CPU and GPU Support

Not all computation modes, methods, and parameters are supported on both CPU and GPU. Differences in CPU and GPU support are listed below.

Computation modes

For the following algorithms, only listed computation modes are supported on GPU:

Algortihm	Supported on GPU
Covariance	batch, online
Principal Component Analysis	batch

For the following algorithms, only listed methods are supported on GPU:

Algortihm	Supported on GPU
K-Means Clustering	defaultDense
Initialization	defaultDense , randomDense
Linear Regression	defaultDense
Moments of Low Order	defaultDense
Stochastic Gradient Descent Algorithm	miniBatch
Covariance	defaultDense
Principal Component Analysis	defaultDense
k-Nearest Neighbors (kNN) Classifier	Brute Force This method is not supported on CPU.
Support Vector Machine Classifier	thunder This method is not supported on CPU.
Decision Forest	hist This method is not supported on CPU.

Algortihm	Notes
Support Vector Machine Classifier	doShrinking is only supported for defaultDense method.
Density-Based Spatial Clustering of Applications with Noise	On GPU, the memorySavingMode flag can only be set to true . On GPU, the weights parameter is not supported.
Kernel Functions	On GPU, the only supported computation mode ( ComputationMode ) is matrixMatrix .
Objective Function	On GPU, only Logistic Loss and Cross-entropy Loss are supported, Mean Squared Error Algorithm is not supported. On GPU, resultsToCompute only computes value , gradient , and hessian .
Logistic Regression	penaltyL1 is not supported on GPU

In This Topic

Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.