RT Crashes with 6800 XT Have Me Stumped

[alexp999]

I recently upgraded my own rig and my Dad now has my old Sapphire RX 6800 XT Nitro that I had from new, a big upgrade from his aging 980 Ti. Wanting to show him what the GPU can do I fired up some games and benchmarks, all good other than some coil whine from his PSU.

That was until I tried to show him some RT effects. With RT on, the Witcher 3 will crash within a few seconds of loading a save, RT off no problem. Cyberpunk will crash about 50% of the time during the benchmark with RT on. Wanting to try something that wasn't the Red engine, I downloaded the UE based Bright Memory benchmark. Again it would crash about 50% of the time.

When I say crash, in all instances it is a soft crash, never a hard crash, the game will crash and trigger the AMD issue detector driver timeout error, along with the engines own crash error. Thinking it may be PSU related I tried OCCT power test with GPU set to +15% and no problem. Any game or benchmark without RT even with the GPU pulling 330W is also fine, again apart from the coil whine.

I used the Bright Memory benchmark to play around with some settings. I found that putting the GPU power tuning settings to Quiet mode (270W TBP), it could loop the benchmark for 30 mins without crashing, conversely putting it up to 15% power increase it would crash 100% of the time. It still crashes in Witcher 3 with RT all the time, but that change in Bright Memory to me was enough to be 99.9% sure it was his 13 year old HX1050, especially as it was whining under load. I assumed the GPU was becoming unstable when trying to run the more intensive RT calculations.

I fit a NZXT C850 (the newer ATX 3.1 version) and although it's "only" 850W with it being designed for the power spikes of modern hardware, it should still be plenty. Cue my frustration when it doesn't fix the issue!!

Things I've tried:
Update BIOS, DDU, latest and 2 previous driver versions, uninstall or stop non-critical background software, including anything which could be polling hardware. Turn Re-sizable bar off, set graphics driver settings to default, try memory sticks individually, run the RAM at less than half its rated speed. Probably more!

I've never had anything stump me quite like this and I'm running out of ideas. I was so sure the PSU was the culprit. His specs for completeness:

Ryzen 3800X under Freezer II Pro 360
Crosshair VI Hero (BIOS 8801)
2x 16GB Corsair 3600 CL18
Sapphire RX 6800 XT Nitro
Was a Corsair HX1050 PSU, now a NZXT C850 (ATX 3.1)
Win 11 Pro

Any ideas at all at this point, if I haven't already tried them I'm up for giving it a go.

TL;DR: Ray Tracing causes crashes, worse with higher power use but it's not the PSU.

 

[tommybhoy]

1. Max the Power Limit in Adrenaline.

2. If RT was fine on your system, reinstall a clean Windows preferably on a spare HD-then you don't 'change' anything and revert back to his current setup if it's not a Windows issue.

Install Windows, gpu driver, MB chipset driver and games that wouldn't run.

 

[Tetras]

I'm 90% sure this has been an issue on the forums before with a 6800 crashing in RT and the fix was a complete reinstall, though, this doesn't suggest that:

When I say crash, in all instances it is a soft crash, never a hard crash, the game will crash and trigger the AMD issue detector driver timeout error, along with the engines own crash error. Thinking it may be PSU related I tried OCCT power test with GPU set to +15% and no problem. Any game or benchmark without RT even with the GPU pulling 330W is also fine, again apart from the coil whine.

I used the Bright Memory benchmark to play around with some settings. I found that putting the GPU power tuning settings to Quiet mode (270W TBP), it could loop the benchmark for 30 mins without crashing, conversely putting it up to 15% power increase it would crash 100% of the time.

 

[Poneros]

The clocks are not stable. If you can run it fine with lower clocks and/or lower power then it's an issue with a clear solution (lower them). RT stresses different parts of the GPU, so just because the GPU runs fine otherwise doesn't mean it's actually stable.

 

[alexp999]

Clean install of Win 11 24H2 on a spare SSD, disconnected from the internet, just installed chipset and GPU driver. Set GPU power to +15% and it crashed during a RT demo.

Witcher 3 was still crashing with RT on and limited to 30 fps, so it's only pulling 150W or so.

The card isn't being overclocked and it was running perfectly fine in my system right up until I moved it over.

Only reason I'm touching the power limit is because it speeds up my testing so everytime I change something it doesn't take too long to see if it fixed it.

I think I'm going to have to put it back in my PC and see if same software and power limit is fine just for a sanity check on the card still being ok.

 

[Rawr]

Clean install of Win 11 24H2 on a spare SSD, disconnected from the internet, just installed chipset and GPU driver. Set GPU power to +15% and it crashed during a RT demo.

Witcher 3 was still crashing with RT on and limited to 30 fps, so it's only pulling 150W or so.

The card isn't being overclocked and it was running perfectly fine in my system right up until I moved it over.

Only reason I'm touching the power limit is because it speeds up my testing so everytime I change something it doesn't take too long to see if it fixed it.

I think I'm going to have to put it back in my PC and see if same software and power limit is fine just for a sanity check on the card still being ok.

Even using less power it can still clock high in a way that might not be stable. Try doing a underclock on afterburner.

 

[alexp999]

Even using less power it can still clock high in a way that might not be stable. Try doing a underclock on afterburner.

I'll give it a try, but it shouldn't need downclocking. It's not that old and hasn't been overvolted or anything to degrade it and I never had an issue with it in my system.

 

[Rawr]

I'll give it a try, but it shouldn't need downclocking. It's not that old and hasn't been overvolted or anything to degrade it and I never had an issue with it in my system.

Well it's more just to narrow down the trigger, and give you a backup option if you can't solve it - you might be able to make a custom curve that retains performance while avoiding anything problematic with the chip.

Though I will say CPUs/GPUs etc can just degrade from use, they don't necessarily need to be overclocked, that just can speed it up since more power/more heat.

 

[alexp999]

Well it's more just to narrow down the trigger, and give you a backup option if you can't solve it - you might be able to make a custom curve that retains performance while avoiding anything problematic with the chip.

Though I will say CPUs/GPUs etc can just degrade from use, they don't necessarily need to be overclocked, that just can speed it up since more power/more heat.

Doing some digging I found this, exactly the same problem. Only way OP seemed to solve it was capping the max boost clock. But it's bugging me why my system didn't have the issue, something about PCI-E 3.0? Or the MB platform? I wonder if there is a PCI-E spread spectrum setting any more.

Sapphire RX 6800 XT Pulse Crashing with Ray Tracing enabled?

Received the same (original) card back, apparently no issues. Result, same issue. Reinstalled Windows, updated drivers and everything. Nothing else but the benchmarks to test and crash. First tried the Port Royal, black screen, back to desktop. Then tried the DirectX Raytraicing feature test...

forums.overclockers.co.uk

@Drumroll did you ever solve the issue beyond capping the boost clock?

 

[alexp999]

Okay so setting the max clock in Adrenaline to 95% stops it crashing in Witcher 3, at least in the few mins I tested it when it would crash 100% of the time before.

The only other thing I've noticed is that GPU-Z does not show any values or sensors for the clock speeds. Not for GPU, boost, VRAM, etc. They are all 0 MHz and no sensor for it. This seems a bit suspicious/coincidental if it's related to the boost clock?

Do I need to reflash the vBIOS or something? Or something else causing it?

 

[Kirby Wurm]

The only other thing I've noticed is that GPU-Z does not show any values or sensors for the clock speeds

Does it report these in another app like MSI Afterburner?