Resolve 16/17 - RHEL/CentOS 7 specific issues

Dear Blackmagic Support Team,

the Linux version behaves very strange when it comes to UI bugs, GPU scopes and performance.

a) Black timeline: when switching from Color to Edit, we often find various elements of the edit page completely blacked out. This occurs faster on longform content, but pretty much happens instantly on some projects. A reboot of the software is required.

b) Color node labels are often destroyed aswell when switching between workspaces for Color and Edit. This happens pretty much instantly when switching for one single time from Color to Edit and vice versa.
Also the performance coupled together with the use of the GPU scopes is not usable at all. When using GPU scopes, we barely get 10-11 fps, when deactivating GPU scopes we instantely get fluid playback on the color page. The Nvidia driver only shows 25% use of VRAM both with GPU scopes enabled or disabled.

We run Resolve 16.2.8/17 Build 39 on CentOS 7.8 with 3x 2080Tis and 1x K4000 as a GUI card. Nvidia driver version is 460.56. Studio version, Decklink 8K Pro running Desktop Video 11.7. Another machine without a Decklink Card and 3x 1080Tis shows the exact same behaviour. Studio version with Dongle. Both in a shared DB environment and on local databases.

I've attached screenshots of the UI bugs to this thread.

Push.

Anyone?

I'm not from BMD, just a linux resolve user ... all I can say is that I've never seen this ... sorry can't really help more.

I'd suggest posting the Debug logs taken just after the problem occured, that might give an indication.

The video / audio track names also disappear.

I get the same errors. Talked to BMD support months ago, but they only told me to update the nvidia driver. I did that but it didn‘t change anything. Didn‘t have time to investigate further.. but tried a complete new install half a year ago, but it didn‘t change anything..

It‘s quite bothersome.

Have similiar specs then you. Centos 7.7
Supermicro
3x1080TI
Lates resolve 16 version

Did you try with a single GPU ?

Hey Lee,

same happened to me - reinstall Resolve, reinstall Nvidia drivers, after I sent them debug logs etc. I've seen this error across facilities, with multiple machines all running CentOS 7.X It's not that big of an issue anymore, will write more at the bottom of this post.

@Lucios Snow:

Yes, that is correct aswell.

@smunaut:

Testing one GPU is not really an option, since we work at HighRes Feature content every single day and every colorist is pounding the GPUs to a maximum. It's not at all viable to "test" those kind of workarounds since we need all the GPU power we can get.

@all the people above: We lessened the use of those Linux workstations in the last couple of months since we got more and more Mac Pros 7.1 which don't behave like this at all. It's really a shame that the Linux version is disregarded by BM by such an extend where you run completely certified hardware, not Ryzen/AMD GPUs/Ubuntu/Arch/whatever distro, and still run into more problems than people with different inofficial configs. Since we work on SAN storage, we can only use RHEL/CentOS based installations anyway.

What I've found since I posted last time is that moving to CentOS 8 or in our case RHEL 8.2/8.3 seems to fix this problem. Since Red Hat released a "free" version of their distro at the end of February, I tested our config with Resolve 16.2.7 (where the switch to a newer cuda version was introduced, btw nice move from BMD to switch major API versions on the last minor release before force feeding us Resolve 17). If you want to test this out, put a new SSD in and quickly set up a Red Hat installation and opening up a GPU intense setup with many OFX plugins and a high project resolution. Keep in mind I only tested the edit and color page, not the Cut and Fairlight page.

Yes it's a shame. Apparently not that enough people run into this problem that BMD fixes it.
Don't get me wrong, I can work with it. It's not that bad, it's just annoying.
Upgrading do CentOS 8 is not working for us, because it's not supported by our SAN and
usually upgrading CentOS versions at all, causes issues with wacom panel and other configs.
So I tend to stay with a CentOS version as long as I can, so there is no unexpected behaviour on
our main workstations.
And so far CentOS 7.7 and 7.8 were the best versions in terms of compatibility with wacom panels etc..

