Tesla's in-house supercomputer received an additional 1600 GPUs, a 28% increase from the figure cited a year ago.
Tesla engineering director Tim Zaman says this would put the machine seventh in the world for GPU count.
The machine now packs a total of 7,360 Nvidia A100 GPUs, which are designed specifically for data center servers, but use the same architecture as the company's higher-end GeForce RTX 30-series cards.
Telsa supercomputer upgrade
Tesla likely needs all the processing power it can get right now. The company is currently working on "neural networks," which are used to process the vast amounts of video data collected by company cars.
The latest update may just be the start of Tesla's high-performance computing (HPC) ambitions.
In June 2020, Elon Musk said that "Tesla is developing a neural network training computer called Dojo to process large amounts of video data", explaining that the planned machine would achieve performance of more than 1 exaFLOP, which represents a quintillion floating point. operations per second, or 1.000 petaFLOPs.
Performance of more than 1 exaFLOP would put the machine among the most powerful supercomputers in the world, as only a few current supercomputers have officially broken the exascale barrier, including The Frontier supercomputer at Oak Ridge National Laboratory in Tennessee, USA. USA
You might even get a job building the new computer. Musk asked his Twitter followers to "consider joining our AI or computer/chip teams if that sounds interesting."
However, Dojo will not be dependent on Nvidia hardware. The planned machine is expected to be powered by Tesla's new D1 Dojo chip, which the automaker says could have specs of up to 362 TFLOPs at its AI Day event.
- Do you want to conduct your own AI research in the cloud? Check out our guide to the best cloud hosting
Via Tom's Hardware(Opens in a new tab)