Intel reportedly preparing region-specific datacenter GPU variants for China market.

Intel is retooling its Data Center GPU Max lineup by discontinuing the GPU Max 1350 and introducing a new GPU Max 1450 SKU later in 2023. The move comes weeks after the departure of Accelerated Computing Group lead Raja Koduri and follows the cancellation of the company's next-gen Rialto Bridge platform. The original GPU Max series, unveiled in November and based on Ponte Vecchio architecture, consisted of three cards: the flagship GPU Max 1550, a 600W liquid-cooled accelerator designed to power the US Department of Energy's Argonne National Lab's Aurora Supercomputer; the GPU Max 1350, a mid-tier 450W air-cooled card; and a third offering. Intel has now expanded support by offering an air-cooled variant of the 1550, prompting the company to "streamline" its product offerings by removing the 1350.

The new Data Center GPU Max 1450 represents a significant shift in Intel's strategy. According to the company, it "has lower I/O bandwidth for different markets and will be able to use air- and liquid-cooling solutions." The emphasis on reduced I/O bandwidth and different markets strongly suggests Intel is preparing to sell the GPUs in China, following US trade restrictions announced last fall that barred the export of processors capable of 600GBps of I/O bandwidth to the country. China represents a substantial market for Intel, accounting for significant revenue, and the growth of AI/ML and HPC services from Chinese cloud providers like Alibaba and Baidu presents a major opportunity for the company.

Intel's strategy mirrors Nvidia's approach to export compliance. Last year, Nvidia announced the A800, a nerfed version of its popular A100 accelerator featuring half the memory and about two-thirds the bandwidth to comply with US export law, and subsequently announced a China-specific version of its next-gen Hopper H100 GPUs called the H800. Daniel Newman, chief analyst at Futurum Research, explained to The Register that "Companies like Intel being able to build a product that can meet the requirements of these export control is a way of addressing the fact that there's still a very large market for AI workloads that don't fit the criteria of what regulators are most concerned about." However, Newman cautioned that Intel, which hopes to claim a substantial chunk of US CHIPS subsidies, must carefully play by the rules and potentially tune down parts in accordance with changing export restrictions.

Intel's GPU roadmap has faced significant uncertainty following recent leadership changes. The company has effectively killed off its high-end dedicated GPU roadmap after Ponte Vecchio, instead pursuing an APU combining CPU and GPU cores codenamed Falcon Shores—though first chips based on that architecture have been delayed until 2025. The GPU line's troubled trajectory accelerated after Intel demoted Raja Koduri to chief architect last winter during a corporate restructuring of the Accelerated Computing Group. Earlier this month, Koduri departed Intel entirely to pursue opportunities outside the company and joined the board of Tenstorrent, a RISC-V-friendly AI chip designer led by semiconductor veteran Jim Keller.