3 ms·
AFAICT most ISAs themselves are not slow. Most modern ISAs can be implemented quickly because the decoder is not the bottleneck and high performance cores do so
by Pet_Ant 2y ago
AFAICT most ISAs themselves are not slow. Most modern ISAs can be implemented quickly because the decoder is not the bottleneck and high performance cores do so much. So it's more like a question of why hasn't someone with a high-performance core slapped a RISC-V ISA on it?
Well, for one they weren't built with RISC-V in the first place so it would take some work to do and test. Second, there isn't currently a market for this. There is not consumer item that has RISC-V and could sell more units by getting a faster one. Third, a lot of cores are probably entangled with ARM IP for now.
None of these are integral to RISC-V itself. If you could rewrite history and swap the ISAs for ARM and RISC-V it'd still have a headwind right now from the late start, and the tail wind of being license free. It's entirely non-technical.
- bhouston 2y agoI completely agree that it is a relative "simple" matter to put a fast core in there, the techniques are known. You may not beat an Apple Mx chip, but you could at least get somewhere close to it, like 50% of the performance. And then RISC-V would take off like crazy.
- Wytwwww 2y agoWho actually has fast core (>=50% of Apple M series) that they could just slap RISC-V on, though? ARM, Intel, AMD, Qualcomm? IIRC the Switch 2 chip is going to use Cortex so even Nvidia can't be bothered to design their own CPU cores. CPUs seem to be a commodity at this point so unless you're willing to invest massive amounts to be able to compete to catch up and overtake them just paying whatever ARM wants (probably pennies) seems like a good deal.
- mycall 2y agoOne thing is true. ARM can't play too many license games in the future, now that RISCV is maturing.
- snvzz 2y ago>You may not beat an Apple Mx chip That's Wei-han Lien, whom later led Tenstorrent's Ascalon design.