llama.cpp binaries compiled specifically for the NVIDIA GB10 (DGX Spark)
Blackwell GPU. These are provided because upstream llama.cpp ships no aarch64 + CUDA
prebuilt binaries — on a Spark you normally have to build from source.[!IMPORTANT] These binaries only run on a machine matching the build target below (aarch64 + GB10/sm_121 + CUDA 13.x runtime). On any other GPU/arch they fail withno kernel image is available for execution on the device.
| llama.cpp version | b9860 (commit fdb1db877) |
| GPU target | CMAKE_CUDA_ARCHITECTURES=121a-real (GB10, sm_121a, native FP4) |
| Platform | Linux aarch64 (ARM Grace) |
| Compiler | GCC 13.3.0 |
| CUDA | 13.x toolkit |
1cmake -B build -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=121a-real -DCMAKE_BUILD_TYPE=Release
2cmake --build build --config Release -j$(nproc)1hf download merve/llama.cpp-dgx-spark-gb10-sm121a --local-dir ./llama-gb10
2cd llama-gb10
3chmod +x bin/llama-*LD_LIBRARY_PATH entry for /usr/local/cuda-13/compat is required
so the Blackwell compat driver libs are found at runtime):1export LD_LIBRARY_PATH="$PWD/bin:/usr/local/cuda-13/compat:$LD_LIBRARY_PATH"
2./bin/llama-server -hf unsloth/NVIDIA-Nemotron-3-Nano-4B-GGUF --port 8081 --n-gpu-layers 99./bin/llama-server --version # version: 9860 (fdb1db877)compute capability 12.1 in the server log.llama-* tools (llama-server, llama-cli, llama-bench,
llama-quantize, llama-embedding, llama-mtmd-cli, ...) plus the required shared
libraries (libggml*.so*, libllama*.so*) under bin/. Test binaries are not included.LICENSE.
This is an unofficial community build and is not affiliated with the llama.cpp project
or NVIDIA.