How To Do GPU Offloading in SQL Server 2017 For Parallelism

The idea of having GPU Offloading in SQL Server 2017 is quite attractive. It might be the key to all your performance woes in SQL Server. However, it is not the case.

GPU Offloading in SQL Server 2017

Currently, GPU acceleration is supported in SQL Server 2017 Machine Learning Services and ‘MicrosoftML’ rxNeuralNet packages only. A Nvidia Cuda enabled GPU is required as a prerequisite. Here are the requirements below:

1. Check to make sure your graphics card is CUDA-enabled. To find what graphic card you are using, there are two ways.
2. One is press Win+R, type “dxdiag” and check under the Display tab(s). The other is open device manager and look under display adapters. Then see if it’s on the supported list of CUDA-enabled products.
3. Install Visual Studio Community 2013.
4. Install NVidia CUDA Toolkit 6.5.
5. This step is optional. Use Visual Studio 2013 to build deviceQuery_vs2013.sln found in C:ProgramDataNVIDIA CorporationCUDA Samplesv6.51_UtilitiesdeviceQuery and bandwidthTest_vs2013.sln found in C:ProgramDataNVIDIA CorporationCUDA Samplesv6.51_UtilitiesbandwidthTest,  you need to set solution configuration to release and solution platform to x64. Once these two solutions are built successfully, open command line from C:ProgramDataNVIDIA CorporationCUDA Samplesv6.5binwin64Release and run deviceQuery.exe and bandwidthTest.exe. Make sure the results are passed.
6. Copy cublas64_65.dll, cudart64_65.dll and cusparse64_65.dll from C:Program FilesNVIDIA GPU Computing ToolkitCUDAv6.5bin to MicrosoftML’s libs directory. MicrosoftML’s libs directory can be found by calling.system.file("mxLibs/x64", package = "MicrosoftML")
7. Download NVidia cuDNN v2 Library (It’s free, but you need to register). Extract somewhere. Copy cudnn64_65.dll from the extracted files into MicrosoftML’s libs directory.

Resources:

See this blog post for more information.

Disclaimer: The Questions and Answers provided on https://gigxp.com are for general information purposes only. We make no representations or warranties of any kind, express or implied, about the completeness, accuracy, reliability, suitability or availability with respect to the website or the information, products, services, or related graphics contained on the website for any purpose.