fix(supraseal): add Turing (sm_75) back to CUDA architectures#1038
Merged
fix(supraseal): add Turing (sm_75) back to CUDA architectures#1038
Conversation
When CUDA 13 support was added, Volta (sm_70) was correctly removed as CUDA 13 dropped it. However, Turing (sm_75) was also removed even though CUDA 13 still supports it. This causes a segfault on `./curio --version` when built on machines with Turing GPUs (GTX 1650/1660, RTX 2060-2080, Tesla T4). The CUDA kernels in pc2.cu are compiled only for sm_80+, and the CUDA runtime crashes during binary initialization when no compatible kernel is found for sm_75. Binaries built on Ampere+ machines work fine because the embedded kernels match. But building locally on a Turing machine produces a broken binary. Adds -gencode arch=compute_75,code=sm_75 back to CUDA_ARCH.
Contributor
|
great catch! |
Contributor
|
My mistake: the AI says no sm_75 even though the docs say it'll work. |
snadrus
approved these changes
Feb 19, 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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem
Building Curio on machines with Turing GPUs (RTX 2060-2080, GTX 1650/1660, Tesla T4) produces a binary that segfaults immediately — even on
./curio --version.Root Cause
When CUDA 13 support was added to
extern/supraseal/build.sh, Volta (sm_70) was correctly removed (CUDA 13 dropped it). However, Turing (sm_75) was also removed even though CUDA 13 still fully supports it:The
CUDA_ARCHflags only included sm_80+:CUDA_ARCH="-arch=sm_80 -gencode arch=compute_80,code=sm_80 ..."The CUDA kernels in
pc2.cuare compiled intolibsupraseal.aand linked into the curio binary. When the CUDA runtime initializes at binary load time and finds no compatible kernel for sm_75, supraseal's native code crashes → segfault beforemain()even runs.Binaries built on Ampere+ machines work fine when copied to Turing machines because the CUDA kernels match the build machine's GPU, and supraseal tasks aren't invoked at startup.
Fix
Affected Hardware
Any Turing GPU building locally with CUDA 13:
Workaround (until this merges)
Or copy a binary built on an Ampere+ machine.