GPU compatability and recommendations

[ralphpeterson]

Hey all,

We're purchasing some new workstations/GPUs that we will use a lot for sleap tracking. Are there any compatibility issues that we should be aware of with newer NVIDIA GPUs (e.g. GeForce RTX 4080 SUPER 16G)?

Also, do you recommend a particular GPU for optimal inference speed?

Thanks!

Ralph

[github-actions[bot]]

Thank you for starting a new discussion!

We appreciate your input and will review it soon.

Warning

A friendly reminder that this is a public forum. Please be cautious when clicking links, downloading files, or running scripts posted by others.

• Always verify the credibility of links and code.
• Avoid running scripts or installing files from untrusted sources.
• If you're unsure, ask for clarification before proceeding.

Stay safe and happy SLEAPing!

Best regards,
The Team

[talmo]

Hi @ralphpeterson,

Any 40xx series NVIDIA GPU should work fine! The faster the better, but inference is most likely going to be bottlenecked by I/O and tracking (if you're doing that and not using ID models).

We don't have systematic benchmarks on newer cards, but we do a lot of inference on A6000, A5000, A40 and 4080 cards.

In general, more GPU memory is preferable so you can handle larger image sizes and increase the batch size for throughput (though in the past we saw a sharp plateau after batch size ~16, probably due to the bottleneck in transferring images to the GPU).

The 4080 SUPER 16G is a great mid-to-high end pick that will get you good bang for your buck at the 16GB GPU memory range.

[ralphpeterson]

Great! Thanks Talmo.