I think that if AMD ever goes the ARM route, it will be for servers and not for mobile/tablets. Or even PCs.
I think it'll depend on what is meant by "going ARM". I could see AMD doing an embedded custom chip that uses standard ARM cores (maybe even slightly tweaked ones). But to me that wouldn't be AMD "going ARM", which I would agree that if AMD takes an ARM initiative on their own, it'd likely be pairing all the logic and interconnect that they're using to make Zen/EPYC good, and then just use chiplets of ARM cores (maybe with a Zen Chiplet that serves as a main controller/OS module; maybe even something weird like Intel's Knight's ___ chips, where they use a big block - like maybe hundreds - of ARM cores to do some simple really parallel task that isn't well suited for GPUs, with a Zen Chiplet handling OS duties).
Which, I was just thinking. What's to stop AMD from buddying up to ARM to play a role in the the cores development? Like especially as ARM is starting to focus on performance oriented chips (aimed at the PC space), where it would likely help both of them (ARM benefits from AMD's knowhow with x86 performance, along with gaining a big partner that could push their cores into servers, and AMD gains a new core - think of it as an intermediate efficiency focused one between mobile focused ARM and workstation/server level x86/64 Zen, which AMD could pair with their GPU and put in laptops and embedded designs; plus it means AMD might not need to differentiate Zen cores as much and can focus them even more, since, my thinking is ARM is what sub 10W, then the intermediate would be for the 10-40W, and AMD can target Zen for the 45W+ markets instead of making sure it can scale down into the like 15-25W range). Plus, wouldn't that be better than being stuck relying on Intel dictating the x86 development?
On top of that, there's definite potential avenues for shared GPU development. From there, AMD can work a cross-licensing IP deal with Qualcomm, where the latter can use AMD GPU IP (which AMD already won in court over GPU IP related stuff that Qualcomm was doing; which is interesting since Nvidia failed at doing the same thing), and AMD can maybe use Qualcomm networking especially wireless IP (even if its delayed, so that AMD can only integrate the previous gen, it'd still be a big boon for AMD, so say as Qualcomm is pushing 5G, AMD would get to integrate LTE/4G stuff; Qualcomm keeps their cutting edge, gains a big partner; Qualcomm can focus on networking stuff, while AMD/ARM develops the CPU/GPU - since Qualcomm was allegedly wanting to get out of that development after they were seeing where Apple was heading and realizing they wouldn't be able to keep up especially as Samsung and others started following suit). They each get to integrate the IP into chips (so AMD wouldn't need to include a separate modem chip for SoC designs, or maybe in the future that chip becomes a chiplet), and since they're focused on different markets it'd be more collaborative, strengthening each in their preferred market, reducing their R&D cost since they'd have the other doing most of the areas they aren't well suited at. They both use TSMC which should make sharing/integrating easier, but it could also open up GF for Qualcomm since AMD has an outsized presence and familiarity with GF, which that could be a further positive, as all 3 being American (although with the Broadcom merger - which hasn't gone through yet so could change too, maybe that wouldn't be true; but then there's plenty of foreign investment in AMD and GF) could help them win military/government deals.
I feel like we'll see the computing market start to resemble the automotive market, where there's lots of R&D sharing and collaboration, especially as performance gains and process improvement become harder to come by, that they'll look for ways to spread the costs. That does already happen, but I feel like it hasn't been where it is in the automotive market, and the thing is, I think the computing market is more primed for it since a lot of companies will be using the same manufacturer (TSMC), and the specifics that they share require high level proficiency (by that, I mean, sure all cars use the same roads and follow similar regulations, but in computing, it has to meet technical standards for interoperability - has to run a standard architecture for software compatibility or else it'll go nowhere unless you do all the development yourself or find a way to translate it and not loose too much performance or cause other problems).
Facebook
Twitter