As part of a strategy shift to compete with Advanced Micro Devices Inc. and Nvidia Corp., Intel Corp. disclosed a few new details on a chip for artificial intelligence (AI) computing on Monday.
During a conference on supercomputing in Germany, Intel unveiled the details of its upcoming chip on Monday，which is internally referred to as "Falcon Shores." will boast a memory capacity of 288 gigabytes and provide support for 8-bit floating point computation. These specific technical features hold significance due to the exponential growth in the size of artificial intelligence models, such as those utilized in services like ChatGPT. Consequently, businesses are actively seeking more potent chips to efficiently process these models.
The information is also among the earliest to emerge as Intel changes its approach to keep up with Nvidia, which dominates the market for AI chips, and AMD, which is anticipated to challenge Nvidia's dominance with a chip dubbed the MI300.
Contrarily, Intel has virtually no market share since its Ponte Vecchio processor, a potential rival to Nvidia, has been plagued by years of delays.
The Ponte Vecchio-based Aurora supercomputer at Argonne National Lab, whose shipments are almost complete, according to Intel, performs better than Nvidia's newest AI chip, the H100.
However, Intel's subsequent chip after Falcon Shores is expected to be released in 2025, which coincides with the probable launch of another chip by Nvidia.
The company is taking its time to rebuild the chip after abandoning its previous plan of merging graphics processing units (GPUs) with its central processing units (CPUs), according to Jeff McVeigh, corporate vice president of Intel's super compute group.
In his conversation with Reuters, Jeff McVeigh, the corporate vice president of Intel's super compute group, highlighted the difficulty in determining a single vendor that could offer the best combination of both central processing units (CPUs) and graphics processing units (GPUs). "While Intel aspires to lead in both CPU and GPU markets, it was challenging to identify a vendor that could consistently provide the optimal combination," he stated. McVeigh emphasized the advantages of offering discrete offerings, explaining that this approach allows users who are at the level of platform to have the freedom to choose the desired ratio between GPUs and CPUs, as well as select from a range of different vendors. The move towards discrete offerings reflects Intel's commitment to providing users with flexibility and choice in their computing solutions.