Skip to main content

software catalog

Intel® D-Series GPU

AI on Intel® D-Series GPU

Scale inference, not infrastructure

Explore the Blueprints

Run it today on Intel® D-Series GPU

The same upstream stack runs these blueprints on Intel® D-Series GPU and on Intel® Arc™ Pro B-series. Pick one, run it, then replicate its numbers on your cards.

Generative AI

Inference on the frameworks you already run

Run generative AI workloads on Intel GPUs using upstream PyTorch and leading LLM serving frameworks like vLLM and SGLang, alongside Intel® Inference Microservices for simplified model deployment and serving. Keep standard model formats and OpenAI-compatible APIs while taking advantage of Intel GPU-optimized kernels and libraries.

  • Upstream first

    Use community framework releases with Intel GPU support directly upstream, with no Intel-specific framework fork required.

  • Broad generative AI workloads

    Run language, reasoning, multimodal, speech, image and video workloads through the open AI-software ecosystem.

  • Leading LLM serving frameworks

    Run vLLM and SGLang directly on Intel GPUs, providing customers access to the leading open-source LLM serving stacks.

  • Right sized

    Start with a single GPU and scale to multi-GPU or distributed inference as model size and workload demand increase.

  • Efficient

    Built for tokens per watt, not just raw throughput.

Agentic AI

Run Open Agent Frameworks on Intel

Build and deploy agentic applications on Intel hardware using open agent frameworks, standard inference APIs, and the models and tools your applications already use.

  1. Open agent frameworks

    Run open agent harnesses such as OpenClaw while keeping the agent layer independent from the underlying inference infrastructure.

  2. Leading inference and serving software

    Run agent workloads using vLLM, SGLang, Dynamo and llm-d on Intel GPUs.

  3. Complete agentic workflows

    Combine model inference with retrieval, memory, tools, MCP servers, databases, APIs and enterprise applications.

  4. Heterogeneous execution

    Use Intel GPUs for model-intensive workloads and Intel Xeon processors for CPU-oriented stages where appropriate.

Intel® Agent Toolkit

Documentation

Each side sized to its stages

Trademark notice

Intel, the Intel logo, Xeon, Arc, Xe, VTune and oneAPI are trademarks of Intel Corporation or its subsidiaries. SYCL, OpenCL and SPIR-V are trademarks of the Khronos Group. CUDA, CUTLASS, CuTe, Nsight, NCCL, cuDNN, cuBLAS, DCGM and Dynamo are trademarks or product names of NVIDIA Corporation. PyTorch, Triton and vLLM are projects of the Linux Foundation or their respective maintainers. Linux is a registered trademark of Linus Torvalds. Kubernetes is a registered trademark of the Linux Foundation. OpenShift is a trademark of Red Hat, Inc. Other names and brands may be claimed as the property of others.