Allow setting custom cuda compute capabilities #2431
Merged
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
As you mentioned in a previous issue, autodetecting the cuda capabilities might cause some issues, where people build dlib on their local fancy GPU, and they deploy on another system with an older GPU...
(I had this problem last week at work, but the other way around, the fancy GPU was on the deployment side, and didn't support
sm_50
)For this reason, this PR is quite conservative: if the user specifies no
DLIB_USE_CUDA_COMPUTE_CAPABILITIES
, then it will use thesm_50
as before.However, setting
DLIB_USE_CUDA_COMPUTE_CAPABILITIES=61,75
will build dlib with those targets instead.What do you think?