Crash with 4K renders

I've been rendering 4K video via the super scale feature for well over a year now. It's been working great!

Recently my machine has been crashing when rendering 4K. This seems to be a new problem...

The computer continues to run, but the fans on my GPUs ramp to max and the screens go black. It will stay in this state until I hard restart manually. I'm guessing my GUPs (2 AMD Radeon RX 580s 8GB) are overheating during the rendering...

Was there a change in the most recent update that could be contributing here? I'm just curious why this all of a sudden started to happen on my machine.

Is there a way to decrease GPU load when rendering with longer render times? Or any other ideas here?


Decode H.264/H.265 using hardware accelerations (AMD) = checked
Adrenalin Edition 22.3.1 - Up To Date
Crossfire enabled
GUP Workload = Graphics

----

Open to any ideas. Thanks!

Just made sure Davinci was updated to the most recent version (which it wasn't) and tried again.

Same results. Crashed around 91% though my video. The hottest of the two GPUs was 81C when it crashed.

I just tried again... but this time with H.264 encoding instead of H.265. Didn't crash.

As a recap - my machine is still crashing when I try to 4K supersize render.

Any ideas?

It's an overheating issue. When I turn my fans up to 95% with MSI afterburner... I get no crash during render. Is there a way to throttle the rendering in DResolve? Or maybe AMD?

Crossfire enabled!

Disable that. Physically break the connection.

Make sure the GPU fans and heatsinks, and any other airways in the computer, are not clogged up with dust.
It's amazing how good the fans are at sucking in dirt.

Case is clean


I just disabled crossfire and ran it again. Seemed to be doing ok and leveling out on temp... but ended up failing around 80% of the render.

When it fails the fans keep running at max and the displays shut off.

Could be bad hardware or driver.

Could be bad hardware or driver.

Drivers are up to date.

That doesn't mean they're properly functioning, though. I recall at least two occasions where the latest Studio driver from nVidia had bugs and folks were advised to roll back for a while.

That doesn't mean they're properly functioning, though. I recall at least two occasions where the latest Studio driver from nVidia had bugs and folks were advised to roll back for a while.

I would think I wouldn't be the only one having issues then. Right?

When I go to preferences and manually select just my discrete GPU (not the one that is running my displays)... it will render all the way though without issue. Max temp gets to 80C on that GPU. Fan levels out at 75% based on stock fan curves.

Render just takes ~2x the time! lol...

It's strange that it stopped working... Both GPUs were having no issues rendering 4k together for months. The only way I can get though a render is to crank both fans up to 100% before starting the render.

I wish there way a way to... I guess use 50% of my display GPU to keep the temp down. Or something like that. So strange.

1) try to render to a intermediate DNX/Prores file
2) open your computer and with a good compressor clean all the dust off. (do it regularly)
3) see if you can go to the BIOS and force all the fans at max speed when render regardless of temperature
4) add fans....
5) switch for a CPU water cooling system if possible - it will keep the CPU cooler and immediately have the heat pushed out instead of lingering in the case
6) reduce/remove overclock.
7) reduce the frame speed for the render form [maximum] to something like [25], it will take longer bu overheat less.
8) use the GPU with the monitor attached for only GUI and use the second one for calculation...

If I try some massive ArriRAW 6.5K footage in an UHD timeline with the new Depth Map function on a PC laptop (Asus ProArt) with Nvidia RTX 3070, it crashes if rendering to H.265 or even DNxHR. It barely gets through with Cineform for output, while the fans are ramping up massively to 4.800 rpm for the CPU and 4.000 for the GPU.
I think you may be hitting similar limits.
My M1 laptop (see below) is eating through that stuff not much faster, but near silently.

I would think I wouldn't be the only one having issues then. Right?

Not necessarily. I wasn't having the issues from the nVidia bugs.

crank both fans up to 100% before starting the render.

Leave 'em that way. Bad idea to throttle cooling fans.

So I'm still having this issue.... Can't get 4K footage to render with both GPUs rendering... I can only get though a render with the (non display GPU) rendering alone.

Recently I flashed the GPU bios on both cards back to factory settings... hoping this may fix it.... Didn't fix it.



Fan's running at 100%. Generally crashes when temp on the second (non display) card hits around 80C.

The computer continues to run, but the displays go black and the computer in unresponsive. The render also stops at this point from what I can tell. Takes a forced re-start to get running again.


----

Drivers are up to date and newest version of Davinci Resolve Studio installed

Log Files:

https://www.dropbox.com/s/y50dxiqxv0tdh ... 3.zip?dl=0

Ok so new development.

I can render the video without issues if using either GPU my itself (I've tested both of them independently). Temps say below 78C when doing this in the solo configuration.

