3 ms·

I wish so much RISC-V to be a success. Modern and standard ISA, worldwide royalty free (unlike x86 and arm). I'll write my code paths directly in assembly wit

I wish so much RISC-V to be a success. Modern and standard ISA, worldwide royalty free (unlike x86 and arm). I'll write my code paths directly in assembly with a conservative usage a macro preprocessor. Take this compilers and language syntaxes planned obsolescence! But I stay humble as I know that without extremely performant implementations (open source/closed source) on the latest silicon process, success will be unlikely or at best somewhat limited.

ksec 3y ago
Tenstorrent has an implementation that promise to have about the same projected performance per watt as Apple's A16. We will have to wait and see if that is true. Although the architect of that chip is also the architect of Apple's M1 aka A15.
- snvzz 3y ago
  On their RISC-V Summit speech slides, they even compared the implementation (Ascalon) with where they expect Zen5 to be, providing similar performance at lower power. In any event, I am hopeful they will deliver something strong on time for it to be competitive. Considering the backgrounds of the people involved, this looks likely.
Rochus 3y ago
As long as there are not much better and cheaper implementations available than we have on the ARM side today, it's unlikely to succeed (the humble opinion of someone who implements embedded systems). And it's also unlikely to be royalty free; even if we don't have to pay royalties for the ISA, there will still be royalties for using patents and such if the implementation is really something. By the end of the day, RISC-V solves only a fraction of the problems (and has the potential to add new ones); personally, I think it is very unlikely that it will ever reach ARM's or AMD's volume.
- hakfoo 3y ago
  I think there's two messages RISC-V supplies, and one is less obvious. "It's free to license" means we don't have to send ARM a nickel. That looks compelling today. But the bigger story is much more the "It's open" angle. That means tomorrow you don't have to have faith in the stewards of the platform. You aren't relying on a fickle short-term license that they can strap stupid terms to next month (see: the ARM accelerator fracas) If you're planning a long-lifecycle product family, you don't want a sudden policy change to scrap half your roadmap, and RISC-V protects you there.
  - Rochus 3y ago
    > you don't want a sudden policy change to scrap half your roadmap, and RISC-V protects you there. I didn't ever experience a situation where we were free of thirdparty IP; RISC-V only covers the ISA; everything else remains the same; so maybe we get less dependent of one company, but a lot of other dependencies remain; whether this is worthwhile remains to be demonstrated.
    - photonbeam 3y ago
      and hopefully you have choices for that third party IP too This just reduces lock-in for CPU IP
musicale 3y ago
I thought Eben Upton's comments were interesting regarding using RISC-V for a Raspberry Pi board. Basically x86 still has much better software support than ARM, which has much better support than RISC-V; and on the hardware side, there were no readily available RISC-V cores that were competitive (presumably in price/performance/power) with their current ARM cores. So RISC-V wasn't a yet a viable option for mainline Raspberry Pi systems. He also complained about current RISC-V hardware implementations having too many feature variations (and this vs. ARM, which is worse than x86 already.) However, he thought that RISC-V would work fine today for microcontrollers where 1) you're running a small/thin software stack and 2) raw performance isn't that important (which as pointed out below seems to be saying that current RISC-V implementations should be able to match the lower computations-per-second of microcontrollers like RP2040/Rpi Pico without exceeding the power/heat/space/cost requirements for that chip/board.) (My own take is that cheap x86/celeron mini-PCs are becoming more competitive with boards like Rpi for use cases where you don't need the credit card form factor or GPIO, and x86 software support is compelling. And I agree that it will take a while for RISC-V to get good hardware and software support, though it seems to be improving every year.)
- Rochus 3y ago
  > raw performance isn't that important. Well, I assume he didn't mean that performance isn't important for microcontroller applications in general; it's usually very important; you generally have to squeeze out the last bit and MCU cycle, because a bigger MCU is usually not feasible in an industrial setting; and some devices do real number crunching.
  - musicale 3y ago
    Well, the deleted words "for...where" (meaning "in cases where") do matter, but... Yes, in fact I was considering different wording when I wrote it to make the distinction, but I think the word he used was "performance." Edited to make this clearer.
    - Rochus 3y ago
      Maybe you can post a reference to the corresponding interview or article.

