Skip to content

Support RTX 50-series GPUs in Linux CUDA Docker builds - #32548

Merged
Tianlei Wu (tianleiwu) merged 2 commits into
mainfrom
copilot/fix-quickgelu-cuda-error
Sep 11, 2026
Merged

Support RTX 50-series GPUs in Linux CUDA Docker builds#32548
Tianlei Wu (tianleiwu) merged 2 commits into
mainfrom
copilot/fix-quickgelu-cuda-error

Conversation

Copilot AI commented Sep 10, 2026

Copy link
Copy Markdown
Contributor

Description

  • Upgrade the default CUDA image from 12.6.1 to 12.8.1.
  • Add compute capability 12.0 to the Docker build architecture list.
  • Align the documented build command with the new CUDA default.

Motivation and Context

The Linux CUDA Dockerfile produced binaries without an RTX 50-series kernel image, causing cudaErrorNoKernelImageForDevice when executing CUDA operators such as QuickGelu on an RTX 5070 Ti. CUDA 12.8 adds the required sm_120 compiler support.

@azure-pipelines

Copy link
Copy Markdown
Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

Co-authored-by: tianleiwu <30328909+tianleiwu@users.noreply.github.com>
Copilot AI changed the title [WIP] Fix CUDA error for QuickGelu node execution Support RTX 50-series GPUs in Linux CUDA Docker builds Sep 10, 2026
@tianleiwu
Tianlei Wu (tianleiwu) marked this pull request as ready for review September 11, 2026 22:37
Copilot AI balanced review requested due to automatic review settings September 11, 2026 22:37
@tianleiwu
Tianlei Wu (tianleiwu) merged commit a7df32c into main Sep 11, 2026
90 of 91 checks passed
@tianleiwu
Tianlei Wu (tianleiwu) deleted the copilot/fix-quickgelu-cuda-error branch September 11, 2026 22:39

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟡 Changes recommended

The unchanged cuDNN 9.5 dependency does not officially support Blackwell/SM120.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Updates the Linux CUDA Docker build for RTX 50-series GPUs.

Changes:

  • Upgrades CUDA from 12.6.1 to 12.8.1.
  • Adds compute capability 12.0.
  • Updates the documented build command.
File summaries
File Description
dockerfiles/Dockerfile.cuda Updates CUDA and GPU architecture defaults.
dockerfiles/README.md Aligns documentation with CUDA 12.8.1.
Review details
  • Files reviewed: 2/2 changed files
  • Comments generated: 1
  • Review effort level: Balanced

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

# If memory is less than 64GB, you may change "--parallel" to "--parallel 4" to avoid out-of-memory error.

ARG CUDA_VERSION=12.6.1
ARG CUDA_VERSION=12.8.1
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Build] on-zero status code returned while running QuickGelu node. No kernel image is available for execution on the device, RTX 5070ti

4 participants