When I turn them both on is when it crashes.

Anyone have any suggestions given this additional info?

I'd blame your power supply then.

Here is a video showing the issue in real time:

Here is a video showing the issue in real time:

That looks like a power issue. How are you supplying power to those 580s? If those are 2 pin you'd want to use separate power cables. With two cards, do you have separate power cables for all? If not, I would bet thats your issue. Especially if you're daisy chaining.

Here is a video showing the issue in real time:

That looks like a power issue. How are you supplying power to those 580s? If those are 2 pin you'd want to use separate power cables. With two cards, do you have separate power cables for all? If not, I would bet thats your issue. Especially if you're daisy chaining.

Each has it's own cable going to a Seasonic 650 Gold

Each has it's own cable going to a Seasonic 650 Gold

You could be tripping the PSU safety. Not sure 650W is enough for two 580 and the rest of your components when at full load.

Another thing could be the BIOS settings for PCIe lanes on those slots. If I'm not mistaken that's a MSI B450 Tomahawk and the second slot is only 2.0. It could be wonky if its set to Auto or 3.0. Try setting the first slot to 3.0 and the second one to 2.0 and see if that helps.

Each has it's own cable going to a Seasonic 650 Gold

You could be tripping the PSU safety. Not sure 650W is enough for two 580 and the rest of your components when at full load.

Another thing could be the BIOS settings for PCIe lanes on those slots. If I'm not mistaken that's a MSI B450 Tomahawk and the second slot is only 2.0. It could be wonky if its set to Auto or 3.0. Try setting the first slot to 3.0 and the second one to 2.0 and see if that helps.

What's interesting to me about tripping the PSU safety is that it runs for a good 4-8 min at full tilt before tripping it. Wouldn't it trip pretty close to being put under max load?

Here are the options in the Bios:




Sorry I just realized that's hard to read...

The options under shipset gen switch is (Auto\gen1\gen2)
The options under PCI_E1 Lanes (Auto\ 8x8x \ 4x4x4x4) - Something like that

Updated Better Pics:

B450

PCIe E1 Gen3
Chipset Gen2 (basically your second slot)
PCIe E1 lanes X8 X8

If it makes no difference switch back to auto.

But if you say it works for a while then crashes it could be also memory related. Are you running XMP?
If yes, try to disable it and retry.

Still looks like a power issue though since it just turns off...

PCIe E1 Gen3
Chipset Gen2 (basically your second slot)
PCIe E1 lanes X8 X8

If it makes no difference switch back to auto.

But if you say it works for a while then crashes it could be also memory related. Are you running XMP?
If yes, try to disable it and retry.

Still looks like a power issue though since it just turns off...

I don't think that's going to work. I changed the options as follows:



And it's no longer showing up in DResolve:




This is my Ram:

And it's no longer showing up in DResolve:




This is my Ram:

Could be because you have a nvme installed that the lanes config don't work in X8/X8.
You can switch back to Auto.

Check BIOS for XMP and disable it if its enabled. 3000Mhz is overclock. 2400 is JEDEC specs.
(Think its called A-XMP or something similar).

If this doesn't "fix" it then I'm not sure what's going on. To me it still looks like a power issue...

Also since you didn't mention, have you updated your drivers to the latest version? There was a fix for AMD crashing with H264/265 in one of the recent drivers.

I reset everything to auto and set A-XMP to disable. It was on Profile 1.

Re-ran the render test and failed.

------

I'll check drivers, but this has been a long term issue.

-----

If this is a power issue. Would replacing my two RX580s with one 6800 XT likely fix my problem?

In addition, if power is the issue... are there any settings in DR that I could select to essentially "render at 1/2 or 3/4 power"?

In addition, if power is the issue... are there any settings in DR that I could select to essentially "render at 1/2 or 3/4 power"?

You already did that by disabling one of the GPUs.

Check my reply on your other topic. One 6800XT should work. 6800 non XT is a safer bet power wise but you'll lose performance. Its hard to pinpoint power issues if you don't have another PSU to test...

P.S Put back XMP on.

In addition, if power is the issue... are there any settings in DR that I could select to essentially "render at 1/2 or 3/4 power"?

You already did that by disabling one of the GPUs.

Check my reply on your other topic. One 6800XT should work. 6800 non XT is a safer bet power wise but you'll lose performance. Its hard to pinpoint power issues if you don't have another PSU to test...

P.S Put back XMP on.

LOL, good point.

I was thinking of potentially trying to under-clock them in MSI Afterburner

---------

Thanks for the P.S.

Also I just updated the drivers, ran the render, and failed again.

----------

On the 6800 XT... I guess there is only one way to find out!

In addition, if power is the issue... are there any settings in DR that I could select to essentially "render at 1/2 or 3/4 power"?

