Another oddity with my 7900xtx/system.

[fizzy]

Hi,

Encountered another repeatable oddity on my system.

Copying a 45GB file between two NVME drives and windows progress bar chugging along at about 2400mb/s or so and then suddenly blank screen and gfx card fans ramp up to 100%. No response from caps lock etc yet disk activity light on case is showing some activity. No choice put to hold power button down and force a shut off.

Power back up and..... no picture? Sounds of life and disk activity, sounds through audio etc just no picture? Quick press of power button and windows does a clean shutdown, so that's working ok. Proper hard power cycle using PSU power switch, back on and...... picture back? It's almost like the GFX card had locked/crashed? Odd.

On loading windows, desktop in low res, and adrenaline complaining about driver issues. Reinstalled latest and all ok again.

So tried copying file again and..... exactly the same issue, and same resolution?

And the odd part - on 3rd try whilst in low res, I uninstalled AMD software and display drivers, tried copying the large file again and it completed with no issues whatsoever?

Done a scan of all drives for errors etc and none found. Done a memtest86 run on memory too, which was fine. If there were memory issues would expect problems whether driver was installed or not.

Confused as to how having the driver in place causes the issue on the large file transfer? Going to do some more testing to see if I can pinpoint anything but wanted to check whether this is anything anybody has seen before?

Spec - ASRock x570 Tiachi, 5950x, 32gb DDR4 3600 8pack black (4 x 8GB), Sapphire 7900XTX nitro+, Seasonic 1000W psu, Windows 11.

 

[fizzy]

Make sure bios / chipset drivers are up to date. Do you get any issues in games.

Already all up to date/latest versions. I did have some issues with Baldurs Gate on initial release but these went after numerous game updates. Not had any issues with anything else that I can recall.

 

[fizzy]

I'd suspect a faulty card here.

It's hard to think that a faulty card could affect nvme file transfers though? If anything I would suspect more driver/amd app related. It's odd that it's a hard crash though, not a blue screen or general error message.

 

[fizzy]

Those NVME’s can get very hot indeed. Might be worth having hw monitor installed to check temps is your case well ventilated.

It's in a corsair Air 540 case with plenty of fans, so airflow not a problem. NVME's are cooled by a full plate across the motherboard. I don't think it's heat though as surely it would have the same issue when I tried when gfx driver was not installed, and it copied fine? Copied files between NVME and SATA without issues but then speed is limited by sata bus there.

Odd thing too is that files are being copied between NVMEs that are NOT the OS drive so not sure why drivers would be effected/corrupting?

 

[fizzy]

How many times has the driver freaked out? If only once, maybe an NVME heat transfer issue. If more than once, I'd suspect the card.

As I said - repeatable. Resolved initial issue, tried again and exact same thing. Will do some more testing tonight. Still baffled by why nvme transfers may be a card issue?

Could just be that the driver corruption "issue" is unrelated and due to just having to forcefully power down the pc, but I can test that tonight too. Odd that file transfer was ok with driver uninstalled though. Need to also test that more in case that was just a fluke too.

 

[fizzy]

Should be ok is not is ok. Monitor your temps of the NVME drives. I would also suggest disabling any networking just in case Microsoft update is not causing a problem.

I'll monitor things tonight.

As stated in another reply though - the file is being copied between 2 nvme drives that are NOT the OS drive so not sure why driver is being affected.

I had disabled the windows update drivers updates via policy/reg edit previously but will double check it's not been reset by windows and disconnected network just to be sure.

 

[fizzy]

I was also going to suggest that. I assume you're not using a riser?

Nope, no riser - direct to MB.

 

[fizzy]

Ok, decided to have a bash this afternoon.

Setup HWInfo to monitor drive temps. Tried copy... no big temp jump or anything reported whilst copying, nvme around 40'sC constantly then screen off, fans full etc again.

Thought for a minute I may have cracked it as chipset fan profile was set to silent (passive fan) until high temp so thought it may be that getting hot so monitored temp and put normal fan profile back one. Copied file ok on a couple of occasions but other attempts issue returned. Chipset temps were fine as well.

Switched PCI mode to PCIE3 for GFX. Issue persists. Only difference is when screen goes off I'm actually able to trigger a safe shutdown by quick press of power button rather than the lockup like symptoms of before. When starting up again, windows reports that card is disabled/down (although display works, full res but guess using MS display driver). Proper power down, switch off then a cold boot and everything fine again.

Tried a couple of drivers versions back too but same thing.

