Intel is retooling its Knowledge Heart GPU Max lineup simply weeks after the departure of Accelerated Computing Group lead Raja Koduri and the cancellation of the x86 titan’s next-gen Rialto Bridge platform.
Particularly, the silicon large is unlaunching its GPU Max 1350 accelerator card in favor of 1 catered to “totally different markets,” an Intel spokesperson instructed The Register Tuesday.
Whereas a lot of the fanfare across the unveiling of Intel’s long-awaited Ponte Vecchio-based GPU Max sequence in November centered on the 600W liquid-cooled 1550 — which can energy the US Division of Power’s Argonne Nationwide Lab’s Aurora Supercomputer — it was simply one among three playing cards introduced on the time.
A kind of playing cards was the GPU Max 1350, which was positioned as Intel’s mid-tier datacenter card, sporting a extra sedate 450W TDP and a standard air cooler. This made it extra accessible for purchasers unprepared to plumb their deployments for liquid cooling.
With all that mentioned, an air-cooled variant of the 1550 is now being provided amid a rethink concerning the 1350.
“We launched the Intel Knowledge Heart Max GPU 1550 (600W) and commenced manufacturing shipments earlier this 12 months,” Intel instructed us immediately. “We’ve got since expanded our assist by providing Intel Knowledge Heart Max GPU 1550 to incorporate air-cooled options.”
Now that the 1550 will also be had in an air-cooled model, Intel says it is “streamlining” its product choices by eradicating the 1350. “We are going to introduce the Knowledge Heart GPU Max 1450 SKU later in 2023, which has decrease I/O bandwidth for various markets and can have the ability to use air- and liquid-cooling options,” the biz instructed us.
The TDP of this new chip is not clear, although arguably the extra attention-grabbing ingredient is Intel’s give attention to different markets and lowered reminiscence bandwidth, which may recommend Intel could also be gearing as much as promote the GPUs in China.
Commerce restrictions introduced final fall barred the export of processors able to 600GBps of reminiscence bandwidth to the Center Kingdom.
Refactoring its datacenter GPU portfolio for the Chinese language market is hardly shocking given how massive a market the Center Kingdom already is for the x86 large. China accounts for a considerable chunk of Intel’s revenues, and, given the take off of AI/ML and HPC providers from Chinese language cloud suppliers like Alibaba and Baidu, Intel could also be seeking to capitalize on a bigger slice of this market.
The Register requested Intel whether or not the playing cards are aimed on the Chinese language market; we’ll let you understand if we hear something again.
Intel would hardly be the primary to go down this route, and there’s no assure that worsening commerce relations between Washington and Beijing will not spoil the plans. Final 12 months, Nvidia introduced a nerfed model of its widespread A100 accelerator known as the A800, which featured half the reminiscence and about two-thirds the bandwidth of the full-fat card with the intention to adjust to US export legislation. The chipmaker has since introduced a China-specific model of its next-gen Hopper H100 GPUs known as the H800.
“China is a large market in itself,” Daniel Newman, chief analyst at Futurum Analysis instructed The Register. “Firms like Intel having the ability to construct a product that may meet the necessities of those export management is a means of addressing the truth that there’s nonetheless a really massive marketplace for AI workloads that don’t match the factors of what regulators are most involved about.”
Nevertheless, Newman notes that Intel, which hopes to say a heft chunk of Uncle Sam’s CHIPS subsidies, must be cautious to play by the principles, and that would require tuning down elements in accordance to altering US export restrictions.
Regardless, the Intel GPU Max 1450’s days are numbered, because the producer has successfully killed off its high-end devoted GPU roadmap after Ponte Vecchio. As a substitute the microprocessor goliath plans to pursue an APU, combining CPU and GPU cores, codenamed Falcon Shores. Nevertheless, the primary chips primarily based on that structure have been delayed till 2025.
Intel’s GPU line has been on unsure footing for the reason that company restructured its Accelerated Computing Group and demoted Koduri again to chief architect final winter. Earlier this month, the corporate’s GPU unit suffered one other blow after Koduri jumped ship to pursue alternatives exterior the corporate. He additionally simply joined the board of Tenstorrent, a RISC-V-friendly AI chip designer led by semiconductor veteran Jim Keller. ®