This is going to ding NVDA, I’m just not sure how much.
CNBC: Tesla working with AMD on self-driving car chip
Sep. 20, 2017 4:14 PM ET|By: Brandy Betz, SA News Editor
CNBC reports that Tesla (NASDAQ:TSLA) is running tests on its own sample chips for autonomous vehicle driving tasks.
The Tesla chips build on Advanced Micro Devices (NASDAQ:AMD) technology through a wafer agreement AMD has with its spin-off, GlobalFoundries.
Creating its own chip will allow Tesla to ease its dependency on Nvidia (NASDAQ:NVDA) GPUs. Tesla already parted ways with chipmaker Mobileye, since acquired by Intel, due to a vehicle accident. Intel is working with Google’s Waymo on self-driving car tech.
AMD shares are up 2.04% aftermarket.
Previously: Intel-backed Waymo cars log 3M miles of real-world driving (Sept. 18)
And here is the defense by an analyst, who is probably quite correct.
Nevertheless, given the U.S. tax system, not much I can do about it anyways as capital gains are far too high and the taxes would be outrageous.
The analyst basically says that it is likely Nvidia will be the primary GPU and AMD will be used for specific AI features as an adjunct and double source. However, from the stories (as I have read them) it is clear that in fact THERE IS NO AMD GPU in this announcement. It is Tesla, rumored (and these rumors like this are true more often than not), to be using AMD IP to build an ASIC for an undetermined aspect, or aspects, of the autonomous driving AI technology.
Whether or not Tesla will utilize an ASIC for the entirety of its technology, or a an adjunct as the analyst comments on, TBD. However, given the large cost of development, and time needed to get to market, it would be astounding if Tesla was investing to create an Apple like (keep it all in house) AI autonomous driving chip to replace everything they have done up to now. But with Musk, who knows.
But that is the gist of the issue above. AMD “IP” is being used by a former AMD chip designer, to build an in-house ASIC (what else could it be) to be used in Tesla’s autonomous system for an undetermined function(s). No mention of Nvidia being displaced.
<<<But that is the gist of the issue above. AMD “IP” is being used by a former AMD chip designer, to build an in-house ASIC>>>
Sounds a lot like Cisco’s complaint against Arista actually, except Tesla is admitting using AMD IP for its former AMD engineer (and thus will probably pay a royalty).
Remember, creating a chip is not a one time process. It is a continuing process of maintaining a technology upgrade roadmap, that includes continuing improvements to the product, software support, etc. unless your chip quickly becomes antiquated. That is why car companies out-source to chip companies, etc.
But the usual rules do not apply to Musk in many cases, so lord knows what any of this may mean or not mean. What it probably means is that there are certain aspects of autonomous driving that are not working to Tesla’s liking, and Tesla is trying to patch up what they consider not solved by their current solutions.
“…Tesla is trying to patch up what they consider not solved by their current solutions.”
Or they are adding functionality for future versions. There’s not necessarily that much difference between the two explanations.