Intel Boosts AI Workloads with Meta Llama 3 Models

Intel has announced the validation of its AI product lineup for Meta Llama 3 8B and 70B models, incorporating Intel® Gaudi® accelerators, Intel® Xeon® processors, Intel® Core™ Ultra processors, and Intel® Arc™ graphics.

In its commitment to democratizing AI, Intel continues to invest in software and AI ecosystems to ensure its products remain at the forefront of innovations in the dynamic AI landscape. Within data centers, Intel’s Gaudi and Xeon processors, equipped with Intel® Advanced Matrix Extension (Intel® AMX) acceleration, offer customers versatile solutions to meet diverse and evolving requirements.

Intel Core Ultra processors and Intel Arc graphics products serve as essential tools for local development and deployment across a multitude of devices, boasting comprehensive support for software frameworks and tools such as PyTorch and Intel® Extension for PyTorch®, essential for local research and development, as well as the OpenVINO™ toolkit for model development and inference.

Intel has conducted preliminary testing and performance assessments for the Llama 3 8B and 70B models, leveraging open-source software including PyTorch, DeepSpeed, Intel Optimum Habana library, and Intel Extension for PyTorch to ensure the latest in software optimizations.

For demanding end-to-end AI workloads, Intel Xeon processors offer optimal performance, with Intel continuously refining results to minimize latency. The Intel® Xeon® 6 processors featuring Performance-cores (code-named Granite Rapids) exhibit a 2x improvement in Llama 3 8B inference latency compared to 4th Gen Intel® Xeon® processors, enabling the efficient execution of larger language models, like Llama 3 70B, in under 100ms per generated token.

Intel Core Ultra processors and Intel Arc Graphics demonstrate impressive performance for Llama 3 models. In initial tests, Intel Core Ultra processors surpass typical human reading speeds, while the Intel® Arc™ A770 GPU, equipped with Xe Matrix eXtensions (XMX) AI acceleration and 16GB of dedicated memory, delivers exceptional performance for LLM workloads.

ASBIS, the official distributor of Intel products, is a leading Value-Added Distributor, developer, and supplier of IT and IoT products across EMEA.