Hey Lee,

that is perfectly reasonably and I understand that - we only have BMD panels connected and did our StoreNext update last week to incorporate the RHEL 8 support. So it seems that you are stuck on 7.X then; I think it's more annyoing to deal with the problem if external colorists work at your facility and your only solution for them is to reboot Resolve - really breaks the workflow depending in the project and the colorist working on the edit page which happens from time to time. Sad really...

yeah thats a problem for sure. It's a bit annoying as we chose the Linux platform specifically for being the
best supported platform for high end work.

I just found, that we usually have little to no problems in the color page, besides node labels that get messed up. But the main issues are happening when doing work in the Edit page. So we do most of our Conform work on Macs now..

I'm on Linux Mint 20.1 and have been noticing strange issues as well. I mainly edit HEVC video from my Fuji X-T3. On the 16.x version of Resolve I could fluidly playback the 4k video, even the 4k60p stuff. Somewhere during the 17 Betas I noticed that the video playback would become choppy. It now is choppy even with 4k30p (29.97) video.

I'm currently running the 460.39 nVidia driver, but I have tried older versions and the behavior is the same. Yes, I'm using the Studio version.

Hey Lee,

yes, that behaviour seems familiar. Node labels are the first thing to go, then some time later it's the edit page. Especially if you open up Resolve in the Color Page and switch to Edit.

@Bill: Sorry, but your issues have nothing to do with our problems and you are running hard- and software which isn't at all what Lee and I use. I should update the thread title to reflect the topic better.

@Lucius Snow:

Yes, that is correct aswell.

I run only only one GPU on Centos 8.2 and the bug is still here on Resolve 16.2.8. I've just upgraded to Centos 8.3 and very latest nVidia drivers. We'll see if it occurs again but i'm pretty pessimist!

Anoter bug by the way: after loading another projet, the title name remains the old one in the Linux software bar. Pretty confusing.

@Lucius Snow:

Yes, that is correct aswell.

I run only only one GPU on Centos 8.2 and the bug is still here on Resolve 16.2.8. I've just upgraded to Centos 8.3 and very latest nVidia drivers. We'll see if it occurs again but i'm pretty pessimist!

Anoter bug by the way: after loading another projet, the title name remains the old one in the Linux software bar. Pretty confusing.

That is unfortunate to hear; maybe I was lucky with the tests on 8.3 or I didn't test it enough; we start a couple of new projects in the next couple of weeks, so I will do additional testing there and will report back. The problem will of course downscale with every Mac Pro additionally purchased, but I refuse that that should be the final option for a facility imho.

Hey Lee and Lucius,

we did a major update to our Supermicro machine and swapped out the 3x 1080Tis with a single (for now) RTX 3090, kept the K620 GUI card and updated to Resolve 17.1.1. Nvidia driver needed to be updated to version 460.73.01.

All I can say at the moment is that this bug has disappeared. We did two features now on that machine and both the black hole and the f'd upped labels work now as intended. Happy to hear from you two again!

Edit: we are still on CentOS 7.8 with that machine.

Hey Lee and Lucius,

we did a major update to our Supermicro machine and swapped out the 3x 1080Tis with a single (for now) RTX 3090, kept the K620 GUI card and updated to Resolve 17.1.1. Nvidia driver needed to be updated to version 460.73.01.

All I can say at the moment is that this bug has disappeared. We did two features now on that machine and both the black hole and the f'd upped labels work now as intended. Happy to hear from you two again!

Edit: we are still on CentOS 7.8 with that machine.

Oh, that is good news! Can I ask, why do you still use the K620 GUI and not just the 3090 as GUI?
BMD support asked me to remove my GUI card, as they thought that might be one of the reasons..

Do you think it has something to do with the driver, or are the multiple 1080Tis the reason?

Thanks for updating on this!

Hey Lee and Lucius,

we did a major update to our Supermicro machine and swapped out the 3x 1080Tis with a single (for now) RTX 3090, kept the K620 GUI card and updated to Resolve 17.1.1. Nvidia driver needed to be updated to version 460.73.01.

