Enum GpuBackend

Namespace

Orleans.GpuBridge.Abstractions.Enums

Assembly

Orleans.GpuBridge.Abstractions.dll

Defines the available compute backend types for GPU and CPU processing.

public enum GpuBackend

Fields

Auto = 0

Automatically select the best available backend based on platform and device capabilities.

When Auto is selected, the system will evaluate available backends and choose the most appropriate one based on performance characteristics and platform support.

CPU = 1

CPU backend for fallback processing when GPU is not available.

The CPU backend provides a fallback option when no GPU backends are available or when CPU processing is preferred. It uses multi-threaded SIMD operations to maximize performance on CPU architectures.

CUDA = 2

NVIDIA CUDA backend for high-performance GPU computing.

CUDA provides optimal performance on NVIDIA GPUs and is available on Windows and Linux platforms. Requires CUDA runtime libraries to be installed.

DirectCompute = 4

Microsoft DirectCompute backend for Windows GPU computing.

DirectCompute is available on Windows platforms through DirectX 11 and later. It provides good integration with Windows graphics APIs and supports most modern GPUs on Windows.

Metal = 5

Apple Metal backend for macOS GPU computing.

Metal provides optimized performance on macOS with Apple's graphics hardware. It offers low-level access to GPU resources and is the preferred backend for Apple platforms.

OpenCL = 3

OpenCL backend for cross-platform GPU and CPU computing.

OpenCL provides broad compatibility across different GPU vendors (NVIDIA, AMD, Intel) and platforms (Windows, Linux, macOS). Performance may vary depending on the specific OpenCL implementation.

Vulkan = 6

Khronos Vulkan backend for modern cross-platform GPU computing.

Vulkan provides low-level GPU access with minimal driver overhead. It supports Windows, Linux, and Android platforms with modern GPUs.

Remarks

This enumeration specifies which compute backend should be used for executing GPU kernels. Each backend corresponds to a different compute API or platform. The device manager uses this information to select the appropriate backend based on platform capabilities and user preferences.