Then, boom. Over night AMD leapfrogged Intel, and now on the third generation has a firm lead in mid range desk top all the way up to high end server silicon. Obviously, Intel has major partnerships with most OEMs, so despite their shortcomings they're still doing strong.
Of course, if ARM or RISCV really is the future this is just a blip. Honestly, I don't see x64 going anywhere for at least 15 years - though hope I can look back at this and roll my eyes someday...
To improve yields they used a number of 8-core "chiplets" manufactured in 7nm tied together with a 14nm I/O die. This meant that while a certain blue competitor was still trying to get 28+ core monolithic dies, AMD could scale to huge core counts without reducing yields by tying together more chiplets.
Their 64-core 128-thread Threadripper workstation parts should have yields almost exactly as good as their low-end CPUs. The Xeon W 28-core monolithic monster? Not so much.
AMD did a lot of things right this time. Intel's got some catching up to do.
(Yes, I'm aware of Samsung.)
ARM generates less than $2B in revenue. Apple posts $275B in revenue. Are ARM's fees just a rounding error to anyone with scale? And ARM knows that it needs to remain competitive on its licensing fees to make sure that people don't move to RISC-V.
I'd guess that a lot of ARM's revenue actually comes from the processor design, not the ISA. ARM will license you cores. RISC-V won't license you cores since they're not designing them.
It's possible that RISC-V will see great things, but I'm kinda thinking that ARM's license fees probably aren't much. Apple especially wouldn't be paying much since they're not licensing cores. Anyone that is licensing cores would need to replace that R&D with their own - which might be more expensive than it's worth. Qualcomm seems to still lean on ARM's designs.
And I think there's certainly a big head start in optimizing things for ARM that will be tough to overcome.
I just think it seems unlikely that current customers will drop ARM to save 0.1% of their revenue - especially if they need to start taking on the costs of designing the chips themselves, contributing to compilers, etc.
The exception I can see is China. China might want a free-and-clear route to their own processors without worrying about other nations denying them access to IP.
ISAs change, and there was a recent article written about this particular situation too: https://codasip.com/2020/12/22/does-isa-ownership-matter-a-t...
I don't really, I was just keeping the comment short and avoiding the ire of any RISC-V fans. It could be either, but it's important to remember that instruction sets don't usually win solely based on merit.
For example, if you read most compiler books - because they were written in the early 2000s or late 90s, they spend almost all of their time discussing various RISC architectures (particularly the Alpha). The (then) 386 is usually mentioned in passing, ARM isn't even mentioned at all.
(You probably can guess what I'm getting towards) They're all dead, somehow ARM lives. It's also not really a question of money, Itanium wasn't that bad and Intel couldn't save it.
If I were God, the only CPU that I really really want to see make it to Silicon (even just to see if they're right) is the Mill. Even if they can't walk the walk it's a really nice concept (I think the belt will see use after their patent expires)
I'm not sure Risc-V will win out over ARM, but if it does, I think the acquisition by nvidia will play a big part of it. It really depends on how open and cooperative Nvidia is with ARM IP now that they have it.
- Typed from my Laptop running Linux
Personally, I don't particularly care whether my computer runs RISC-V or not (for the reason above, it's only open source at the edges) - I'm more concerned with the openness of the rest of the machine
There's this weird obsession with blaming AMD shortcomings on Intel...
AMD can't produce enough chips for OEMs so they get left out in places. End of story.
It's fun to joke about "14nm++++++" but it just so happens actually being able to stock your chips is a pretty big competitive advantage.
Take for example the 'Ultrabook' branding. It's an Intel trademark, and for a few years that word was _everywhere_. However, to get the shiny badge on your system you had to give a lot of control to Intel over the design and characteristics [1]. That's not to say AMD doesn't do the same thing to a degree - They own the term "APU" and will never let Intel refer to an i7 with 'Iris Xe Graphics' as an APU, Never.
The point is that Intel for a long time had leverage over most of the PC industry and used it to get exclusivity over certain parts of the market.
---
[1] https://www.eweek.com/mobile/intel-ultrabook-partners-look-f...
I mean take things like the Ultrabook... you realize your article is literally a decade old?
I remember those days Intel was literally inventing a term and paying developers to come up with "ultrabook app ideas" and stuff. Now the usage of the term is pretty much toothless: https://store.hp.com/us/en/dlp/amd-ultrabook
And that was a bizarre example to choose anyways if you're willing to dig into ancient history, there was a point OEMs were afraid to work with AMD because of Intel...
-
Now a days their iron grip has waned, but AMD literally cannot produce. It's that simple.
