Another oddity with my 7900xtx/system.

[fizzy]

At a loss now. Taken out the PCIE4 NVME, and used it in my PS5. Copied pretty much everything from PS5 internal mem to the drive and it was fine, didn't report any issues and flew along.

Checked in HWInfo to confirm gfx card in PCIE4 mode (showing link speed as 16.0 GT/s, i.e. PCIE4) and ran 3dmark PCI Feature test - which I would think would max out the PCIE4 bus to the card and reveal any issues. Test ran fine, reported a bandwidth of 14.13 GB/s.

Copied same 45GB file between remaining two NVMEs both ways multiple times and no issues.

I guess next test would be to get another PCIE4 NVME drive and see if problems reoccur with that, and/or get a high speed USB-C NVME enclosure and test the existing PCIE4 drive with that, and retry the copies etc.

 

[fizzy]

Does all seem a bit odd and does point to motherboard issue. I see you can get pcie to nvme adaptors on rainforest pretty cheap. Wonder if its worth getting one of those ? I would put money on it not being the GPU though.

Edit. ah it seems overclockers do them to https://www.overclockers.co.uk/?query=nvme%20adaptor

From what I can tell all those adapters are PCIE3 cards. Whilst it would allow testing of the nvme drive itself it wouldn't test out the theory of issues relating to PCIE4. I could also use a usb-c nvme enclosure in same way.

Whereas I don't think it's the GPU persay, the situation has not been helped by the GPU/GPU driver issues clouding the problem. Can't see why card would be switching off/going into panic mode or whatever it's doing and windows disabling the driver/device in device manager too. Didn't happen with the Radeon VII - which would suggest either driver or PCIE4 issue. I did still have issues when forcing the 7900xtx to PCIE3 in bios though. Also tried forcing the chipset link to PCIE3 too which made no difference. Not really sure how effective it is restricting a PCIE4 card to PCIE3 in that manner over using a physical PCIE3 card.

For now everything appears stable again with the PCIE4 NVME removed, but will keep an eye on things. Will think about options and what I'll need to get/try to finally track down the problem. Or just get a large SATA SSD instead.

 

[fizzy]

Well - was playing an incredibly intensive game (not!) - Frost Punk - for about 45 mins and.... picture went off, gfx fans ramped up and had to reboot. On reboot.... yup, card showing as disabled in windows device manager. So same issue - just for some reason it was easy to reproduce with the file copying between NVME drives. <sigh>

So - just noticed new driver out (althoug doesn't mention anything about driver timeouts etc) which I've installed and will see if that makes any difference.

 

[fizzy]

No difference with later driver. Even tried going back a version or two - 21.8 etc - same issue. Had a number of occurances whilst playing Frost Punk. Switched to my Radeon VII, installed drivers and played for hours without issues, even completed the game.

Does seem something related to the 7900XTX, but still not 100% sure whether it's card, drivers or some other hardware issue with board/cpu. Memory is still testing out fine.

 

[fizzy]

Ok, lots of head scratching going on here.

To recap - Issue is that main pc has starting having odd issues when copying between nvme drives, and also it seems to be having issues in games too. In all cases, screen goes black, GPU fans go full whack and I need to reboot the pc. Seems to be running in background as I can still hear audio and able to triger a gracefull shutdown using power button. On reload windows has the GFX card down as disabled in device manager. No number of reboots or power offs sort it - I need to re-enable in device manager and reboot to get things back. Using windows 11. Also get it running various graphic stress tests in 3dMark as well, sometimes as soon as it tries to start, other times after a few cycles of the demo.

- Tried various versions of the drivers, used AMD clean utility etc numerous times, still the same.

- Tried resetting bios to defaults, disabled SmartAccess Memory - still the same. There was a slightly later bios which I also installed, but still the same.

- Ran Memtest86 for full scan and no issues found.

- Tried setting PCIE speeds to 3 for GFX slots as well as chipset interconnect - issue persisted.

- Tried my old Radeon VII card instead and everything fine then - played games (frost punk) for hours, stress tests ok etc. Put doubts on a general hardware issue?

- Tried the 7900XTX in another pc - older spec (Intel 4690k, windows 10, Corsair HX750W psu, PCIE3 slots) - played games (i.e Frost Punk) for a good while no issues. 3dmark stress tests also worked fine with a complete test of the DirectX 12 ultimate test etc - GPU was at 100% usage, temps looked fine in HWInfo.

- Tried moving the OS disk from the old PC to Main PC just to check if it was a Windows 11 thing - nope, does it on main pc with Windows 10, and does the same thing in device manager etc. Not seen that before, would think a reboot/power cycle would reset everything.