All I can say at the moment is that this bug has disappeared. We did two features now on that machine and both the black hole and the f'd upped labels work now as intended. Happy to hear from you two again!

Edit: we are still on CentOS 7.8 with that machine.

Thanks for the feedback. I still use v16 at the moment.

for the edit page, both on centos 6 and 7, occasionally the timeline editing window stop updating, either showing a old timeline or plain garbage.

however there is a quick way to refresh it without restart.

1) locate the [timeline view options] - open it
2) toggle the [stacked timelines] twice

that usually give back control to me.

Hey Lee,

we use the K620 because of our fibre channel DVI extenders and with the RTX 3090 we would have needed another adapter so we kept the K620 for now. A GUI card is also not the root of the problem, since our other machine without a GUI card had the same problems.

Since we had the problems with 3x 1080 Tis and the initial Resolve 17.0 release, we can only assume that it has to do either with a Resolve update or the GPU/VRAM itself. The two projects we did both were classical "out of VRAM" projects, containing a mix of 6K/8K footage in a 4K timeline project and lots of OFX plugins, so it's good to see that the VRAM isn't a concern anymore.

In the end we should do a final test with this setup and Resolve 16.2.7 and see if we get the black hole again. Until then the rest is guesswork.

Update: after 2,5 weeks we encountered the first black hole again. So it's still there and I have to backpaddle on my previous statements. It seems to happen way less often, so that's an improvement (?!). We have to roll back to 16.2 for an old project anyway, so I will test if the RTX 3090 results in the bug being less present than before.

Thanks Walter for the tip btw

Thank you for the update!!

Any update on this issue?

I just update our main supermicro to the latest v17 and the issue still persists..
Also I encountered a new gui issue that I have to switch twice to the window page till I‘m able to move a window with the mouse. I can still move it via the panel, but if I am switching from another page to the window page it doesn‘t let me move a window via the mouse right away..
This is quite bothersome, never had this previously, the only thing I updated was resolve to v16.
Any help / info from BMD would be greatly appreciated!!


Supermicro box
Centos 7.7
3x 1080TI
Nvidia driver 460.32.03

Any update on this issue?

I just update our main supermicro to the latest v17 and the issue still persists..
Also I encountered a new gui issue that I have to switch twice to the window page till I‘m able to move a window with the mouse. I can still move it via the panel, but if I am switching from another page to the window page it doesn‘t let me move a window via the mouse right away..
This is quite bothersome, never had this previously, the only thing I updated was resolve to v16.
Any help / info from BMD would be greatly appreciated!!


Supermicro box
Centos 7.7
3x 1080TI
Nvidia driver 460.32.03

Used 17.3 today. Still the same behavior.
It‘s not every single time, just about every other time. I can live with the labels being gone, but having to go into the window panel twice to adjust a window via mouse is just very annoying..

Hello Lee,

I was a bit out of the loop because we added another two 7.1s Mac Pros to our setup and the Supermicro now sits on a shelf without any power plugged in. I have given up on Blackmagic providing any fixes for this issue because we are not the focus group any more since a couple of years and all my tickets didn't receive any answer or bugfixes since about 4-5 months. I still have no bugfixes for their buggy Teranex / Decklink hardware so we had to buy more Aja converters to fix our SDI needs. Sorry I have no good news for you :/

All the Best,

@Lee did you try another window manager by any chance ?

TBH I'm not even sure what you mean "switch twice to the window page till I‘m able to move a window".
What's the "window page" ?

@Lee did you try another window manager by any chance ?

TBH I'm not even sure what you mean "switch twice to the window page till I‘m able to move a window".
What's the "window page" ?

Window page = davinci resolve tab where all the windows / masks are

No I haven‘t tried another window manager.. I‘m about to switch a Mac Pro soon anyway and I‘m still staying on v16 till then. It‘s just too buggy on Linux at the moment..

Ah ok, I see (window page).
All I can say is here it works fine :/

I've definitely had bugs in Resolve, but then I've seen people on windows reporting the same bugs, so I know at least those are not OS specific.