It'll be interesting to see what comes next. Certainly they want to ship lots of big fancy awesome high margin things. But I'm worried that AMD is at a crossroa

It'll be interesting to see what comes next. Certainly they want to ship lots of big fancy awesome high margin things. But I'm worried that AMD is at a crossroads & is making more and more sacrifices that make the tech inaccessible to most. If AMD really wants to seed the market with new tech that will have a big impact, the tech needs to be available at a variety of price points. Programmers need to be able to get their hands on it & monkey around. I'm really afraid for AMD for a lot of reasons that they'll fail miserably with what's coming next here. Their Ryzen 7040U laptop cpus are about to ship, with once-ballyhooed XDNA aka "Ryzen AI" on the chips. But so far there is zero word of how developers can make use of this. It seems to be a very flexible FPGA system, but there's been such chronic non-delivery of tools or support or info on how folks will be able to use this that it's just draw droppingly stunning. Will AMD provide some runtimes for some popular ML toolkits & that's that? Ideally this seems like a very flexible tool that could be used for so much- but only if AMD actually lands support for folks to make use of this interesting consumer chip. AMD will be launching even bigger integrated FPGAs in the future, is the word. We should expect to see them create some higher I/O products, with big networking, would be my guess, based on what Xilinx FPGAs are good at, as well as AI. But right now Xilinx FPGAs require incredibly onerous cumbersome old-world tooling, most of which has incredibly expensive subscriptions, and even more expensive licenses if you want to use the various features already on the FPGAs. The whole PC then web era arose from people making excellent use of commodity hardware, but it's just so damned hard to imagine people actually being able to use the upcoming FPGA systems very well at all. I think AMD/Xilinx really needs to do soul searching, if FPGA is to be involve in this Number One Priority at all, and decide whether they want to sell chips, or sell expensive vendor tools + licensing. One path leads to people figuring out how to use your systems well & mass market success, the other keeps your efforts consigned to niche interests & ill adapted for the world. More GPUs is a fine possibility. But again AMD just isn't there yet. ROCm has done a ton of work trying to add more and more CUDA libraries to the table, but it's never enough. Thankfully more modern ML stacks like OpenXLA/IREE are more vendor neutral & it seems like this is less of a roadblock for the future, albeit will be an ongoing cramping issue. But what's far worse- like the above two major issues- is that ROCm is only officially supported on a very limited select number of cards, most of which cost a fortune. You can run CUDA on a huge range of cards. Again, whether AMD actually can change their tune and get total ecosystem success is just absolutely critical, in my mind, and I don't know if AMD gets that or not. There's no signs right now that AMD actually wants the open source community using their FPGAs or the GPUs, given the price-points & janky support offered. AMD needs to become a company that has a very different philosophy on software & drivers, that works much more actively to get the tech they make usable by much wider audiences of developers. It's unclear that any of this will happen. Just combining all the AI groups into one at AMD doesn't make me super hopeful here. If this group actually tries to & does a good job making consumer gear also AI and open-source capable, I'll quickly reverse my gloom here. But I think AMD has dollar signs in their eyes and doesn't understand that they can't just chase big customers. Even with recent wins with PyTorch support & what not, I just can't stress enough that they need a street presence, they need street game, to maintain any real relevance. They need mindshare. Will AMD fight for that or not?