ComputingIntel Unlocks Arc A-Series...

Intel Unlocks Arc A-Series GPUs for Faster AI & Deep Learning with PyTorch Support

Follow Us On

Intel has expanded the capabilities of its Arc A-Series desktop GPUs, enabling accelerated artificial intelligence (AI) and deep learning workloads through new software optimizations for the popular PyTorch machine learning library.

In the latest update to the Intel Extension for PyTorch (IPEX), the chipmaker has added official support for leveraging the Xe Matrix Extensions (XMX) AI engines found inside Arc Alchemist GPUs like the A770. This allows PyTorch models to tap into the dedicated AI hardware acceleration of the Arc GPUs.

“GPUs are well suited for large language model workloads as they excel at massive data parallelism and high memory bandwidth,” said Intel’s AI software team. “The Arc A-Series Graphics, including the Arc A770, provide capability for accelerated execution of deep learning models with up to 512 XMX Engines.”

- Advertisement -

IPEX extends PyTorch with Intel’s latest hardware optimizations, giving developers an easy way to offload AI model execution to Arc’s XMX AI cores across Windows, Linux and WSL2 environments.

Intel demonstrated the new functionality by running the Llama 2 7B and Llama 2-Chat 7B large language models (LLMs) on the Arc A770 GPU, achieving significant inference speedups over CPU-only operation.

“To run models like Llama 2 on Arc GPUs, simply import intel_extension_for_pytorch and move the model and data to the ‘xpu’ device,” the company explained in technical guidance.

The move unlocks Arc’s AI potential for a wide range of deep learning use cases like natural language processing, image recognition, and data analytics – areas traditionally dominated by Nvidia and AMD’s gaming GPUs.

- Advertisement -

While still playing catch-up in graphics performance, Intel is aggressively optimizing its Arc line to secure a foothold in the lucrative and rapidly evolving AI market segment.

As larger language models and more complex AI workloads become commonplace on PCs, having dedicated AI acceleration could give Arc GPUs an edge over Nvidia’s CUDA-dependant offerings according to some industry watchers.

Intel hopes enabling out-of-the-box PyTorch acceleration on shipping Arc hardware will help it gain ground with AI developers exploring GPU-accelerated deep learning on desktop systems.

more from the author
Latest Stories