SK hynix and Tesla are reportedly discussing $725 million SSD deal for Dojo datacenters

11 hours ago 3
From Nvidia A100 to Tesla Dojo
(Image credit: Tesla)

Tesla is in talks with SK hynix regarding purchasing enterprise-grade SSDs worth around $725 million for its AI supercomputers based on the wafer-scale Dojo processors, reports the Korea Economic Daily. If the negotiations come to fruition, this could be one of the largest single SSD purchase agreements ever. Yet, the report is unofficial, so take it with a grain of salt.

Details about the deal are scarce to put it mildly, but the Korea Economic Daily reports mentions SK hynix's Solidigm subsidiary that makes enterprise-grade storage devices. For now, Solidigm is the world's only maker of eSSDs that can store 61.44TB of data with its D5-P5336 drive designed specifically for storing large amounts of data used for AI training. The report does not directly state that the talks are about procurement of these eSSDs, though it is clear that Tesla needs vast amounts of storage and significant storage density for its AI training workloads. 

Like other companies, Tesla is training its AI models these days and spends fortunes on hardware. In late August, Tesla introduced its Cortex AI cluster featuring 50,000 Nvidia H100 GPUs and 20,000 of Tesla's Dojo AI wafer-sized processors that will be used for AI training. It is unclear which storage devices H100-based servers use (probably those that have been certified by Nvidia or rather actual server maker), but it looks like Dojo-based machines are set to use enterprise-grade SSDs from SK hynix.

TSMC

(Image credit: TSMC)

These servers with 20,000 Dojo processors inside will be used, as Tesla CEO Elon Musk puts it, to train AI that solves real-world problems. In particular, it is expected that Dojo machines will be used to train Tesla’s full self-drive (FSD) capability, one of the competitive advantages that Tesla cars have.

While reports about AI supercomputers tend to focus on AI processors and GPUs and their power consumption, storage is equally crucially important for training large AI models as slow storage can slow down the whole process, whereas power-hungry drives can increase its costs. To that end, winning a supply contract with Tesla is a big deal for SK Hynix as it can serve as proof of the superiority of its storage technologies. That is, assuming a deal actually comes to fruition.

Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.

Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.

Read Entire Article