From 774d4fc3249ccfb568096c8b6bd9c6d54541be77 Mon Sep 17 00:00:00 2001 From: Gabriel Stein Date: Wed, 12 Aug 2026 13:39:02 -0700 Subject: [PATCH] build: add Pascal (sm_61) to CUDA architectures MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The published CUDA whisper-server carried kernels for Turing (7.5) through Blackwell only. On Pascal cards (GTX 10 series, compute 6.1) the server starts, loads the model into VRAM, then aborts at the first kernel launch with 'no kernel image is available for execution on the device' — a two-day field debugging incident on a GTX 1080 Ti. Adding 61 ships Pascal SASS; its embedded PTX also covers Volta (70) via driver JIT. Release notes updated to match. --- .github/workflows/build-binaries.yml | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) diff --git a/.github/workflows/build-binaries.yml b/.github/workflows/build-binaries.yml index 13218ecc10a..4692ba21571 100644 --- a/.github/workflows/build-binaries.yml +++ b/.github/workflows/build-binaries.yml @@ -15,8 +15,11 @@ permissions: contents: write env: - # RTX 20-50 (Turing through Blackwell). sm_120 requires CUDA Toolkit >= 12.8. - CUDA_ARCHITECTURES: "75;80;86;89;120" + # GTX 10 series through RTX 50 (Pascal through Blackwell). sm_120 requires + # CUDA Toolkit >= 12.8. Without sm_61 the server loads the model on a Pascal + # card and aborts at the first kernel launch ("no kernel image is available"); + # 61's embedded PTX also covers Volta (sm_70) via driver JIT. + CUDA_ARCHITECTURES: "61;75;80;86;89;120" jobs: build-macos-arm64: @@ -711,7 +714,7 @@ jobs: - **All other platforms**: No requirements. ## Supported NVIDIA GPUs (CUDA builds) - Compute capabilities 7.5+ (RTX 2000 series through RTX 40 series) + Compute capabilities 6.1+ (GTX 10 series and newer) files: release/* draft: false prerelease: false