ksec 3y ago

Tenstorrent has an implementation that promise to have about the same projected performance per watt as Apple's A16. We will have to wait and see if that is true. Although the architect of that chip is also the architect of Apple's M1 aka A15.

snvzz 3y ago
On their RISC-V Summit speech slides, they even compared the implementation (Ascalon) with where they expect Zen5 to be, providing similar performance at lower power. In any event, I am hopeful they will deliver something strong on time for it to be competitive. Considering the backgrounds of the people involved, this looks likely.

Rochus 3y ago

As long as there are not much better and cheaper implementations available than we have on the ARM side today, it's unlikely to succeed (the humble opinion of someone who implements embedded systems). And it's also unlikely to be royalty free; even if we don't have to pay royalties for the ISA, there will still be royalties for using patents and such if the implementation is really something. By the end of the day, RISC-V solves only a fraction of the problems (and has the potential to add new ones); personally, I think it is very unlikely that it will ever reach ARM's or AMD's volume.

hakfoo 3y ago
I think there's two messages RISC-V supplies, and one is less obvious. "It's free to license" means we don't have to send ARM a nickel. That looks compelling today. But the bigger story is much more the "It's open" angle. That means tomorrow you don't have to have faith in the stewards of the platform. You aren't relying on a fickle short-term license that they can strap stupid terms to next month (see: the ARM accelerator fracas) If you're planning a long-lifecycle product family, you don't want a sudden policy change to scrap half your roadmap, and RISC-V protects you there.
- Rochus 3y ago
  > you don't want a sudden policy change to scrap half your roadmap, and RISC-V protects you there. I didn't ever experience a situation where we were free of thirdparty IP; RISC-V only covers the ISA; everything else remains the same; so maybe we get less dependent of one company, but a lot of other dependencies remain; whether this is worthwhile remains to be demonstrated.
  - photonbeam 3y ago
    and hopefully you have choices for that third party IP too This just reduces lock-in for CPU IP

musicale 3y ago

I thought Eben Upton's comments were interesting regarding using RISC-V for a Raspberry Pi board. Basically x86 still has much better software support than ARM, which has much better support than RISC-V; and on the hardware side, there were no readily available RISC-V cores that were competitive (presumably in price/performance/power) with their current ARM cores. So RISC-V wasn't a yet a viable option for mainline Raspberry Pi systems. He also complained about current RISC-V hardware implementations having too many feature variations (and this vs. ARM, which is worse than x86 already.) However, he thought that RISC-V would work fine today for microcontrollers where 1) you're running a small/thin software stack and 2) raw performance isn't that important (which as pointed out below seems to be saying that current RISC-V implementations should be able to match the lower computations-per-second of microcontrollers like RP2040/Rpi Pico without exceeding the power/heat/space/cost requirements for that chip/board.) (My own take is that cheap x86/celeron mini-PCs are becoming more competitive with boards like Rpi for use cases where you don't need the credit card form factor or GPIO, and x86 software support is compelling. And I agree that it will take a while for RISC-V to get good hardware and software support, though it seems to be improving every year.)

Rochus 3y ago
> raw performance isn't that important. Well, I assume he didn't mean that performance isn't important for microcontroller applications in general; it's usually very important; you generally have to squeeze out the last bit and MCU cycle, because a bigger MCU is usually not feasible in an industrial setting; and some devices do real number crunching.
- musicale 3y ago
  Well, the deleted words "for...where" (meaning "in cases where") do matter, but... Yes, in fact I was considering different wording when I wrote it to make the distinction, but I think the word he used was "performance." Edited to make this clearer.
  - Rochus 3y ago
    Maybe you can post a reference to the corresponding interview or article.