Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add support FP8 for Ada Lovelace and Hopper #74

Open
2 tasks done
neurogen-dev opened this issue May 4, 2023 · 2 comments
Open
2 tasks done

Add support FP8 for Ada Lovelace and Hopper #74

neurogen-dev opened this issue May 4, 2023 · 2 comments
Labels
enhancement New feature or request

Comments

@neurogen-dev
Copy link

Is your feature request related to a problem? Please describe.

TensortRT 8.6.1 update was released and, judging by the list of changes, it added the ability to build with the fp8 flag. Since Cuda 12.1 Ada Lovelays also support FP8, as does Hopper. Perhaps being able to build with FP8 would give good acceleration for these GPUs.
image

Describe the solution you'd like

Build engine with FP8

Describe alternatives you've considered

No response

Additional context

No response

Validations

  • Read the docs.
  • Check that there isn't already an issue that asks for the same feature to avoid creating a duplicate.
@ddPn08 ddPn08 added the enhancement New feature or request label May 4, 2023
@ddPn08
Copy link
Owner

ddPn08 commented May 4, 2023

thank you. That's something I would love to implement. However, there is probably no machine in my environment where I can try fp8, so it may take some time.

@neurogen-dev
Copy link
Author

thank you. That's something I would love to implement. However, there is probably no machine in my environment where I can try fp8, so it may take some time.

I have an RX 4090 and if you needed help testing (on Windows 11 or Ubuntu 22.04), I could help you.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement New feature or request
Projects
None yet
Development

No branches or pull requests

2 participants