enne2/qwentts.cpp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

build: split CUDA Docker images into cuda12 (12.9.x) / cuda13 (13.3.x), fix Blackwell Ultra arch gate (#23)

Browse Source 
The CMakeLists.txt >=12.8 CUDA-architecture branch claimed 121a-real
(Blackwell Ultra) alongside 120a-real, but nvcc from CUDA 12.8.1
rejects compute_121 ("nvcc fatal: Unsupported gpu architecture
'compute_121'"). Confirmed against a real build: 12.8.1 compiles
cleanly with 120a-real but not 121a-real; 12.9.2 compiles both. Split
the >=12.8 branch into >=12.9 (full Blackwell + Ultra) and >=12.8
(Blackwell only, no Ultra).

Since no single CUDA toolkit spans the full arch range -- 12.9.x is
the newest 12.x that still emits Pascal (61-real) SASS, 13.x drops
Pascal entirely but is otherwise more current -- the single :cuda
image can no longer serve both audiences. Split it into two CI
variants, :cuda12 (12.9.2, Pascal through Blackwell Ultra) and
:cuda13 (13.3.1, Turing and newer). The Dockerfile's own --target
cuda stage is unchanged; the CUDA_BUILD_IMAGE/CUDA_RUNTIME_IMAGE ARGs
now default to 12.9.2 (was 12.4.1) and the CI matrix overrides them
per variant.

Both variants were build-tested against nvidia/cuda:12.8.1 and 12.9.2
locally and exercised with real end-to-end synthesis requests on an
actual sm_61 card (GTX 1070 Max-Q) -- RTF ~0.4 on both, no kernel
image / arch mismatch errors.

Co-authored-by: Gary <gitea@gerasch.dev>
This commit is contained in:
Stefan Gerasch
2026-08-07 11:26:45 +02:00
committed by GitHub
co-authored by Gary
parent 37d62e42a0
commit a8a7716b53
4 changed files with 83 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files
CMakeLists.txt
+5 -2
View File
@@ -51,14 +51,17 @@ endif()
# supported card and doesn't need to seed forward JIT compat for anything
# older, unlike the 75-virtual baseline which does that for 7.5+. CUDA 13
# removed offline compilation for pre-Turing architectures, so 61-real is
# only emitted on 12.x toolkits; Blackwell (120a/121a) needs 12.8+.
# only emitted on 12.x toolkits. Blackwell (120a) needs 12.8+; Blackwell
# Ultra (121a) needs 12.9+ (nvcc 12.8 rejects compute_121).
# Users can override with -DCMAKE_CUDA_ARCHITECTURES=native for local builds.
if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
find_package(CUDAToolkit QUIET)
if(CUDAToolkit_FOUND AND CUDAToolkit_VERSION VERSION_GREATER_EQUAL "13.0")
set(CMAKE_CUDA_ARCHITECTURES "75-virtual;80-virtual;86-real;89-real;120a-real;121a-real")
elseif(CUDAToolkit_FOUND AND CUDAToolkit_VERSION VERSION_GREATER_EQUAL "12.8")
elseif(CUDAToolkit_FOUND AND CUDAToolkit_VERSION VERSION_GREATER_EQUAL "12.9")
set(CMAKE_CUDA_ARCHITECTURES "61-real;75-virtual;80-virtual;86-real;89-real;120a-real;121a-real")
elseif(CUDAToolkit_FOUND AND CUDAToolkit_VERSION VERSION_GREATER_EQUAL "12.8")
set(CMAKE_CUDA_ARCHITECTURES "61-real;75-virtual;80-virtual;86-real;89-real;120a-real")
else()
set(CMAKE_CUDA_ARCHITECTURES "61-real;75-virtual;80-virtual;86-real;89-real")
endif()