Tesla Cortex Supercomputer cluster to add 100k Nvidia H100/H200 chips

Tesla is preparing its supercomputer cluster with 100k of H100 and H200 Nvidia AI chipsets for new self-driving and robotics training capacities.

Since the first quarter, the company has been working on Giga Texas expansion. The construction at the south side of the facility will house these new computing machines. Last month, Tesla CEO, Elon Musk shared a visual of the new supercomputer cluster and the work in progress to complete the project.

The EV maker is a big H100 buyer and deployed more than 35000 of these chips for self-driving training. However, it planned to add up to 50,000 to expand the training capability. Musk gave an update on the Giga supercomputer cluster on Saturday and inspected the progress. He also named the cluster “Cortex”.

This new cluster contains 100,000 Nvidia H100 and H200 chipsets tagged with large storage for training purposes. Tesla is weighing big on its two future businesses including Full Self Driving (FSD) and Optimus robot.

Tesla FSD (image credit: Tesla)

Both of these require high computing power. Therefore, it is ramping up computing power and support these two segments. Once online, the cluster will enable FSD and Optimus architecture to add new data and speed up training capabilities. Also, major improvements could become a reality when this cluster goes online.

A previous report confirms that the facility will use 130MW of power consumption and cooling for this supercomputer. Tesla will increase 500MW over the next 18 months. Furthermore, new fans are installed to support cooling for all computing hardware.

A few weeks ago, we also saw the first-ever images of the Tesla Dojo AI rack. These are designed in-house by Tesla engineers and look different from traditional data center racks. Under the hood, it has self-designed chipsets and the other half of the Dojo will use Nvidia chips.

Tesla Dojo Cluster Pictures (Image Source – Tesla/X)

With these moves, Tesla is likely to enable this new supercomputer cluster anytime in August or the following month. However, we’ll have to wait for an official confirmation.

(source)