Class TensorCoreConfiguration
- Namespace
- DotCompute.Backends.CUDA.Configuration
- Assembly
- DotCompute.Backends.CUDA.dll
Comprehensive configuration for Tensor Core operations on NVIDIA GPUs. Tensor Cores provide specialized hardware acceleration for matrix operations.
public class TensorCoreConfiguration
- Inheritance
-
TensorCoreConfiguration
- Inherited Members
Properties
DataType
Gets or sets the data type string for Tensor Core operations. Supported types include TF32, FP16, BF16, INT8, and FP8.
public string DataType { get; set; }
Property Value
Default
Gets a default configuration optimized for general use.
public static TensorCoreConfiguration Default { get; }
Property Value
EnableAutoMixedPrecision
Gets or sets whether to enable automatic mixed precision.
public bool EnableAutoMixedPrecision { get; set; }
Property Value
EnableProfiling
Gets or sets whether to enable performance profiling.
public bool EnableProfiling { get; set; }
Property Value
EnableSparsity
Gets or sets whether to enable sparsity acceleration.
public bool EnableSparsity { get; set; }
Property Value
ForAI
Gets a configuration optimized for AI/ML workloads.
public static TensorCoreConfiguration ForAI { get; }
Property Value
ForHPC
Gets a configuration optimized for HPC workloads.
public static TensorCoreConfiguration ForHPC { get; }
Property Value
MinMatrixSize
Gets or sets the minimum matrix size to use tensor cores.
public int MinMatrixSize { get; set; }
Property Value
Precision
Gets or sets the precision mode for Tensor Core operations. Options include "High" for maximum precision, "Medium" for balanced performance, and "Fast" for maximum throughput with reduced precision.
public string Precision { get; set; }
Property Value
PreferredDataType
Gets or sets the preferred data type for computations.
public DataType PreferredDataType { get; set; }
Property Value
UseTransformerEngine
Gets or sets whether to use the Transformer Engine on Hopper.
public bool UseTransformerEngine { get; set; }