Remove sd-cpp ROCm support on Windows - #3616
Merged
Merged
Conversation
On Windows the sd-cpp ROCm build delivers no GPU acceleration at all: it runs at CPU speed, and on one test it is twice as slow as CPU. Measured on release-v2026.39 CI (run 35124275540), per test, cpu / rocm / vulkan seconds, all three passes on one runner in a single job: test_004_image_generation_with_steps 2.9 / 3.0 / 0.4 test_005_image_generation_with_cfg_scale 4.0 / 4.1 / 0.6 test_006_image_generation_with_seed 2.9 / 3.0 / 0.5 test_020_upscale_basic 39.8 / 39.9 / 0.7 test_015_image_edit_basic 5.7 / 11.7 / 0.9 Linux is unaffected and keeps ROCm: the same tests there show ROCm running 5-6x faster than CPU (test_004 2.3 -> 0.4, test_020 32.3 -> 6.2). The failure is silent, which is why it survived. The Windows ROCm pass reports "ROCm (TheRock) runtime ready for sd-cpp:rocm", passes all 22 tests and exits OK, so nothing in CI flags it and a Windows user selecting ROCm for image generation gets CPU-speed inference with no indication. Narrow the rocm support row to Linux in the descriptor, which is the single source of truth for the generated support matrices and docs. Drop the now-unreachable Windows ROCm asset filename, and the rocm pass from the Windows stable-diffusion CI job. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QY95EN9FjNYMyStH3idZZB
github-actions
Bot
added
engine::sd
jeremyfowers
marked this pull request as ready for review
September 17, 2026 17:17
superm1
approved these changes
Sep 17, 2026
github-merge-queue
Bot
removed this pull request from the merge queue due to failed status checks
Sep 17, 2026
ramkrishna2910
approved these changes
Sep 18, 2026
github-merge-queue
Bot
removed this pull request from the merge queue due to failed status checks
Sep 18, 2026
github-merge-queue
Bot
removed this pull request from the merge queue due to failed status checks
Sep 20, 2026
github-merge-queue
Bot
removed this pull request from the merge queue due to failed status checks
Sep 20, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
@ramkrishna2910 this PR removes sd:rocm windows support because its clogging up CI, doesn’t work, and we haven’t had bandwidth to fix it. We can revert this PR whenever we have bandwidth to provide fixes.
Summary
On Windows,
sd-cppon ROCm delivers no acceleration over CPU. On one test it is 2x slower than CPU.Source: run 35124275540,
release-v2026.39.Test .exe - stable-diffusionrunscpu,rocmandvulkanin one job on one runner, so this is a direct comparison. Seconds per test:test_004_image_generation_with_stepstest_005_image_generation_with_cfg_scaletest_006_image_generation_with_seedtest_020_upscale_basictest_015_image_edit_basicLinux is unaffected and keeps ROCm. Same tests,
Test .deb - stable-diffusion:test_004test_005test_006test_020_upscale_basictest_015Documentation
Breaking Changes
On Windows,
sd-cppconfigs withbackend: "rocm"no longer resolve. Migration:vulkan, which is 6-50x faster on the same hardware per the table above, orbackend: "auto". Linux unaffected.Generated by Claude Code