New 6 GPU miner - problem looking for solutions

[akadamson]

I have a fairly new 6 GPU rig.  It's on a Asus Prime Z270-A MB and uses risers.  The GPU's are all EVGA Kingpin 1080 Ti's.

The problem is really rather strange and I'm looking to see what others might suggest as a form of debugging.

I can launch any of the mining software under either windows or Linux and the miner will perform flawlessly.  Even OC it up to 800+ h/s per card on equihash with no issues.

But it's when I go to stop the software that I have a problem.  One of the cards, and it seems to be the same one every time.  Will then lock up the computer with an IRQ pending request.  I can see this under linux as a stuck process called IRQ/XXX-nvidia (the XXX is a 3 digit number).  Once I unplug the card from the riser, that will clear and the system utilization will return to normal.  Same thing happens under windows 10.

Again, this does NOT effect the function when the mining software is running, just after... but any time you stop the software, this will happen and if you don't want to risk some hardware damage by unplugging that card, you have to do a hard reset.

I've done all the obvious stuff.  swapped risers, swapped slots, etc and the problem always follows the card.  As an aside I have noticed that this card is always the lowest of the batch of 6 from a h/s standpoint.

Once I remove this card from the system, I never see this issue on 5 cards.

Any thoughts on what I should try before I attempt to get warranty replacement from EVGA?

Thanks in advance,

[1714965742]

1714965742

[1714965742]

1714965742

[1714965742]

1714965742

[1714965742]

1714965742

[1714965742]

1714965742

[Rath_]

Did you try plugging your GPU directly into PCIe slot? Does Windows Device Manager report any errors? If it doesn't change anything then I would recommend you sending your GPU to the manufacturer.

[leonix007]

I have a fairly new 6 GPU rig.  It's on a Asus Prime Z270-A MB and uses risers.  The GPU's are all EVGA Kingpin 1080 Ti's.

The problem is really rather strange and I'm looking to see what others might suggest as a form of debugging.

I can launch any of the mining software under either windows or Linux and the miner will perform flawlessly.  Even OC it up to 800+ h/s per card on equihash with no issues.

But it's when I go to stop the software that I have a problem.  One of the cards, and it seems to be the same one every time.  Will then lock up the computer with an IRQ pending request.  I can see this under linux as a stuck process called IRQ/XXX-nvidia (the XXX is a 3 digit number).  Once I unplug the card from the riser, that will clear and the system utilization will return to normal.  Same thing happens under windows 10.

Again, this does NOT effect the function when the mining software is running, just after... but any time you stop the software, this will happen and if you don't want to risk some hardware damage by unplugging that card, you have to do a hard reset.

I've done all the obvious stuff.  swapped risers, swapped slots, etc and the problem always follows the card.  As an aside I have noticed that this card is always the lowest of the batch of 6 from a h/s standpoint.

Once I remove this card from the system, I never see this issue on 5 cards.

Any thoughts on what I should try before I attempt to get warranty replacement from EVGA?

Thanks in advance,

The problem is that the manufacturer wont admit that your card is faulty, I wonder what would you tell them to describe your problems? I'm pretty sure you wont tell them you are doing mining.

back to GPU prob, Ensure that you have a pretty decent PSU to handle those monstrous power hunger cards.

[akadamson]

The problem is that the manufacturer wont admit that your card is faulty, I wonder what would you tell them to describe your problems? I'm pretty sure you wont tell them you are doing mining.

back to GPU prob, Ensure that you have a pretty decent PSU to handle those monstrous power hunger cards.

Yes, that is always the dilemma.  1600W supply with a 750W so power isn't the problem.... I have made some progress.  These cards will do well over 800 h/s on equihash at 100% power... but there seems to be a relationship between core and mem.  E.g. by default they overclock it appears so even a 0+ on core will run them at a 2000mhz core clock, they are garanteed for 2025 core. and can run up to 2100mhz.  So I've backed the core clock back to +0 and the mem clock back to +550 (from the +850 I was running). Seems to have made a difference.  I noticed however when I pushed the mem clock back down, the core clock decreased as well.  Curious.  haven't changed the power and maybe that is why it's limiting internally.

Anyway, it will run DSTM just fine on any of the numbers above.... but this one card, when you stop DSTM, will go into in a race condition with it's IRQ's.   So the battle continues, but I have made some progress