Something is definitely unstable somewhere, and for some reason when the card plays up it almost seems like it crashes internally and needs a proper power cycle to recover oddly. Possibly some wierd card bios issue? I've no idea why drivers/card is affected like this, not seen it on other cards. Just remembered it has the bios switch so may try the other position which I think is safe/standard settings?

I'll also try putting in my old radeon VII again tonight (which is a PCIE 3 card) and do same tests to rule out any board, nvme drive or cpu based issues. Certainly an odd one.

 

[fizzy]

What card did you have in before ? Did you have the issue before the 7900XTX ? Just seems like it could be motherboard bios related, maybe drop in voltage to pcie whilst copying large file, is it possible to go back a motherboard bios ?

Previous card was a Radeon VII, which I still have and will probably try that later. Don't recall any issues with it - other than not being quite as fast as expected. It is a PCIE 3 card though - the 7900XTX is my first PCIE 4 card.

The ASRock Taichi X570 I have has always been a bit..... finicky. May have to look into settings since a lot are on auto, and auto o/c etc. It has odd quirks like when you manually enter the PCI bus speed as 100 it switches back to auto.

I've got 3 NVME drives installed - a WD Black SN850 (Gen 4) in top slot, Samsung 970 Evo Plus in center slot and a WD Black SN750 (Gen 3) in lower slot. Issues I'm having is copying the big file between the two WDs. Samsung is my OS drive and when copying to that it just doesn't get a high sustained throughput - suspect as it's the OS drive and used a lot by other things.

I've been told it's generally not a good idea to revert to earlier bios versions - and there are warnings as such on their website. When I flash a new version I've always reloaded defaults and then entered xmp mode etc afterwards so all fresh settings.

More testing this evening I think and see if I can find anything obvious.

 

[fizzy]

Further digging tonight and found that there is actually a newer bios available - 5.50 - so installed that, loaded defaults etc. Tried file copy again several times and all fine. Thought I might have cracked it. However just tried again and it's done it again. Was watching a video at the time too, and could still hear audio etc so not locked up. This is in PCIE 4 mode. Was able to shutdown windows cleanly using power button press.

On loading windows, once again it was reporting gfx card as down as disabled (yet it was still working) in device manager. Tried reenabling it and got an error from AMD adrenalin about not being able to load driver. Rebooted windows and all ok again once more.

All I have in the windows event logs are a couple of these driver alerts warnings, which don't point to much.

Log Name: System
Source: Display
Date: 16/01/2024 21:48:57
Event ID: 4101
Task Category: None
Level: Warning
Keywords: Classic
User: N/A
Computer: MainPC
Description:
Display driver amduw23g stopped responding and has successfully recovered.

It's all rather frustratingly annoying. Especially as it didn't recover at all, just sat on black screen with gfx fans full blast.

 

[fizzy]

Have you tried reseating all the GFX card power connectors to the card and PSU?

Yes, and double/triple checked that I have seperate power cables for all 3 PCIE power connectors.

I've been playing games recently with no issues, which is when I would expect issues with high power draw etc... but when copying between NVME drives on desktop?? The draw must be miniscule is comparison, unless the file transfer progress bar takes far more power than it should.....

I've just downloaded and run the AMD cleanup tool, which cleared out all AMD related files. Then reinstalled latest chipset drivers and latest gfx driver again. Will see if that makes any difference.

Copied big file across again (about 45GB - at about 2,500 MBs).... and worked ok for the couple of times I tried it. Will see if this stability persists as it lulled me into a false sense of a fix earlier this evening too...

Still can't understand why copying a large file between NVME drives is causing the gfx card/driver to wig out. It doesn't make sense to me. Would expect possibly a file error message if it was a drive fault etc.

 

[fizzy]

Nope, still doing it. I've been copying from the WD Gen 3 to the WD Gen 4, so tried the other way this time and screen went off pretty much straight away. Same issues with gfx card disabled in device manager on reboot. Had to re-enable in device manager, reboot (after adrenalin said couldn't load driver) and all was ok.

It's so repeatable and same thing happening every time, I can't see it being a random fault now.

edit: This is crazy. Issue occurs - and on reboot device manager has card as disabled. Even if I do a proper power switch off it stays that way. I have to enable it again in DM. Windows says enabled but still problems, message pops up from Adrenaline that it can't load driver but then on reboot again everything is fine. I can't understand why it would stay disabled after a cold start etc.

