Intel Gaudi logo

 

Create, Migrate, and Optimize Your AI Models with

Intel® Gaudi® Software

Access Intel® Gaudi® accelerators along with model references, libraries, containers, and tools for training and deploying GenAI and LLM.

Models

Optimize, fine-tune, debug, and profile your model to meet your performance targets.

Documentation

Access the most recent documentation or repositories on GitHub*.

Performance Data

Review training and inference model performance data on the Intel Gaudi AI accelerator.

Support

Ask questions in our forum or browse our GitHub.

The Latest from Intel Gaudi Software

 View Blog

{"limitDisplayedContent":"showAll","collectionRelationTags":{"relations":{"OR":["etm-96a2a467cd514dff85b6c153b364b0fa","etm-8ad69a00500f46bcbfde5ad9f3f1422f","etm-a9b846f3b3e24e3a88a53659a892cd29"],"AND":["etm-b21347d4d9474d5081968efe45da1418","etm-D68A7C5C-B64D-4332-8263-EF62D1A680E7"]},"featuredIds":[]},"collectionId":"833854","resultPerPage":4.0,"filters":[],"coveoRequestHardLimit":"1000","accessDetailsPagePath":"/content/www/us/en/secure/design/internal/access-details.html","collectionGuids":["etm-D68A7C5C-B64D-4332-8263-EF62D1A680E7","etm-b21347d4d9474d5081968efe45da1418"],"cardView":true,"sorting":"Newest","defaultImagesPath":"/content/dam/www/public/us/en/images/uatable/default-icons","coveoMaxResults":5000,"coveoSplitSize":8,"fpgaFacetRootPaths":"{\"fpgadevicefamily\":[\"Primary Content Tagging\",\"Intel® FPGAs\",\"Intel® Programmable Devices\"],\"quartusedition\":[\"Primary Content Tagging\",\"Intel® FPGAs\",\"Intel® Quartus Software\"],\"quartusaddon\":[\"Primary Content Tagging\",\"Intel® FPGAs\",\"Intel® Quartus Software - Add-ons\"],\"fpgaplatform\":[\"Primary Content Tagging\",\"Intel® FPGAs\",\"Intel® FPGA Platforms\"]}","newWrapperPageEnabled":true,"descendingSortingForNumericalFacetsName":"[\"Intel® Quartus® Prime Pro Edition\",\"Intel® Quartus® Prime Lite Edition\",\"Intel® Quartus® Prime Standard Edition\",\"Quartus® II Subscription Edition\",\"Quartus® II Web Edition\"]","columnsConfiguration":{"idColumn":false,"dateColumn":false,"versionColumn":false,"contentTypeColumn":false,"columnsMaxSize":0},"dynamicColumnsConfiguration":[{"name":"DynamicColumn_version","type":"version","gtv":"Version","width":135,"selected":true}],"updateCollateralMetadataEnabled":true,"relatedAssetsEnable":true,"disableExpandCollapseAll":true,"enableRelatedAssetsOnExpandAll":false,"disableBlueBanner":true,"isICS":false,"isUPE":false}

FAQs

Frequently Asked Questions

Visit the Intel® Tiber™ AI Cloud to access a node with eight Intel Gaudi 2 accelerators. Access is granted on an hourly basis for a reasonable cost. For details, see Get Access.

Use the GPU Migration Toolkit preinstalled in the Intel Gaudi software. The toolkit converts Python* code with CUDA* and other GPU-specific commands to code that Intel Gaudi accelerators can understand. The toolkit runs in real time and does not modify the original code. You need to add the GPU_migration library into the model script. The GPU Migration Toolkit User Guide details how to ensure the model is functional on Intel Gaudi accelerators.

Find fully documented and optimized models at the following GitHub* repositories: Model References and Optimum for Intel Gaudi accelerators from Hugging Face*. These repositories contain instructions on how to download the dataset and run the models.

Setup provides information on how to migrate models to the Intel Gaudi processor, as well as videos and direct links to detailed documentation. For first-time users of Intel Gaudi processors, refer to the Intel Tiber AI Cloud Quick Start Guide.

Start with the Optimum for Intel Gaudi accelerators library. This is a dedicated library that allows all Hugging Face Transformer- and Diffusion-based models to run on Intel Gaudi accelerators. To get started, see Using Hugging Face.

We recommend that most users run the Docker* image for PyTorch* from Intel as this contains all the Intel Gaudi software, drivers, and libraries needed to run models successfully. To pull and run Docker images for Intel Gaudi accelerators, refer to the Docker installation instructions.

Performance numbers for training and inference, including the latest MLPerf* performance numbers, can be found on Model Performance Data.

Visit the Intel Gaudi AI Accelerator Developer Forum to post questions and see responses from Intel team members and the broader community.

To identify the most current version of software, see the Release Notes. To find the current version of Intel Gaudi software on your platform, see the Software Stack Verification section in the Intel Gaudi software documentation.

Yes. Use the Tensor Processor Core (TPC*) software development kit for Intel Gaudi accelerators to write kernels. This is a TPC-C-based kernel library, so custom CUDA kernels need to be converted to TPC kernels for Intel Gaudi accelerators. For more information, see TPC Programming.

If you have questions not answered here, post them in the Intel Gaudi AI Accelerator Developer Forum. For specific questions related to using PyTorch* models on GitHub*, post in the Issues tab of the Model-References GitHub repository.

Stay Informed


Register for the latest Intel Gaudi AI accelerator developer news, events, training, and updates.

Sign Up