Tesla of electric car bibcock (the Full Self-Driving(FSD) that Tesla) announced to roll out proper motion research and development a few days ago drives oneself chip, use design of double processor chip, and control Cheng production with the 14nm FinFET of Samsung. Releasing brand-new drive oneself chip while, tesla also compares the product below the banner and NVIDIA photograph, emphasize the chip that its design by oneself, have apparent ascensive; at the operation ability of NVIDIA product relatively to this, NVIDIA criterion recently fight back, think the comparison of tesla not ” accurate ” , and emphasize NVIDIA drive oneself platform has high-powered AI not only, still have at the same time have ” open sex ” .
NVIDIA own machine (Rob Csongor of vice president of Autonomous Machines) branch expresses, undeniable, tesla is rising to drive automatically the manufacturing level of the car. Above all, tesla uses double AI to handle chip, every chip includes CPU, GPU and deepness study accelerator (Deep-learning Accelerators) , operation ability can amount to 144 TOPS, can from all sorts of surrounding data is collected in cinematograph, radar, supersonic wave and algorithm of deepness nerve network. Additional, tesla also emphasizes them will developing next generation to drive oneself continuously chip, in order to achieve taller than 144 TOPS figure efficiency.
Csongor points out, tesla reiterates ceaselessly, drive automatically the car is promotion drives the key of security, efficiency and convenience sex, also be the future of automobile industry at the same time, because this needs extraordinary operation efficiency. This point of view with NVIDIA is same, this also is the matter that be designed before us and makes NVIDIA Xavier SoC. Xavier is had but process designing CPU, GPU and deepness study accelerator, data processing speed is 30 TOPS.
In the meantime, because answer to drive oneself expensive operation demand, NVIDIA also has double chip solution ” DRIVE AGX Pegasus ” , xavier tie-in GPU, make efficiency of single processor operation promotes to 160 TOPS; reach two processor conformity same platform, data processing efficiency is as high as 320 TOPS.
Csongor thinks, high speed carry is development to drive automatically the key of the car, tesla and NVIDIA all do not pass to development; toward this one party, tesla drives its FSD oneself chip and Xavier photograph are inaccurate quite. Because FSD drives oneself,chip is double chip solution, and Xavier is odd chip, both fiducial not identical. If want relative word, should compare with photograph of DRIVE AGX Pegasus, and the 320 TOPS operation of DRIVE AGX Pegasus FSD of tesla of efficiency criterion outclass drives oneself the 144 TOPS of chip.
Csongor explains further, additional, what main application of NVIDIA Xavier SoC is Level 2(or Level 2+) is automatic drive support system (ADAS) , and rather than is complete drive automatically, because drive automatically,need more powerful operation efficiency. In the meantime, tesla is in brief also by accident establish the data of Xavier SoC, xavier can provide the operation of 30 TOPS, but the data of tesla shows 21 TOPS however.
Csongor emphasizes, useless doubt ground, tesla develops high-powered chip, promotion drives oneself operation level is right way, this is prospective direction. Nevertheless, the platform of NVIDIA besides expensive operation efficiency, more important is to have ” open sex ” , can develop for industry, let car manufacturer can be offerred high-powered drive oneself; and only at present themselves uses Tesla processor.