Table of Contents

Class AcceleratorInfo

Namespace
DotCompute.Abstractions
Assembly
DotCompute.Abstractions.dll

Represents information about an accelerator device.

public class AcceleratorInfo
Inheritance
AcceleratorInfo
Inherited Members

Constructors

AcceleratorInfo()

Parameterless constructor for test compatibility. Creates a default AcceleratorInfo for testing.

[SetsRequiredMembers]
public AcceleratorInfo()

AcceleratorInfo(AcceleratorType, string, string, long)

Legacy constructor for backward compatibility with tests.

[SetsRequiredMembers]
public AcceleratorInfo(AcceleratorType type, string name, string driverVersion, long memorySize)

Parameters

type AcceleratorType
name string
driverVersion string
memorySize long

AcceleratorInfo(AcceleratorType, string, string, long, int, int, Version?, long, bool)

Extended constructor for tests.

[SetsRequiredMembers]
public AcceleratorInfo(AcceleratorType type, string name, string driverVersion, long memorySize, int computeUnits, int maxClockFrequency, Version? computeCapability, long maxSharedMemoryPerBlock, bool isUnifiedMemory)

Parameters

type AcceleratorType
name string
driverVersion string
memorySize long
computeUnits int
maxClockFrequency int
computeCapability Version
maxSharedMemoryPerBlock long
isUnifiedMemory bool

AcceleratorInfo(string, string, string, AcceleratorType, double, int, int, long, long)

Constructor for tests with full parameters.

[SetsRequiredMembers]
public AcceleratorInfo(string name, string vendor, string driverVersion, AcceleratorType type, double computeCapability, int maxThreadsPerBlock, int maxSharedMemory, long totalMemory, long availableMemory)

Parameters

name string
vendor string
driverVersion string
type AcceleratorType
computeCapability double
maxThreadsPerBlock int
maxSharedMemory int
totalMemory long
availableMemory long

Properties

Architecture

Gets the GPU architecture name (e.g., "Ampere", "RDNA2", "Xe", "Apple M1"). This is derived from the Capabilities dictionary or vendor-specific information.

public string Architecture { get; init; }

Property Value

string

AvailableMemory

Gets or sets the available device memory in bytes.

public long AvailableMemory { get; init; }

Property Value

long

Capabilities

Gets or sets device-specific capabilities.

public Dictionary<string, object>? Capabilities { get; init; }

Property Value

Dictionary<string, object>

ComputeCapability

Gets or sets the compute capability or version.

public Version? ComputeCapability { get; init; }

Property Value

Version

ComputeCapabilityVersionSafe

Gets the compute capability as a non-nullable version for test compatibility. Returns a default version (6.0) if ComputeCapability is null.

public Version ComputeCapabilityVersionSafe { get; }

Property Value

Version

ComputeUnits

Gets or sets the number of compute units.

public int ComputeUnits { get; init; }

Property Value

int

DeviceIndex

Gets or sets the device index (for multi-device systems).

public int DeviceIndex { get; init; }

Property Value

int

DeviceType

Gets or sets the device type (e.g., "CPU", "GPU", "TPU").

public required string DeviceType { get; init; }

Property Value

string

DriverVersion

Gets or sets the driver version.

public string? DriverVersion { get; init; }

Property Value

string

Extensions

Gets the collection of backend-specific extensions supported by this accelerator. Examples: OpenCL extensions, CUDA PTX features, Metal feature sets.

public IReadOnlyCollection<string> Extensions { get; init; }

Property Value

IReadOnlyCollection<string>

Features

Gets the collection of hardware features supported by this accelerator. This provides a strongly-typed view of accelerator capabilities.

public IReadOnlyCollection<AcceleratorFeature> Features { get; init; }

Property Value

IReadOnlyCollection<AcceleratorFeature>

GlobalMemorySize

Gets or sets the total global memory size in bytes. This represents the total amount of memory accessible by compute kernels.

public long GlobalMemorySize { get; init; }

Property Value

long

Id

Gets or sets the unique identifier for this device.

public required string Id { get; init; }

Property Value

string

IsUnifiedMemory

Gets or sets whether the device uses unified memory.

