3 ms·
All Teslas need to do is run inference. I'm fairly certain training is all done either in the cloud or on datacenter GPUs. It would be very interesting if Goo
by illumin8 8y ago
All Teslas need to do is run inference. I'm fairly certain training is all done either in the cloud or on datacenter GPUs.
It would be very interesting if Google was selling Tesla EOL TPU v1 units. The specs are very good for what Tesla is trying to do.
4 years is not a long time considering this is a proprietary technology that even GCP customers haven't had access to for very long. It will likely be a few more years before any TPU-like technology is available in the mass market.