Skip to content

feat: Add SM121/GB10 (DGX Spark) Blackwell-class GPU support #31740

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
seli-equinix wants to merge 13 commits into
base: main
Choose a base branch
from
Open

feat: Add SM121/GB10 (DGX Spark) Blackwell-class GPU support #31740

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
13 commits
Select commit Hold shift + click to select a range
4c5443c
feat: Add SM121/GB10 (DGX Spark) Blackwell-class GPU support
seli-equinix Jan 4, 2026
3515f5e
fix: Handle empty/whitespace tool_call arguments in chat preprocessing
seli-equinix Jan 28, 2026
75d6d43
feat: Auto-configure TRITON_PTXAS_PATH for SM121/GB10
seli-equinix Jan 28, 2026
ef871fb
fix: Add SM121 (12.1) arch support for CUTLASS SM120 kernels
seli-equinix Jan 28, 2026
4c49c64
feat: Add NVIDIA GB10 MoE tuning config for FP8
seli-equinix Jan 28, 2026
3e1f5e4
fix: Address Copilot review comments for SM121/GB10 support
seli-equinix Feb 3, 2026
0d37a9f
Add metrics visualization with plot plugin
seli-equinix Feb 3, 2026
1e1090a
fix: Allow FP8 block-scale MoE on SM121 via FlashInfer CUTLASS backend
seli-equinix Feb 16, 2026
28c10db
fix: Split has_flashinfer_nvfp4 from has_flashinfer_cutlass_fused_moe
seli-equinix Feb 17, 2026
ff8f796
fix: Handle raw torch.fx.Node references in AOT compile serialization
seli-equinix Feb 18, 2026
061c2fe
fix: GB10 nogds + add Qwen3.5 MoE tuning config (E=256,N=512)
Mar 5, 2026
57b86c3
fix: Prefer FlashInfer for ViT attention on SM121/GB10 (Flash Attenti…
Mar 11, 2026
1a629ce
fix: Add SM121 to CUDA_SUPPORTED_ARCHS (was filtered out by arch inte…
Mar 11, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 11 additions & 9 deletions CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -94,10 +94,10 @@ find_package(Torch REQUIRED)
# This check must happen after find_package(Torch) because that's when CMAKE_CUDA_COMPILER_VERSION gets defined
if(DEFINED CMAKE_CUDA_COMPILER_VERSION AND
CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 13.0)
set(CUDA_SUPPORTED_ARCHS "7.5;8.0;8.6;8.7;8.9;9.0;10.0;11.0;12.0")
set(CUDA_SUPPORTED_ARCHS "7.5;8.0;8.6;8.7;8.9;9.0;10.0;11.0;12.0;12.1")
elseif(DEFINED CMAKE_CUDA_COMPILER_VERSION AND
CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12.8)
set(CUDA_SUPPORTED_ARCHS "7.0;7.2;7.5;8.0;8.6;8.7;8.9;9.0;10.0;10.1;12.0")
set(CUDA_SUPPORTED_ARCHS "7.0;7.2;7.5;8.0;8.6;8.7;8.9;9.0;10.0;10.1;12.0;12.1")
else()
set(CUDA_SUPPORTED_ARCHS "7.0;7.2;7.5;8.0;8.6;8.7;8.9;9.0")
endif()
Expand Down Expand Up @@ -530,9 +530,9 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
# The cutlass_scaled_mm kernels for Geforce Blackwell SM120 (c3x, i.e. CUTLASS 3.x) require
# CUDA 12.8 or later
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
cuda_archs_loose_intersection(SCALED_MM_ARCHS "12.0f" "${CUDA_ARCHS}")
cuda_archs_loose_intersection(SCALED_MM_ARCHS "12.0f;12.1f" "${CUDA_ARCHS}")
else()
cuda_archs_loose_intersection(SCALED_MM_ARCHS "12.0a" "${CUDA_ARCHS}")
cuda_archs_loose_intersection(SCALED_MM_ARCHS "12.0a;12.1a" "${CUDA_ARCHS}")
endif()
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND SCALED_MM_ARCHS)
set(SRCS
Expand Down Expand Up @@ -648,9 +648,9 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
# The nvfp4_scaled_mm_sm120 kernels for Geforce Blackwell SM120 require
# CUDA 12.8 or later
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
cuda_archs_loose_intersection(FP4_ARCHS "12.0f" "${CUDA_ARCHS}")
cuda_archs_loose_intersection(FP4_ARCHS "12.0f;12.1f" "${CUDA_ARCHS}")
else()
cuda_archs_loose_intersection(FP4_ARCHS "12.0a" "${CUDA_ARCHS}")
cuda_archs_loose_intersection(FP4_ARCHS "12.0a;12.1a" "${CUDA_ARCHS}")
endif()
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND FP4_ARCHS)
set(SRCS
Expand Down Expand Up @@ -699,10 +699,12 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
endif()

# CUTLASS MLA Archs and flags
# Note: CUTLASS MLA only supports SM100/SM103, NOT SM12x (GB10/SM121)
# SM12x devices use TRITON_MLA instead
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
cuda_archs_loose_intersection(MLA_ARCHS "10.0f;11.0f;12.0f" "${CUDA_ARCHS}")
cuda_archs_loose_intersection(MLA_ARCHS "10.0f;10.3f;11.0f" "${CUDA_ARCHS}")
else()
cuda_archs_loose_intersection(MLA_ARCHS "10.0a;10.1a;10.3a;12.0a;12.1a" "${CUDA_ARCHS}")
cuda_archs_loose_intersection(MLA_ARCHS "10.0a;10.1a;10.3a" "${CUDA_ARCHS}")
endif()
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND MLA_ARCHS)
set(SRCS
Expand Down Expand Up @@ -818,7 +820,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")

# moe_data.cu is used by all CUTLASS MoE kernels.
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
cuda_archs_loose_intersection(CUTLASS_MOE_DATA_ARCHS "9.0a;10.0f;11.0f;12.0f" "${CUDA_ARCHS}")
cuda_archs_loose_intersection(CUTLASS_MOE_DATA_ARCHS "9.0a;10.0f;11.0f;12.0f;12.1f" "${CUDA_ARCHS}")
else()
cuda_archs_loose_intersection(CUTLASS_MOE_DATA_ARCHS "9.0a;10.0a;10.1a;10.3a;12.0a;12.1a" "${CUDA_ARCHS}")
endif()
Expand Down
Loading
Loading