAM. Otherwise all 4 machines are the same.

2/4 have EVGA x299 Micro (131-SX-E295) - latest BIOS
2/4 have EVGA x299 Micro 2 (121-SX-E296) - latest BIOS
2/4 have G.SKILL Sniper X Series 64GB (4 x 16GB) 288-Pin DDR4 SDRAM DDR4 3600 (PC4 28800) Desktop Memory Model F4-3600C19D-32GSXKB
2/4 have G.SKILL Ripjaws V Series 64GB (4 x 16GB) 288-Pin DDR4 SDRAM DDR4 3600 (PC4 28800) Desktop Memory Model F4-3600C19D-32GVRB

4/4 have:
Intel 9800X Skylake-X CPU (lapped)
Noctua NH-D15 coolers with NT-H1 paste (all temps under 80C at load)
All are using XMP settings for 3600Mhz RAM @ cas 19
400W Seasonic platinum PSU (actually one has a 850W to see if that was the problem)
Headless with junk 2nd hand graphics cards
120GB SSD
Win10 64 bit
Cheap microatx Cases
All are running default BIOS settings except for XMP and the one I’m underclocking to reach stability
All are plugged into a KillOWatt to measure consumption.
All are stable under 29.4 (no AVX512) and unstable @ 29.5b9 (AVX512)

The thing that stands out is the XMP. 3600 MT/s is high for this platform. Both of my Skylake X boxes have trouble at this speed.

But since there's 4 machines here failing the same way, it seems unlikely this would be the cause.

It's worth turning off the XMP anyway to see if you still see the instability. But I doubt this is the cause. Memory instability is highly variable across even identical systems. So it seems unlikely that it would put all 4 systems right on the edge of stability.

Originally Posted by preda

IMHO, it would be better to keep the failing system untouched and failing, and run mprime95 again on it, maybe even the same exponent again, to hopefully fail in the same way -- in order to find out why PRP didn't detect the error. That's more important than getting the one system stable.

The core problem is not why the system failed, but why the PRP didn't catch it.

There are two things we're trying to solve here, and they are not conflicting. We already know the settings that are failing - stock settings. And it can be easily returned to those settings at any time.

So while we are blocked trying to solve the software issue, it is still safe to try to (in parallel) debug the hardware issue as well. There's also multiple boxes here exhibiting the same issue. So if we're extra paranoid, just leave one of them untouched.

[chalsall]

Originally Posted by preda

The core problem is not why the system failed, but why the PRP didn't catch it.

That is one of two problems we're currently "working".

Some of us prefer to work with stable kit before trying to figure out why the software failed.

But your point is valid. Possibly simon389 will agree to not fix one of his machines, so George can run tests on it. To figure out why the software didn't correctly determine that the hardware was insane.

[R. Gerbicz]

Originally Posted by preda

run mprime95 again on it, maybe even the same exponent again,

If he is getting/using different shift value for PRP then he will squaremod very different numbers.

Originally Posted by chalsall

But your point is valid. Possibly simon389 will agree to not fix one of his machines, so George can run tests on it. To figure out why the software didn't correctly determine that the hardware was insane.

Agree.

[simon389]

Originally Posted by chalsall

Just to be clear, you "lapped" your kit!?!?!? And you're wondering why it's not working reliably?

They had problems before lapping the IHS. I lapped to try and get temps down (and the temps did indeed lower but the errors remained). I have lapped many times. No problems ever.

[chalsall]

Originally Posted by simon389

I have lapped many times. No problems ever.

With all due respect, you are reporting problems now.

And you will not be able to return them as defective, since you have modified them from spec.

[Mysticial]

Originally Posted by chalsall

With all due respect, you are reporting problems now.

And you will not be able to return them as defective, since you have modified them from spec.

If I had to bet (without much confidence), I would put my money on a mobo/ram incompatibility issue.

The same mobo, with the same memory brand and type (albeit with different decorations). I've seen these sorts of things show up all the time on my own hardware. Even for memory that's on the QVL of the mobo. I've had multiple instances where the QVL has failed me. Quite often, downclocking doesn't work, and the only solution is to change either the mobo or the memory.

The only thing that doesn't really support this is that the errors are AVX512-sensitive. Perhaps the AVX512 makes the workload sufficiently memory-bound to stress the memory in a way that's not possible with just AVX or scalar.

I find it hard to believe that sanding down the IHS would cause problems like this. I'd expect it to be no issues at all, unrelated issues (like temperature), or catastrophic issues like a missing memory channel, or complete failure of the chip. And then you have the fact that it's 4/4 machines here - all with the exact same symptoms.

[chalsall]

Originally Posted by Mysticial

If I had to bet (without much confidence), I would put my money on a mobo/ram incompatibility issue.

Perhaps simon389 would be willing to take some direction from you wrt MB and RAM compatibility, using only the kit they have on hand.

This might only involve some BIOS settings (set at the lowest possible, initially), to experiment....

[simon389]

Lapping was a really smart decision IMHO. Yes, it killed my warranty, but the IHSes were all rounded in the middle and temps were all over the place. It’s amazing, actually, how uneven CPUs and heat sinks are, even of the premium variety.

Originally Posted by chalsall

Perhaps simon389 would be willing to take some direction from you wrt MB and RAM compatibility, using only the kit they have on hand.

This might only involve some BIOS settings (set at the lowest possible, initially), to experiment....

More than happy to