Earlier this morning, Tesla (TSLA) shared the most recent photos from the inside of its Autopilot/FSD AI machine learning GPU supercluster, aka Dojo.
These pictures of the supercomputer racks and the overall data center look much more futuristic than any similar system developed yet. The server racks of the Dojo cluster carry the cyber design theme from the Cybertruck (and what’s more cyber than an AI training supercluster of processors).
If we look side-by-side, even the Elon Musk-owned xAI GPU supercluster looks outdated in front of Dojo. Nvidia, SuperMicro, and xAI built this huge machine learning AI GPU supercluster in just 122 days. Dojo, on the other hand, has been years in the making.
Last year in July, Tesla CEO Elon Musk shared the following pictures of the Dojo 1 data center. The most recent pics shared by Tesla today show an even better overall design and further refinement compared to last year (2024).
Tesla practices a continuous improvement (Deming Cycle) in all of its business areas. The new pictures of the Dojo GPU cluster are a testament to this work philosophy put into work ethic at maximum practicality.
The views of the Dojo supercomputer show Tesla’s approach to minimalistic design has transitioned from its vehicles to even the FSD data halls. The clean design and management of fiber optic cables and the thermal management plumbing are at the finest you can observe at any large data center.
A single large cable can be seen going up to the roof from each supercomputer rack of the Dojo data hall. The roof of the data hall also doesn’t show an incomprehensible and massive bundle of wires.
The following photo of Elon Musk, most probably taken at the xAI Colossus supercluster, shows how messy and unintelligible these networking wires can be at a huge data center. xAI uses Colossus to train the Grok AI chatbot.
Tesla’s Dojo 1 AI machine learning supercomputer consists of Nvidia H100 GPUs in combination with the HW4/AI4 computers. These are the same AI4 computers found in Tesla vehicles built in or after January 2023.
According to Elon Musk, the current version of the Dojo supercomputer uses AI4:Nvidia in a ~1:2 ratio. Future versions of the Dojo supercluster will include the more advanced AI5 Tesla processors that are ~10X faster compared to the AI4 chips.
Tesla currently uses multiple GPU clusters and data centers for Autopilot Full Self-Driving (FSD) Vision AI data storage and machine learning. The Cortex supercluster at Giga Texas is the recently built, more robust one.
Dojo has been slowly ramping up with the number of GPUs and Tesla AI processors over the years. However, Tesla teasing these new photos ahead of the Q1 2025 Earnings Call hints that the automaker will reveal something significant in this area on the upcoming call.
Stay tuned for constant Tesla updates. Follow us on:
Google News | Flipboard | X (Twitter) | WhatsApp Channel | RSS (Feedly).
Related Tesla News
- Starship Flight 14: SpaceX performs a single-engine static fire test on Ship 41, Starbase updates (videos)
- Here’s how to win a chance to attend the Cybercab launch event, Tesla holds the date announcement
- Tesla (TSLA) gears up for Cybercab robotaxi launch, leases an 80 wireless charger location in Austin, Texas
- Tesla to reportedly unveil the next-gen Roadster in August, with SpaceX thrusters of-course
- Tesla (TSLA) launches Certified Pre-Owned Program in the US, in-warranty cheapest Model Y or Model S $25K, Model 3 $20K
- Tesla rolls out FSD v14.1 Lite after HW3 ECU overheating issues in FSD v14 Lite
 posts new pictures of the Dojo supercomputer data center. Let's compare how it is better than xAI's Nvidia Colossus supercluster for Grok.){kind=link}