By Jim Ringold
The head coach and referee combined is the computer central processor (CPU) of an electric vehicle (EV) such as a Tesla. As we covered in Parts 1 and 2, it took the steady process of inventions and improvements in “solid state” electronic components to make the EV even possible. Previous electric cars needed mechanical switches and rheostats (a veritable resistor to provide variable current flow) to speed up and slow down an electrical vehicle. A rheostat was a very inefficient way to control speed gobbling up the battery amps. This technology advancement is doubly true of the central computer that controls all in a Tesla.
The Tesla Model S, introduced in June of 2012, was the first “clean sheet of paper” car from Tesla; a car using Elon’s “First Principles” design philosophy. This was starting from scratch and putting little weight on what had been done previously. Sometimes the current state of technology is the best solution, as in the case of disk brakes. But, usually, the advancement in car component design is based on what was used previously; can the current tooling make it; utilize the current vendor; and is it cheap to make? Tesla had an advantage in that nothing came previously. Vendors were not in place and it didn’t have to be the cheapest solution, only the best solution. So let’s skip to the details of the central processor of the Model S.
The CPU of choice turned out to be made by NVIDIA. We can safely assume that of all the CPUs out there, this was the best solution available. And it met the need, successfully managing the Model S’s systems — the Tegra 3 3D Visual Computing Module (VCM) for the main, vertical, iPad-sized center screen; and the Targa 2 VCM for the devices instrument display. This Targa System on a Chip (SoC) contained eight specialized processors. The Model X utilizes the same components.
This NVIDIA GPU uses a 16 nanometer (NM) size production process. (This will be referenced later in the description of Tesla’s propriety GPU.) It is physically considered a medium–sized processor chip that has a transistor count of 4,400 million. We have come a long way from the discrete (single) transistor of the 1950s. NVIDIA started as a producer of a video graphics processor (GPU) for PC computers with fast and high-resolution products. Video game players loved them. The graphics abilities for Tesla’s SoC are needed because of video camera inputs, graphics dashboard displays, and also audio system control.
In March of 2018, Tesla announced a Media Control Unit (MCU) upgrade version called 2.0. An older Intel Processor was reintroduced as the Atom CPU was adopted for graphics. Tesla utilized this Atom processor to display the faster, improved center display graphics. This upgrade allowed Tesla Theater and Arcade to be displayed.
In the Model S refresh in 2021, the center touchscreen was turned horizontal, similar but larger than the horizontal screen introduced in the Model 3 and, subsequently, the Model Y. But back to the CPU/GPU.
As Tesla realized that the “Autopilot” brand of driver assist was going to need additional computer power to successfully obtain the goal of “Full Safe Driving,” a more powerful CPU/GPU would be required. Since Tesla was utilizing what was at the time the best NVIDIA had to offer, Elon did what he has consistently done, bring the CPU/GPU design in house. The result was Hardware Version 3 (HW3). (For the record, there was an interim Hardware Version HW2.5 that filled the gap until HW3 was ready.)
Elon Musk hired Jim Keller as head of Autopilot Hardware and used his deep expertise in CPU design. He was a chip designer of legendary portions that had designed Apple and AMD CPUs. An AMD computer chip needs to be designed as a “Jack of All Trades,” suitable for a wide market to utilize as a CPU. This general purpose CPU has a lot of features and circuitry that a car-specific CPU does not need.
Thus, after Elon convinced Jim that he was serious and willing to go the distance with this expensive project, the HW3 was developed specifically for a vehicle application including self-driving. Model 3, S, and X vehicles with HW2/2.5 who had purchased the “Full Safe Driving” feature were upgraded to HW3 at no cost to the owner of the car; a free upgrade of a magnitude that is unheard of in the automotive industry — if they could even do such a thing. As the “Full Self Driving” beta now available to some “safe driving” owners proves, HW3 is capable of doing the job.
The new Tesla GPU is, I understand, made by Samsung in a modern Korean fab shop. When HW3 was needed, the ultimate in a smaller nanometer chip was not then available from Samsung. Remember, smaller is faster in chips, because the electrons have less distance to travel. So, a smaller chip is faster, and as a bonus, it runs cooler. Since Elon and Jim are forward thinkers, HW4 was designed at the same time as HW3. It will be a smaller chip, rumored to be down to 7nm, also to be manufactured by Samsung. Samsung is building a new modern fab shop in Austin, Texas, close to Giga Texas. Again, the rumor mills are saying the Cybertruck will use HW4 as well as a new suite of cameras. Also rumored is a retrofit to FSD Model S/X vehicles that will be delivered with HW3.
Which brings us up to the present day on Tesla’s GPU as we know it. Obviously, there will be faster and more capable specialized processors in Tesla’s future, but we will stop here as not to demoralize Tesla’s competition any further.
One last thing: Tesla spends the extra money to place its GPU on a printed circuit board (PCB) that is fighter aircraft quality. And like a fighter, reliability is hyper important with full self-driving. To that end, Elon put two of the HW3 GPUs on the PCB. always looking over each other’s shoulder to assure 100% processing reliability. Even a Boeing 747 doesn’t have redundant processors!
Next, in part four: why the rheostat had to be replaced to have a practical, reliable, and efficient EV. Again, solid-state technology had to advance and come to the rescue.
It is harder than it looks. Early Model 3 electronics circuit schematic:
Appreciate CleanTechnica’s originality? Consider becoming a CleanTechnica Member, Supporter, Technician, or Ambassador — or a patron on Patreon.
PlatoAi. Web3 Reimagined. Data Intelligence Amplified.
Click here to access.