[PR #464] Update vendored CUTLASS to v4.8.0 #468

New issue

Open

opened 2026-10-26 21:18:14 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 21:18:14 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepGEMM/pull/464
Author: @fhgffy
Created: 10/4/2026
Status: 🔄 Open

Base: main ← Head: cutlass-4.8.0-sm107

📝 Commits (1)

84dfabd Update vendored CUTLASS to 4.8.0 for SM107 (Rubin) support

📊 Changes

1 file changed (+1 additions, -1 deletions)

View changed files

📝 third-party/cutlass (+1 -1)

📄 Description

Update the vendored CUTLASS submodule from v4.2.1 (f3fde583) to v4.8.0 (098de2a6), the exact upstream release tag. This brings in the CuTe/CUTLASS building blocks for SM107, but is a dependency update rather than a claim that all DeepGEMM kernels have been validated on Rubin.

DeepGEMM already includes the CUTLASS version in its JIT cache key. The pinned DeepJIT generates architecture targets from the device's major/minor capability; no DeepJIT pointer change is included.

Validation: initialized both submodules, verified the v4.8.0 tag and version macros, checked all 26 referenced CUTLASS/CuTe headers, and reviewed the architecture dispatch and cache key. No CUDA build or SM90/SM100/SM107 device test was run: this machine has no NVCC toolkit or supported target GPU. The upstream release notes describe initial SM107 support and note that running Rubin kernels requires the R615 driver associated with CUDA 13.4 GA; the R610 developer-preview driver is insufficient.

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/deepseek-ai/DeepGEMM/pull/464 **Author:** [@fhgffy](https://github.com/fhgffy) **Created:** 10/4/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `cutlass-4.8.0-sm107` --- ### 📝 Commits (1) - [`84dfabd`](https://github.com/deepseek-ai/DeepGEMM/commit/84dfabd36d9dcf8829b955e1bc729f4498743ca7) Update vendored CUTLASS to 4.8.0 for SM107 (Rubin) support ### 📊 Changes **1 file changed** (+1 additions, -1 deletions) <details> <summary>View changed files</summary> 📝 `third-party/cutlass` (+1 -1) </details> ### 📄 Description Update the vendored CUTLASS submodule from v4.2.1 (`f3fde583`) to v4.8.0 (`098de2a6`), the exact upstream release tag. This brings in the CuTe/CUTLASS building blocks for SM107, but is a dependency update rather than a claim that all DeepGEMM kernels have been validated on Rubin. DeepGEMM already includes the CUTLASS version in its JIT cache key. The pinned DeepJIT generates architecture targets from the device's major/minor capability; no DeepJIT pointer change is included. Validation: initialized both submodules, verified the v4.8.0 tag and version macros, checked all 26 referenced CUTLASS/CuTe headers, and reviewed the architecture dispatch and cache key. No CUDA build or SM90/SM100/SM107 device test was run: this machine has no NVCC toolkit or supported target GPU. The [upstream release notes](https://github.com/NVIDIA/cutlass/releases/tag/v4.8.0) describe initial SM107 support and note that running Rubin kernels requires the R615 driver associated with CUDA 13.4 GA; the R610 developer-preview driver is insufficient. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi added the

pull-request

label

2026-10-26 21:18:14 +00:00

bsakaguchi referenced this issue

2026-10-26 21:18:15 +00:00

[PR #468] [CLOSED] [DeepGEMM] retire actlize v0.5.0 and use v1.0.0 only #471