Tesla’s stock price has skyrocketed over the last few months, when arguably, not many people are buying or even contemplating new cars. Car manufacturers are making ventilators. The question then arises if this a sane investment and if the price of the stock going to rise further. If it does, what would be the ceiling ?
Laying those (very important) questions off to the side, I want to turn your attention to something technologically more important, that has perhaps not received appropriate spotlight. While the world has been rightfully busy with a pandemic, Tesla unveiled, without too much pomp and show, a new computer.
The FSD (Full Self-Driving) Computer promises a 21x performance improvement over current hardware and at a lower cost, with the capability to process up to 2300 frames per second. This changes the game for safety and autonomy and showcases a strong technological upper hand that Tesla can now claim over rivals. The new platform can be retrofitted onto any Tesla vehicle made since October 2016, and as engineers/scientists it is not hard to see the challenging constraints on form factor and thermal characteristics one would have to overcome to achieve that.
The FSD is purpose-built for autonomy, with 72 TOPS for deep learning inference and also includes a set of CPUs and GPUs and video encoders – all with a low TDP ceiling of 40W. The system is actually equipped with two of the FSD chips [260 -sq micrometer, 6B transistors, 14-nm FinFet by Samsung] running independent OS-s for safety and redundancy.
Tesla’s boutique Neural Network Accelerator (NNA) is optimized to detect a predefined set of road-relevant objects such as lane markers, pedestrians and other vehicles at a very high frame rate. The NNA speeds up various operations such as Convolutions and Pooling (in a very non-trivial way) while keeping the power requirements low by using a 32 MB Static RAM (instead of a DRAM) to store interim weights and activations.
What I really liked was that Tesla engineers published their efforts in IEEE Micro. Granted this is not a peer reviewed paper; still, the article is quite revelatory and is very well written – from an engineering viewpoint. I guess when you are far ahead of the pack, you do not worry too much about losing competitive advantage.
Technology in business is not fully manifested until it commands additional revenue (or productivity) and in this case it seems that Tesla will not have to wait too long. It has already increased the price of the fully self-driving option by a $1000 – to $8000. Difficult to say whether the chip is directly responsible but the correlation is not hard to see.
It has probably not escaped the readers’ radar that recently Apple has announced their plan to migrate, over the next two years, to Apple Silicon chips instead of Intel. Tesla’s game is also from the same playbook. Both Apple and Tesla are owning their supply chain and cementing their commanding role in the ecosystem.
As Jack Welch would say, “Control your destiny; or someone else will”. And that deserves a surge in the stock price.
Journey of the Journeyman 