Is there an official Ray Docker image that supports CUDA 11.4?

1

I’m working with legacy NVIDIA hardware (Tesla K80), which only supports:

  • Driver: 470.xx

  • CUDA: 11.4

Ray’s public Docker images on Docker Hub contain GPU variants for cu121, cu123, cu124, etc., but I cannot find any Ray image tagged with CUDA 11.4 (e.g., cu114) or anything compatible with older GPUs.

I want to confirm whether Ray already provides an official CUDA 11.4 image — or if support for older CUDA versions has been intentionally dropped.

Context

  • GPU: Tesla K80

  • driver: 470.256.02

  • CUDA: 11.4

  • Ray version: 2.47.1

  • python version: 3.10

  • Platform: Ubuntu 24.04

2

Ray does not provide official Docker images for CUDA 11.4 (cu114) as of recent releases. The available GPU images on Docker Hub and in the official build scripts are for CUDA versions 11.7, 11.8, 12.1, 12.3, 12.4, 12.5, 12.6, and 12.8 (e.g., cu117, cu118, cu121, cu123, cu124, cu125, cu126, cu128), with no mention of cu114 or CUDA 11.4 support. This is confirmed by the absence of cu114 tags in the official image lists and build files, and by user discussions noting the lack of such images for legacy hardware. Support for older CUDA versions like 11.4 appears to have been dropped in favor of newer CUDA versions only Ray Docker README CUDA versions list Anyscale base images Ray Discourse.

Would you like more detail on how to build a custom Ray image for CUDA 11.4, or guidance on alternatives for legacy GPU support?

Sources:

Hint: Mention @RunLLM in the post for followups.

3

how to build a custom Ray image for CUDA 11.4