Enum CudaArchitecture

Namespace

DotCompute.Backends.CUDA.Types

Assembly

DotCompute.Backends.CUDA.dll

Represents NVIDIA GPU architectures for optimization targeting. Each architecture has specific capabilities and optimization strategies.

public enum CudaArchitecture

Fields

Ada = 2

Ada Lovelace architecture (SM 8.9). Third-generation RT cores, fourth-generation Tensor cores, and improved efficiency. Examples: RTX 4060, RTX 4070, RTX 4080, RTX 4090, RTX 2000 Ada, L40.

Ampere = 1

Ampere architecture (SM 8.0, 8.6). Enhanced Tensor cores, improved RT cores, and better async compute. Examples: RTX 3060, RTX 3070, RTX 3080, RTX 3090, A100, A40.

Hopper = 3

Hopper architecture (SM 9.0). Designed for datacenter, featuring Transformer Engine and improved multi-instance GPU. Examples: H100, H200.

Turing = 0

Turing architecture (SM 7.5). Features include RT cores, Tensor cores, and improved integer operations. Examples: RTX 2060, RTX 2070, RTX 2080, T4.