You already did that by disabling one of the GPUs.

Check my reply on your other topic. One 6800XT should work. 6800 non XT is a safer bet power wise but you'll lose performance. Its hard to pinpoint power issues if you don't have another PSU to test...

P.S Put back XMP on.

LOL, good point.

I was thinking of potentially trying to under-clock them in MSI Afterburner

---------

Thanks for the P.S.

Also I just updated the drivers, ran the render, and failed again.

----------

On the 6800 XT... I guess there is only one way to find out!

As I've said in the other topic, if you can wait for the new gen do that.
But if it works with 1 card and its stable and not with 2 it sure does sound like some sort of power issue.

You could try undervolting and set some power limits to see what happens but its a realm I'm not comfortable giving advice over the internet .

Good luck with that.

In resolve there is a setting in export to limit the number of frames you can render, usually it is at "max", but you can scale it down to something like 5 or 10 (don't remember the specifics), that should limit the power load

In resolve there is a setting in export to limit the number of frames you can render, usually it is at "max", but you can scale it down to something like 5 or 10 (don't remember the specifics), that should limit the power load

Is this what you're talking about?

Yes, limit to 10 or 5 and see if you can get thru the render…

So I ran some render testing.... a.) to see how each card performs b.) to see how they perform with different "DR render speeds" c.) to see how the 6800 XT compares if I end up getting on.


Test 1 was just Card 1 alone
Test 2 was just Card 2 alone
Test 3 was both cards at 10 render speed
Test 4 failed with both cards are 25 speed

Would be nice to have a 17.5 speed! On test 3 the second card barley worked at all... I was watching the % usage... and on Text 4 it was darn near maxed out the whole time.

What Windows Event Viewer says?


You can use HWINFO to check how much power your system is spending. It shows for GPU's and CPU
https://www.hwinfo.com

What Windows Event Viewer says?


You can use HWINFO to check how much power your system is spending. It shows for GPU's and CPU
https://www.hwinfo.com

I've never been in Windows Event Viewer, not sure what I'm looking for there.

On the HWINFO program... pretty cool. It shows my Card 1 is working at around 305W and Card 2 is working at around 250W when rendering (peaks). The CPU around 60W,

What Windows Event Viewer says?


You can use HWINFO to check how much power your system is spending. It shows for GPU's and CPU
https://www.hwinfo.com

I've never been in Windows Event Viewer, not sure what I'm looking for there.

On the HWINFO program... pretty cool. It shows my Card 1 is working at around 305W and Card 2 is working at around 250W when rendering (peaks). The CPU around 60W,

Sounds like a power issue. Remember 650W is not really 650W...there's overhead and efficiency loss.
305+205+60 not counting for the other components is pretty close to 650.

Yes, as other have said and the HWINFO data you collected seem to reinforce it seems your PSU is at limit with 2 cards.

Thanks for the help yall. It's interesting the variance on the two cards at 100% utilization. The card that runs at the lower wattage aslo has the better rendering time... but it's the card that is not running the display.

Anyways... If I can get one newer card (hopefully the 6800XT) that keeps the wattage below 500... I think I'll be in the clear.

Get a 850w gold PSU and swap it, should not cost you more than a couple of hundreds bucks and you’re in business…

(Of course the next thing will be heat…. But who count!)

Get a 850w gold PSU and swap it, should not cost you more than a couple of hundreds bucks and you’re in business…

(Of course the next thing will be heat…. But who count!)

I took this advice! I just installed an 850w gold in my machine... about $120 bucks (cheaper than a 3080 for sure....)


These are the results on the same test render I did above with individual cards... this render would crash my machine every time when running both cards:

----------------------------------------------------------------------
Test Render 5 (SS4K/DefaultMP4Settings/H265) - (Both Cards at render speed 100 new PSU):

Set Up:
RemoteDesktop Open (right screen)
Chrome and all browsers Closed
Starting GPU temps (Card1 = 37, Card 2 = 25)
Fans both running at 95%

Results:
Peak Temp = (Card 1 = 80, Card 2 = 69)
Peak Wats = (Card 1 = 162, Card 2 = 131)

Time to Render = 05:33 min

---------------------------------------


That's a 46% reduction in rendering times with a new PSU! Thanks yall

Wonderful

Well I'm back! ... I went to render a 4K video that's larger than the "test render" I was using for evaluating success in the above posts.

About 58% of the way though it crashes like it used to. Even at 50% rendering speed with both GPUs running. It gets all the way though with one GPU running... Makes me think it's an issue with GPU#1....

More so than a power issue.

Check in your BIOS that both cards receive the right amount of “lanes”, and try to swap gpu1 to gpu2 (or test each one ondividually)