None of this is making any logical sense. Will see if I can figure anything new out tomorrow.

 

[fizzy]

Did you try your old card.

Not yet, I'll dig it out tomorrow.

 

[fizzy]

Hmm... just had a thought. When did AMD start supporting Windows 11 direct storage?

 

[fizzy]

Wouldn't automatically put the blame straight onto a faulty GPU, could be a mb pcie lane problem due to 3 onboard nvme drives.

I'm not specifically - just trying to figure the issue out and why the gfx drivers/card keep playing up when there is an issue.

As far as I recall, the top NVME slot is directly connected to the CPU, and the other two are connected to the X570 chipset. So moved my OS drive NVME into the top slot, and the two WD NVMEs into the middle and lower slots.

Tried copying file between the WD drives again.... and same issue. At least it rules out an issue with the top M2 slot and cpu connection to it. Again, card was disabled in Windows DM on reboot and had to enable and reboot again to bring it back to life. I believe top PCIE slot is also directly connected to CPU.

Done crystal disk benchmark on each WD drive and that was fine and got expected read/write speeds. Disk checks are all ok too.

I think it's getting to the stage where I'll have to start doing parts swaps to isolate cause - don't have any other board or cpu to try though. OCUK don't seem to have X570 boards listed anymore, so would probably need to go for a B550 board if I was to buy a replacement.... which only has 2 M2 slots anyway.

 

[fizzy]

You might want to email the manufacturer. I know it is unlikely, but it is possible there's some kind of compatibility or setting issue and they might have the stuff on hand to test it for you.

I built the system myself, a few years ago so all out of hardware support now. I've built my own PCs (and plenty of other peoples too) for years.

 

[fizzy]

I don't mean as general tech support, more like a possible BIOS issue that they need to address.

Will be lucky getting tech support from ASRock on this, or any really on a board this old. I believe most of the bios is based on the amd AGESA component that they have no say over, and their latest bios is already using 1.2.0.B. The previous bios I was using used AGESA 1.2.0.8, so tried at least 2 different bios/agesa versions with same issue.

Used 2 PCIE3 nvmes for ages without issues, and added in a 3rd PCIE4 NVME later. Only started noticing this issue when moving large files between nvmes at high speed. I also have a couple of SATA SSDs and no issues moving files between them and the NVME drives. It's all very odd and a pretty specific scenario, just complicated by the strange behaviour of the GFX.

 

[fizzy]

If it's within DSR I'd send it back, wait for the pricedrop to happen and try again with another brand (or the Pulse, had that excellent card).

The GFX card I've had several months now. Stress testing, playing games etc is fine. It's only on this specific scenario I'm seeing this odd issue.

More testing I think - although individual components so far have all been testing ok.

 

[fizzy]

Ok, some further testing. Curious to check whether the issue is with the NVME drive locking up or just something relating to GFX drivers.

So, cleared the WD Gen4 drive, and kicked off a robocopy job in command line to copy all contents from WD Gen3 drive (both 1TB drives, source about 80% full). Off it started and within a few seconds, display off and GFX fans ramp up full. Left it for a while until disk activity light stopped. Power cycled pc and checked..... all contents copied, so disks were operating fine whilst the display was off. Had to renable card in windows device manager again and reboot to get it fully working again. Still no idea why the GFX card is being affected like this.

I've dug out my Radeon VII which I will try - but it's a PCIE Gen 3 card rather than 4, different arch etc so likely uses a different subset of drivers in the package. Will give it a go for the sake of it.

 

[fizzy]

Ok - turns out that the radeon VII driver is in a completely different driver set (vega-polaris rather than RDNA), and latest is 23.11.1. Uninstalled existing drivers, installed the Radeon VII ones.

Tried robocopy job again and looked like it had completed. Checked and the dest drive had dissappeared. Restart and checked - showing as half full so didn't properly complete. Tried again and after a while robocopy started reporting "A device which does not exist was specified". Drat. GFX stayed fine though. Drive temps didn't change much either in hwinfo. Strange that copy completed ok before.

Ok - at least this seems to have rulled out a few things. The other NVME drives are PCIE3 models, so either I've a fault on the PCIE4 nvme, or there's an issue with PCIE4 on my machine for some reason. Will have to dig through the bios and see if I can force the NVME and GFX modes to PCIE3 and see if problem persists then.

If it does turn out to be somehow related to PCIE4 modes, I'm guessing that's going to be a cpu or board issue. Oh the joys of PCs and fault finding.