public bool IsUnifiedMemory { get; init; }

Property Value

bool

LanguageVersion

Gets the Metal Shading Language version supported by this device. Example: "2.4", "3.0", "3.1"

public string LanguageVersion { get; }

Property Value

string

LocalMemorySize

Gets or sets the local memory size in bytes.

public long LocalMemorySize { get; init; }

Property Value

long

MajorVersion

Gets the compute capability major version. For CUDA: The major version of compute capability (e.g., 8 for CC 8.6). For other backends: Derived from ComputeCapability or default to 0.

public int MajorVersion { get; }

Property Value

int

MaxClockFrequency

Gets or sets the maximum clock frequency in MHz.

public int MaxClockFrequency { get; init; }

Property Value

int

MaxComputeUnits

Gets or sets the maximum number of compute units available on this accelerator. For GPUs, this typically represents streaming multiprocessors or compute units. For CPUs, this represents the number of CPU cores or logical processors.

public int MaxComputeUnits { get; init; }

Property Value

int

MaxMemoryAllocationSize

Gets or sets the maximum memory allocation size in bytes.

public long MaxMemoryAllocationSize { get; init; }

Property Value

long

MaxSharedMemoryPerBlock

Gets or sets the maximum shared memory per block in bytes.

public long MaxSharedMemoryPerBlock { get; init; }

Property Value

long

MaxThreadExecutionWidth

Gets the maximum thread execution width for Metal GPUs. This represents the SIMD width of the GPU (typically 32 for Apple GPUs).

public int MaxThreadExecutionWidth { get; }

Property Value

int

MaxThreadsPerBlock

Gets or sets the maximum threads per block.

public int MaxThreadsPerBlock { get; init; }

Property Value

int

MaxThreadsPerThreadgroup

Gets the maximum threads per threadgroup for Metal GPUs. This is equivalent to MaxThreadsPerBlock but uses Metal terminology.

public int MaxThreadsPerThreadgroup { get; }

Property Value

int

MaxWorkGroupSize

Gets or sets the maximum work group size (alias for MaxThreadsPerBlock).

public int MaxWorkGroupSize { get; init; }

Property Value

int

MaxWorkItemDimensions

Gets the maximum work-item dimensions supported by this accelerator. This is typically 3 (x, y, z) for most modern GPUs.

public int MaxWorkItemDimensions { get; init; }

Property Value

int

Remarks

This determines how many dimensions you can use when launching kernels. Most GPU compute APIs support 3D work-item grids (x, y, z).

MemorySize

Gets or sets the total device memory in bytes (alias for TotalMemory).

public long MemorySize { get; }

Property Value

long

MinorVersion

Gets the compute capability minor version. For CUDA: The minor version of compute capability (e.g., 6 for CC 8.6). For other backends: Derived from ComputeCapability or default to 0.

public int MinorVersion { get; }

Property Value

int

Name

Gets or sets the friendly name of this device.

public required string Name { get; init; }

Property Value

string

RecommendedMaxWorkingSetSize

Gets the recommended maximum working set size for Metal unified memory. This represents the optimal amount of memory to use for best performance.

public ulong RecommendedMaxWorkingSetSize { get; }

Property Value

ulong

SupportsFloat64

Gets or sets whether this accelerator supports double-precision (64-bit) floating-point operations. This is important for scientific computing applications requiring high precision.

public bool SupportsFloat64 { get; init; }

Property Value

bool

SupportsInt64

Gets or sets whether this accelerator supports 64-bit integer operations. This affects the availability of long and ulong data types in compute kernels.

public bool SupportsInt64 { get; init; }

Property Value

bool

TotalMemory

Gets or sets the total device memory in bytes.

public long TotalMemory { get; init; }

Property Value

long

Type

Gets or sets the device type for legacy compatibility.

public string Type { get; }

Property Value

string

Vendor

Gets or sets the vendor name.

public required string Vendor { get; init; }

Property Value

string

WarpSize

Gets the warp/wavefront size for this accelerator.

  • NVIDIA CUDA: 32 threads
  • AMD ROCm: 64 threads (wavefront)
  • Intel: 8-32 threads (SIMD width)
  • CPU: 1 (no warp concept)
public int WarpSize { get; init; }

Property Value

int

Remarks

This is critical for optimizing memory access patterns and thread synchronization. Applications should align thread block sizes to multiples of WarpSize for optimal performance.

Edit this page