Skip to content

Conversation

@jenshannoschwalm
Copy link
Collaborator

@jenshannoschwalm jenshannoschwalm commented Jan 20, 2026

For testing purposes we can start darktable via --conf resourcelevel="xxx" setting resources to defined values - if the machine is itself sufficient.

This new resourcelevel simulates behaviour on todays common systems: 16GB available memory, 64MB single buffer, 1GB mipmap cache and 5.5GB of available graphics memory (like on a 8GB card) for better testing situations.

@elstoc for the manual it would be a one-liner in special topics -> memory&performance tuning -> darktable resources

"common2026” (16GB ram, 64MB single buffer, 1GB thumbnail cache, 5.5GB OpenCL memory)

@wpferguson
Copy link
Member

Hmmm learned something new today. Guess I need to go RTFM (again).

@TurboGit TurboGit added this to the 5.6 milestone Jan 20, 2026
For testing purposes we can start darktable via --conf resourcelevel="xxx" setting
resources to defined values - if the machine is itself sufficient.

This new resourcelevel simulates behaviour on todays common systems:
16GB available memory, 64MB single buffer, 1GB mipmap cache and 5.5GB of available graphics memory
for better testing situations.
@elstoc elstoc added the documentation-pending a documentation work is required label Jan 23, 2026
@elstoc
Copy link
Contributor

elstoc commented Jan 23, 2026

That's assuming quite a lot of GPU memory used by the OS. I don't think mine ever gets nearly that high (not that I'll be using this setting anyway)

@jenshannoschwalm
Copy link
Collaborator Author

That's assuming quite a lot of GPU memory used by the OS.

That's how opencl memory is currently calculated by default :-) So use that for simulations how common user systems would behave.

I very well remember a) the discussions how much OpenCL memory dt should be allowed to use b) people "complaining" we have a too conservative strategy and c) my frustrating experiments to get the memory we "currently have available on the system". Unfortunately there is no portable way to know as not in specs. All vendors implement some sort-of but not cross-vendor. A check-after-alloc as we do all over the OpenCL code works on some systems, on some it doesn't.

Wayland Gnome plus some firefox tabs ... it's adding up :-)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

documentation-pending a documentation work is required scope: debugging

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants