Intel Reveals Future AI Strategy and GPU Advancements at Hamburg Exhibition

Hamburg, Germany - In a high-performance computing exhibition held in Hamburg, Germany, on Monday, Intel, currently undergoing a strategic transformation, unveiled the latest details of its future AI computing strategy. Since reentering the independent GPU market, Intel's GPUs have been continuously gaining momentum.

Intel's GPUs have been making rapid advancements since their reentry into the independent GPU market. Addressing the market's most anticipated next-generation Max series GPU chip, Falcon Shores, Intel's preliminary specifications indicate that it will feature high-bandwidth memory (HBM3) with a capacity of 288GB, supporting 8-bit floating-point operations, and offering a total bandwidth of up to 9.8TB/s. The demand for more powerful chips is insatiable, especially for artificial intelligence models like ChatGPT, which continue to grow in size. By comparison, Intel's earlier Max 1550 chip, released earlier this year, had HBM2e memory specifications of 128G.

Intel stated that the company is nearing completion of the delivery of its Aurora supercomputer based on Ponte Vecchio to the Argonne National Laboratory in the United States. Intel also emphasized that the performance of this system will surpass that of the HPC using Nvidia's H100 chip.

The Argonne National Laboratory, which made significant contributions to the "Manhattan Project," also announced its plans to utilize this supercomputer to develop a series of generative AI models for the scientific research community. The complete Aurora system consists of 63,744 GPUs, 21,248 CPUs, and 1,024 DAOS storage nodes. Furthermore, the Cambridge Open Zettascale Laboratory at the University of Cambridge has deployed the first Max GPU testbed in the UK and has achieved positive early results in molecular dynamics and bioimaging applications.

Additionally, despite the strengthened export control of advanced chips to China by the United States, American companies are unwilling to give up the Chinese market, which accounts for 60% of global chip procurement. Intel is expected to follow Nvidia's lead and introduce chips specifically designed for the Chinese market. On April 12th, Intel spokesperson confirmed on Tuesday that the data center GPU series will be adjusted, with the removal of the Max GPU 1350 and the expected launch of the Max GPU 1450 later this year, meeting the requirements of "different markets" by reducing I/O bandwidth.

Jeff McVeigh, Vice President and General Manager of the Supercomputing Business Unit at Intel, stated on Monday that the company is investing time in redesigning chips after abandoning the strategy of combining CPUs and GPUs. While the company aspires to have the most powerful CPU and GPU on the market, it is challenging to achieve the optimal combination of both from a single supplier. Conversely, having independent products allows for competition at the platform level.