Some interesting stuff here.
"-Tesla is working with AMD to refine its new chip, which will likely reduce its reliance on Nvidia.
-GlobalFoundries CEO Sanjay Jha said it’s working with Tesla on a chip.
-Tesla has more than 50 employees involved in the project, including chip star Jim Keller."
I wonder if this will lead to any imrprovements on AMD’s consumer GPU’s and what improvements there is going to be
Interesting. I knew they were using nVidia chips all right, so doubly interesting they would switch to AMD, though the Jim Keller part might have something to do with that. As soon as zen was out the door he left AMD again on good terms for Tesla, not hard to imagine a few kind words about his previous employer once he took up the job at Tesla having some influence.
The interesting part will be the continued updating and development of the existing cars once the new chips emerge. They will have to maintain two lines of the AutoPilot, one to work with the nVidia hardware and another to work with the AMD one. I imagine also that nVidia and AMD having the relationship they do that the hardware will have to be/going to be anyway quite different in how it deals with the information. Also nVidia being who they are likely have some amount of proprietary code in their system so the two lines thing again becomes an issue.
Unless the systems are plug and play with the existing sensors and systems, they could recall the existing cars for maintenance and replace the brains and then only have the deal with the one line of computers and its code.
I wonder also if this also helps AMD with their GlobalFoundries minimum commitment thing where X percentage of AMDs chips have to be made by GF. AMD are working with a few different foundries these days and have had delays and supply issues form GF before and it was speculated they were trying to get away from them some what in the run up to Zen, but the contract meant they still had to provide GF with a level of orders. I wonder is this will take over that commitment and they can go on and spin the big chips like Zen and the GPUs are the faster more reliable foundries like Samsung. Would add an interesting incentive, but I am very not knowledgeable about this so this is all just free wheel thinking and speculation, just mentioned it as I see GF specifically mentioned above as a point.
This is not nearly as much of an issue as you make it out to be. Code very close to silicon has to be written twice and put into libraries.The vast majority can be shared however.
And this is assuming they need duplicate code at all. If they use a vendor-independent API to begin with nothing needs to be duplicated at all.
Take a look at TensorFlow for example. If you were to write a program that utilizes deep-learning via TensorFlow there is no need to care about the underlying hardware at all. It’s all abstracted away.
Good stuff, I did not know. I was just figuring the good old proprietary nVidia tech might get in the way.
Assuming this is the third “custom design win.”
Makes sense with the Keller connection and his K12
The stock market it liking it for sure
This could be a big win for AMD. Yay competition!