A large number of sensors on Apple’s self-driving prototype car

On November 11, 2020, Apple’s self-developed chip M1 was officially unveiled. This M1 chip is a sign of Apple’s expansion from the field of mobile phones to areas other than mobile phones. This chip can be used in car cockpits and unmanned driving with a little modification. Apple has acquired Intel’s baseband team and will also be able to launch 5GV2X chips in the future.


The actual Apple M1 is just a test of water. Apple’s A14 is well known to be two large Firestorm cores and four small Icestrom cores. The number of transistors is 11.8 billion, the transistor density is 134 million per square millimeter, and the L2Cache is 8MB. M1 follows the design of A14, the transistor density is exactly the same, but two large cores are added, L2Cache is added, and one or two dedicated hard cores are added. There is nothing new besides this. Even just a simple addition, even optimization is not done, the extra 4MB of L2Cache is a plug-in, and it may be more appropriate to call it L2.5.


In doing so, Apple reused the A14 design and diluted the cost. In the future, Apple will launch a series of optimized chips with stronger performance. In terms of GPU computing power, M1 is already twice that of Nvidia’s Xavier. Nvidia’s Xavier has a GPU computing power of 1.3FLTOPS (FP32). In terms of deep learning, Xavier is relatively high, with 30TOPS (INT8). But it is easy for M1 to achieve Orin’s 200TOPS (INT8).


Apple believes that multi-core is meaningless, and the general computing power of the CPU will drop a lot on certain occasions. Therefore, Apple advocates multiple dedicated cores or hard cores. The core of M1 includes image processing, video encoding and decoding, audio processing, encryption and decryption, and neural network acceleration. Used in the cockpit or autonomous driving field, audio processing, encryption and decryption, and video codec can be replaced with binocular parallax, optical flow, and ISP.


M1 is almost the same as A14, and R&D costs can be ignored. The cost of A14 is about US$75-80, and M1 can be a little lower, about US$70. Even if M1 increases FPU to 200TOPS (simply increasing FPU hardly increases R&D costs, stacking more MACs), its price will be Far lower than the price of Nvidia Orin, about 1/2-1/3 of Nvidia. However, Apple will not compete head-on with Nvidia. Apple will not sell chips. Apple will still build its own ecosystem. This time it is the electric vehicle ecosystem.


In addition to Apple, the performance of Qualcomm Ride is also sufficient to compete with Nvidia Orin. Qualcomm has an annual shipment of at least 600 million pieces, and it can also share a lot of costs, including R&D costs and hardware costs.


After the automobile enters the era of electrification and intelligence, two major changes have occurred. One is that the threshold for building cars has been greatly reduced, and the second is that the importance of chips has greatly increased. In the era of fuel vehicles and non-intelligence, Apple cannot replicate its success in the field of mobile phones in the automotive field, but in the era of electrification and intelligence, Apple can replicate its success in the field of mobile phones.


Post time: Aug-09-2021

Send your message to us:

Write your message here and send it to us