At the Supercomputing Conference 2023 (SC23), Intel has unveiled groundbreaking advancements in high-performance computing (HPC) and artificial intelligence (AI) acceleration.
The company showcased its leadership in performance across various workloads, integrating Intel Data Center GPU Max Series, Intel Gaudi2 AI accelerators, and Intel Xeon processors.
In collaboration with Argonne National Laboratory, Intel presented updates on the Aurora generative AI (genAI) project, highlighting the 1 trillion parameter GPT-3 LLM on the Aurora supercomputer. The unique architecture of the Max Series GPU and the Aurora system capabilities enabled the remarkable achievement. Deepak Patil, Intel’s Corporate Vice President, emphasized the company’s commitment to providing diverse technology choices to meet the evolving needs of the HPC and AI communities.
Key Highlights:
Generative AI for Science: Argonne National Laboratory shared progress on the Aurora genAI project, a collaboration aiming to create foundational AI models for diverse scientific disciplines. The models, trained on vast datasets, will support research in biology, cancer, climate science, cosmology, and materials science.
Performance Demonstrations: Intel and Argonne demonstrated the acceleration of various scientific workloads on the Aurora supercomputer, including brain connectome reconstruction, chemical process modeling, cosmology code runs, and drug-screening AI inference applications.
HPC and AI Performance: Intel presented new benchmarks and optimizations, showcasing the superior performance of Intel Max Series GPUs and CPUs in comparison to competitors. Results included improved support for large language models (LLMs) like GPT-J and LLAMA2.
Future Developments: Intel announced plans for the Intel Gaudi3 AI accelerators in 2024, building on the success of Gaudi2. The upcoming accelerator is expected to deliver increased compute power, networking bandwidth, and on-board high-bandwidth memory, addressing the growing demand for high-performance computing of LLMs.
Software Development Advances: Intel revealed features for its 2024 software development tools, enhancing open software development with oneAPI multiarchitecture programming. The tools aim to empower developers to extend AI and HPC capabilities on Intel CPUs and GPUs, offering faster performance and improved productivity.
Market Momentum: Intel highlighted its market momentum with new supercomputer deployments featuring Max Series GPU and CPU technologies. The company emphasized its commitment to AI and HPC, with ongoing projects and collaborations, including the oneAPI Center of Excellence with the Texas Advanced Computing Center (TACC).
Next-Generation GPU: Intel’s Falcon Shores, the next-generation GPU for AI and HPC, was introduced, leveraging Intel Gaudi and Intel Xe IP. With a single GPU programming interface built on oneAPI, Falcon Shores aims to provide a seamless transition for applications built on current Intel Gaudi AI accelerators and Max Series GPUs.
