Tesla advocates creating "world's best processor" for self-driving – TechCrunch



[ad_1]

On the occasion of "Independence Day" today, Tesla detailed the new custom chip that will run the autonomous driving software in its vehicles. Elon Musk has peremptorily termed it "the best chip in the world … in an objective way". This may seem difficult, but it should definitely be done.

Called for the moment "fully autonomous computer" or "FSD computer", it is a specially designed high-performance chip (designed by Samsung Texas) solely for the sake of self-sufficiency and security. The question of whether it really outperforms its competitors is not a simple question and we will have to wait for more data and further analysis to say more.

Pete Bannon, a former Apple chip engineer, reviewed the specifications of the FSDC. While numbers may be important for software engineers working with the platform, what is more important at a higher level is the satisfaction of various requirements specific to autonomous driving tasks.

Redundancy is perhaps the most obvious characteristic for audiovisual vehicles. The FSDC consists of two duplicate systems juxtaposed on the same table. It is an important choice, although barely unprecedented, simply because dividing the system in two naturally divides its power. Therefore, if performance was the only indicator (if it was a server, for example), you would never do it.

In this case, however, redundancy means that if an error or damage creep in one way or another, they would be isolated from one of the two systems and a reconciliation software the would detect and report them. Meanwhile, the other chip, on its own power systems and storage, should not be affected. And if something happens that breaks both at the same time, the architecture of the system is the least of your worries.

Redundancy is a natural choice for audiovisual systems, but it is made more acceptable by the extreme levels of acceleration and specialization that are possible today for neural network-based computing. A normal graphics processor, as you would in your laptop, will become familiar with graphics calculations, just as a special computing unit for neural networks will defeat even a graphics processor. As Bannon notes, the vast majority of calculations are a specific mathematical operation and bring huge benefits in terms of performance.

Combine this with RAM RAM and high speed storage and you will have very few bottlenecks when it comes to running the most complex components of autonomous driving systems. The resulting performance is impressive and enough to resonate with pride Musk during the presentation:

"How could it be that Tesla, who has never designed a chip before, designs the best chip in the world? But it is objectively what happened. Not better by a small margin, better by a large margin. "

Take this with a grain of salt, as Nvidia's engineers, Mobileye, and other autonomous driving issues would certainly disagree with this statement for various reasons. And even if it's the best chip in the world, there will be a better one in a few months – and no matter, the hardware is only as good as the software it uses. (Thankfully, Tesla also has an incredible talent on this side.)

(A quick note for a terminology you may not be familiar with: OP The abbreviation is Operation for Second, and it is measured in billions and trillions of dollars nowadays.FLOP is another common term, meaning operations floating point two, they concern more precise calculations, often used by supercomputers for scientific calculations: one is neither better nor worse than the other, and they should not be compared directly or considered as exchangeable .)

High performance computing tasks tend to drain the battery, such as transcoding or editing HD video on your laptop, and the dust disappears after 45 minutes. If your car did that, you'd be angry, and rightly so. Fortunately, a side effect of acceleration tends to be efficiency.

The entire FSDC works with about 100 watts (or 50 per computing unit), which is pretty low – it's not a cellphone chip, but it's way below what a desktop or a high-performance laptop would shoot, less than many GPUs. Some AV-oriented chips pull more, others less, but Tesla claims to have more power per watt than its competitors. Again, it is difficult to immediately look at these claims, given the closed nature of the development of audio-visual equipment, but it is clear that Tesla is at least competitive and can very well beat its competitors on some important parameters.

Two other features specific to the AV, present on the chip, although they are not duplicated (the computation paths converge at a given moment), are a block work of the processor and a security layer. Lockstep means that it is very carefully applied that the timing on these chips is the same, ensuring that they process exactly the same data at the same time. It would be disastrous not to be synchronized with each other or with other systems. Everything in audiovisual vehicles relies on a very precise timing while minimizing delays. Robust locking measures are therefore put in place to keep it clear.

The security section of the chip checks commands and data cryptographically to monitor, essentially, hacking attempts. Like all audio-visual systems, it is a finely oiled machine and interference should never be tolerated – lives are at stake. Thus, security closely monitors the input data. and exit to detect any suspicious item, such as spoofed visual data (to make the car believe that there is a pedestrian, for example), to change the output data (for example to prevent it take the appropriate precautions to detect a pedestrian).

The most impressive part of all perhaps is that all this custom chip is backwards compatible with the existing Teslas system, that it can be directly inserted and that it will not cost as much. Exactly what the system itself costs Tesla and how much you will be charged as a customer. It will probably vary. But even if it is the "best chip in the world", it remains relatively affordable.

Part of that could come from a manufacturing process in 14 nm rather than the process below 10 nm that others have chosen (and to which Tesla might eventually have to migrate). For energy savings, the better, and as we have already established, efficiency is the word of the order.

We'll know more once we get a little more objective – objectively objective objections, excuses to Musk – tests on this chip and its competitors. Know for now that Tesla is not left out and that the FSD calculator should be enough to keep your model 3 on the road.

[ad_2]

Source link