Tesla switches from Intel to AMD chip for its Model Y in China


Posts: 1,584   +1,407
They will make better APUs when DDR5 arrives. You know what the primary limitation is on APUs? Bandwidth.
Why do you think Fiji had HBM? Why did Vega use HBM? Why did Nvidia develop GDDR6X? Bandwidth. You're not going to be able to match a 6700 XT with only the bandwidth from DDR5 ever. It's just not possible. Besides, not even GDDR6 was enough bandwidth for the 6700 XT: they needed to engineer a giant on-die solution for infinity cache. That takes giant amounts of die space.

No, it's not possible to have something on the AM4 or AM5 socket with infinity cache, a good CPU, AND a 40+ CU GPU. It's literally not possible on any node TSMC currently has due to size constraints, let alone memory bandwidth.
@Shadowboxer remains smugly ignorant of die size, memory bandwidth, and physical constraints. Probably ban for trolling?
Shadow isnt the troll dude, you are! Its definitely possible to get more powerful APUs than the ones we get but AMD wouldn’t make as much money that way. Same reason why AMD don’t sell cheap Ryzen 5000 CPUs or Radeon GPUs any longer. Their board clearly aren’t interested in low margins. They want to extract as much money as they possibly can from the market and supplying APUs which contain High end CPUs and GPUs sell for less than separate Ryzen and Radeon products, which they cant even meet the demand for so why would they bother?

Remember the Intel CPU with Radeon integrated graphics back in 2018 the i7 8809? That had the performance of an RX 580 roughly and was much more powerful than a 5700G today lol. It was also based on a larger die process than todays APUs and could easily it onto a single package. Its living evidence that its possible to get more powerful APUs. Although I hear AMD abandoned driver support for it, shame on them.