> the x86_64 patents should expire this year. Apple could have probably made their own x86 chip without paying royalties. The free/open architecture that we've always dreamed of, might turn out to be the one we're already using.
I've never seen anyone mention this before. Is this really in the cards? Or are all the extensions patented as well, so that a hypothetical Apple (or whoever else takes it on) x86 wouldn't be able to use x86 without licensing it?
Author has clearly not heard about NexGen! They implemented a RISC architecture that code morphed x86 i586 design. They were then bought out by AMD who shortly thereafter adopted the design for their own architecture. Then Intel had no choice but to adopt NexGen architecture as well lool. We all use NexGen Architecture now. The most famous microprocessor no one's heard of. I only know because NexGen was based in Milpitas and that's where I live, so Milpitas Architecture is like hometown pride.
As for patents, Xed is tiny (only ~4kb) and it allows us to distinguish the which instructions belong to which microarchitectures, e.g. k8, sse3, sse4, avx. That gives us a clear accurate picture of the intellectual property rolloff, thereby enabling a fabless chip or emulator to simply ignore the encumbered parts of the encoding space. The tradeoff is you can't patent troll Intel (due to Xed being licensed Apache 2.0) and I think that's fair.
Nexgen was great, the lack of a FPU at that moment in time however was fatal at launch.
Quake was just starting to ignite the world and pentium performance at 486 prices would have done more, but no fpu... it ran NT fine enough although my memory wants to say Microsoft detected it as a 386 so that would have cut NT 4 out of the picture.
I thought knowledge of them was more common but I guess not
The TL;DR seems to be that if you restrict yourself to implementing an ISA from 20 years ago, your chip can indeed be patent free, but that isn't going to run modern software - maybe not even modern x86 32-bit software!
So ISAs are patentable? Even if you reimplement them from scratch, it's covered? Can anyone ELI5 that in the light of the Oracle vs Google "APIs are not copyrightable" battle? Seems like an ISA (sans implementation) is comparable to an API, so is the difference patents versus copyright? Are APIs patentable?
I haven't investigated it in detail personally, but my understanding is that Intel has or had troll patents on the obvious (and therefore not legitimately patentable) way to go about implementing various x86 features. So if you reimplement it from scratch, the way you naturally go about implementing it will just happen to match 'Intel's' supposed 'inventions' which they have patents on. (Eg, I think they had a patent on register renaming at one point.)
I think neither the basic idea of register renaming nor solutions to do it in a way that speeds up the processor (conceptually, each register reference has to go through a logical register-physical register lookup) are so obvious that they are guaranteed to be unworthy of patenting.
If Intel had a patent on this, it must have been on some (possibly tiny) variation, though, as IBM had register renaming in hardware in 1967 (See https://en.wikipedia.org/wiki/Tomasulo_algorithm) and the POWER1 had it in 1990.
Even if the patent on the instruction set expired, that's only half the battle (if that). When you licence ARM you're getting both the instruction set design AND the implementation.
AFAIK Apple mostly only licenses the ISA and then runs a completely custom implementation, so in this case that's not relevant.
I think the sibling comment is more accurate: while the core of the ISA might effectively be public domain soon, there are plenty of extensions in common use that aren't. If you implement a custom x86 design without those a lot of software won't be able to run or it will run with significantly worse performance.
The assumption I'd made was that they'd left the actual core mostly alone. They've gone a lot further than everyone else in what they've put around it though, e.g. large caches and other accelerators.
I'd be interested in being pointed to any references as to what they've actually done, not that much information is likely to be public domain.
My source is "I know somebody who knows somebody who works in Apple's hardware design team" so take it with a grain of salt but it seems that their core is almost entirely custom, or at least customized to the point where it's effectively a different core.
And it makes sense for them to do it too, they clearly have the money and people to pull it off and it gives them a big competitive advantage. It's not for nothing that they push for ARM on the desktop as well, this way they'd have a tighter control on their ecosystem than they've had for decades (maybe ever), both soft- and hardware.
Yes. So, if you want to reimplement a CPU from 1999, you will probably be fine without any kind of license. If you want to add features or use new methods, however, you may run into infringement issues.
I've never seen anyone mention this before. Is this really in the cards? Or are all the extensions patented as well, so that a hypothetical Apple (or whoever else takes it on) x86 wouldn't be able to use x86 without licensing it?