- Pc has been fine for general usage - web browsing, video watching etc.

So unless there is a weird driver thing going on here when in an AMD system, it's looking like something not right with board or cpu in my main rig. I'm using a 1000W Seasonic Platinum PSU, could try the 750W in main pc. It coped ok with the 7900XTX in the old pc, but then the AMD 5950x is a somewhat more power hungry cpu.

I used to have a 3700x in it originally but that went onto a B450 board for somebody else and I don't have access to it now - would have been useful to part swap. All I can think of now is to try a different board or cpu to try and rule out either, although have been using this setup for a good few months and only recently has it become this unstable.

Sorry for the rambling, been trying a lot of things and seeing if I can figure out things logically and wanted to get thoughts down while I remembered what I had done.

 

[fizzy]

What is the exact model number of these PSUs?


When you used the 7900 XTX in your other PC and it was fine, were you using the same monitor, display cable and refresh settings?

PSU's are Seasonic Platinum 1000W (that's what's on the label) and a Corsair HX750W which is a good few years old.

Yes, I was using the same monitor, display cable and refresh settings in the other pc, as well as same keyboard mouse etc - basically sat it on top of the main pc and moved cables over.

Just had a bit of a result - used the HX750W with main PC and just completed a full 20 loop stress test of 3D Mark Time Spy, which is wasn't doing before. Was a bit of coil whine from it though so don't want to push it too much.

I did go over all the power cables and plugs to make sure all seated before, so I think it's looking like the Seasonic is on it's way out. Guess I'll only know if I get a new PSU and try that. Let the hunt begin.

 

[fizzy]

What date is it? There were known issues with some PSUs from pre-2019. Though, I assume the card worked in this system before?

Been using it with the card since I got the card earlyish last year, so yeah it's been ok since then. I think the PSU is getting on for 5 or 6 years old, maybe longer so probably due a replacement anyway. Can't recall seeing any build date or anything on it but will double check.

 

[fizzy]

Going to need exact model of that 1000w power supply

That is the exact model? Just found the box and it just says "Seasonic Platinum Series 1000W.". Probably pretty irrelevant now though since testing ok with the HX750 so will be looking to replace it anyway.

edit: Just seen on the side label is says SS-1000XP Active PFC F3.

 

[fizzy]

Be worth checking its warranty as they are usually very long.

Indeed - think it's 7 years on this one? I'll have to see if I can find the receipt. Model was apparantly released in 2011 or somewhere around there looking at reviews, so there is a chance I may have had it long enough to be out of warranty. :O

Also remembered I had a thermaltake psu tester. Small unit with an lcd screen that checks voltages etc. Quick check on that and voltages look ok generally, although under the very light load of the tester the 12v rail is at exactly 12.0v. Others I tested are usually a bit over that, say 12.2v+ I'm guessing that the 12v rail voltage has dropped a little over time, enough to effect the GPU under load. Infact reviews show it as 12.4v or so on light load and dropping as load increases. That's my current theory anyway.

Now to decide which PSU to replace it with. Fair few more 1000w+ PSUs now than there was when I bought this one.

 

[fizzy]

Ordered a seasonic vertex px-1200, bit of future proofing and confirmed it has 3 seperate pcie power cables. Also supposedly newer standards cater for transient spikes better (which I was aware of).

Should arrive early next week so will see how it goes.

 

[fizzy]

New PSU received today and fitted. Completed a full 20 loop stress test of both Time Spy and Speed Way benchmarks in 3d mark without any issues. Also copied all contents from one NVME to another and that was fine too.

Fingers crossed hope this means the issue is now resolved. Will keep an eye on things for a while though.

Quite surprised that the new 1200w PSU is actually smaller than the 1000w one I was using before.

 

[fizzy]

Would never have guessed this had anything to do with a PSU

Indeed - also the fact that it was making the card play up whilst doing that too was confusing. Wouldn't have thought that was anything to do with transient spikes or card power draw etc jut working on desktop stuff. Perhaps there was a general drop in the 12v rail under high cpu/io load too which was causing instability. Who know.

Will see how it goes with the new one (although looking good so far). I haven't got the equipment to do a full PSU extensive test of the old one to see how it's behaving.

Hard to tell though whether it was in fact the psu not liking the gfx card, or the other way around. Could be the 7900XTX is a bit "fussier" than my radeon VII etc. I guess I could try the old rig with the 7900XTX and the old 1000w psu and see if it shows issues then - but not sure I want to be part swapping yet again when things are stable at the moment.