"tuned GPU memory" option grayed out as "not available"

Andrea_Mingaroni · July 11, 2025, 9:49am

Hi.
I use a notebook with amd CPU and alzo a dedicated nvidia gtx 3060 card.
I just migrated to fedora and compiled as usual darktable from source.
Under the “processing” tab in the options, I get this in the OpenCL section:

Is it normal that the “tuned gpu memory” says “not available”?
I don’t remember with the previous ubuntu installation but it may be that it was active.

Also, should I use “multiple GPU” as on the notbook there also is an integrated GPU on the CPU, or better use “very fast GPU”?

Thanks in advance for any help/suggestion.

kofa · July 11, 2025, 9:53am

Run from the command line and see if both GPUs are detected. The setting is only available if you have multiple GPUs.

Andrea_Mingaroni · July 11, 2025, 10:29am

I think that it sees only the discrete one.

kofa · July 11, 2025, 11:07am

Check recent threads, there has been discussion about getting AMD up and running. Although I must say, you may be better off just using the Nvidia card: it’s probably so much faster than the embedded one that you may even lose performance if some work gets executed on that one.

Andrea_Mingaroni · July 11, 2025, 11:50am

Yeah I know, anyway the core of my post was indeed about the “tuned gpu memory” option being greyed out.

g-man · July 11, 2025, 12:22pm

That feature is now disabled from the GUI. It is not a safe option that can result in dt using more memory than what it is available. You can still enable it via darktablerc but I discourage you from turning it on.

I think the plan is to was to remove it from GUI in the future but I don’t recall the GitHub discussion in detail.

kofa · July 11, 2025, 12:37pm

I already answered your question yesterday on Facebook, and I quoted the documentation of the setting (unfortunately, it’s not visible on the UI, if the option is unavailable).

github.com/darktable-org/darktable

data/darktableconfig.xml.in

93638c905


      
          <dtconfig prefs="processing" section="opencl" capability="multiopencl">
            <name>opencl_tune_headroom</name>
            <type>bool</type>
            <default>false</default>
            <shortdescription>tuned GPU memory</shortdescription>
            <longdescription>if enabled on a system with multiple OpenCL devices you may specify a safety margin per device (headroom, default is 600MB)</longdescription>
          </dtconfig>

Andrea_Mingaroni · July 11, 2025, 1:33pm

I didn’t see your response on FB until now, don’t know why. I only saw the other answer directing me here. Sorry.

Andrea_Mingaroni · July 11, 2025, 1:35pm

Perfectly clear. I was afraid that it was a compilation error on my system.

Terry · July 11, 2025, 10:30pm

I am on DT 5.3 Windows and this option is available on my laptop. So should I use it or not? From @g-man response above I presume I shouldn’t.