8000 CUDA: update build CTK version to 12.8 by thevishalagarwal · Pull Request #13360 · ggml-org/llama.cpp · GitHub
[go: up one dir, main page]
Skip to content

CUDA: update build CTK version to 12.8 #13360

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
wants to merge 3 commits into
base: master
Choose a base branch
from
+51 −36
Open

CUDA: update build CTK version to 12.8 #13360

Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
1307bb8
cuda: compile sm120 for ctk 12.8
thevishalagarwal May 7, 2025
c54c98f
remove whitespaces
thevishalagarwal May 7, 2025
d971e0a
update to ctk 12.8
thevishalagarwal May 12, 2025
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
Next Next commit
cuda: compile sm120 for ctk 12.8
  • Loading branch information
@thevishalagarwal
thevishalagarwal committed May 12, 2025
commit 1307bb87c12d672677be450254762b6cd407ef7c
31 changes: 23 additions & 8 deletions ggml/src/ggml-cuda/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@ if (CUDAToolkit_FOUND)
# 80 == Ampere, asynchronous data loading, faster tensor core instructions
# 86 == RTX 3000, needs CUDA v11.1
# 89 == RTX 4000, needs CUDA v11.8
# 120 == RTX 5000, needs CUDA v12.8
#
# XX-virtual == compile CUDA code as PTX, do JIT compilation to binary code on first run
# XX-real == compile CUDA code as device code for this specific architecture
Expand All @@ -23,19 +24,33 @@ if (CUDAToolkit_FOUND)
# The default behavior for a non-native is to build virtual architectures as needed to cover all features needed
# for best performance and to also build real architectures for the most commonly used GPUs.
if (GGML_NATIVE AND CUDAToolkit_VERSION VERSION_GREATER_EQUAL "11.6" AND CMAKE_VERSION VERSION_GREATER_EQUAL "3.24")
# Use the GPUs available on this system
set(CMAKE_CUDA_ARCHITECTURES "native")
elseif(GGML_CUDA_F16 OR GGML_CUDA_DMMV_F16)
if (CUDAToolkit_VERSION VERSION_GREATER_EQUAL "11.8")
set(CMAKE_CUDA_ARCHITECTURES "60-virtual;61-virtual;70-virtual;75-virtual;80-virtual;86-real;89-real")
else()
set(ARCH_LIST "")

# Base architectures - depending on feature flags
if (GGML_CUDA_F16 OR GGML_CUDA_DMMV_F16)
# FP16 support (Pascal and newer)
list(APPEND ARCH_LIST "60-virtual")
else()
set(CMAKE_CUDA_ARCHITECTURES "60-virtual;61-virtual;70-virtual;75-virtual;80-virtual;86-real")
# Maxwell and newer
list(APPEND ARCH_LIST "50-virtual")
endif()
else()

# Always included after base architecture assuming CUDA toolkit version is 11.1 or higher
list(APPEND ARCH_LIST "61-virtual" "70-virtual" "75-virtual" "80-virtual" "86-real")

# Version-dependent architectures for newer GPUs
if (CUDAToolkit_VERSION VERSION_GREATER_EQUAL "11.8")
set(CMAKE_CUDA_ARCHITECTURES "50-virtual;61-virtual;70-virtual;75-virtual;80-virtual;86-real;89-real")
else()
set(CMAKE_CUDA_ARCHITECTURES "50-virtual;61-virtual;70-virtual;75-virtual;80-virtual;86-real")
list(APPEND ARCH_LIST "89-real")
endif()

if (CUDAToolkit_VERSION VERSION_GREATER_EQUAL "12.8")
list(APPEND ARCH_LIST "120-real")
endif()

set(CMAKE_CUDA_ARCHITECTURES ${ARCH_LIST})
endif()
endif()
message(STATUS "Using CUDA architectures: ${CMAKE_CUDA_ARCHITECTURES}")
Expand Down
0