Compute Capability
A version identifier for NVIDIA GPU architectures, composed of major and minor version numbers, which determines the CUDA features and instruction sets supported by the GPU.
Timeline (last 90 days)
When a GPU's compute capability is below the minimum version supported by the CUDA toolchain, nvcc may refuse to compile outright, or the compiled output may fail to load at runtime.
Compute Capability is a version number NVIDIA defines for its GPU architectures, expressed as "major.minor" (e.g., 8.x for Ampere, 8.9 for Ada Lovelace, 12.0 for Blackwell), encoded directly in the GPU hardware identity and read by the driver via PCI configuration space and GPU registers; it does not depend on upper-layer software or platform labels and cannot be forged by editing product descriptions or the virtualization layer (MIG partitioning only affects the visibility of memory and compute units, not the architecture version number).
Different compute capabilities correspond to different instruction sets, tensor core generations and CUDA feature support, and determine whether frameworks such as PyTorch and vLLM can enable new precision formats like FP8/FP4.
All Facts (3)
When a GPU's compute capability is below the minimum version supported by the CUDA toolchain, nvcc may refuse to compile outright, or the compiled output may fail to load at runtime.
50%UnverifiedCompute Capability is a version number NVIDIA defines for its GPU architectures, expressed as "major.minor" (e.g., 8.x for Ampere, 8.9 for Ada Lovelace, 12.0 for Blackwell), encoded directly in the GPU hardware identity and read by the driver via PCI configuration space and GPU registers; it does not depend on upper-layer software or platform labels and cannot be forged by editing product descriptions or the virtualization layer (MIG partitioning only affects the visibility of memory and compute units, not the architecture version number).
50%UnverifiedDifferent compute capabilities correspond to different instruction sets, tensor core generations and CUDA feature support, and determine whether frameworks such as PyTorch and vLLM can enable new precision formats like FP8/FP4.
50%