3 ms·

I don't really see how the ISA is relevant, maybe that it affects 3% of a chip performance, 15%, or perhaps even 30% but either way it is nothing in the grand s

I don't really see how the ISA is relevant, maybe that it affects 3% of a chip performance, 15%, or perhaps even 30% but either way it is nothing in the grand scheme of things. Our software is order of magnitudes slower than it could be, RISC-V or x86 doesn't change anything.

snvzz 3y ago
>I don't really see how the ISA is relevant There's definitely better and worse ISAs e.g. some are unjustifiably complex. Yet the key advantage of RISC-V is not at all technical, and lies in its licensing scheme. There is no need to ask for permission to implement RISC-V and thus be able to leverage the best ecosystem out there, which RISC-V is rapidly building. There is also no need to ask for permission to extend it with custom extensions, to suit your specialized needs. Using the RISC-V trademarks is only a little more complicated, as it requires proving standard compliance. Yet this aligns well with most implementations; Compliance is desirable as it enables tapping into existing software and tooling, which is much cheaper than building and maintaining bespoke solutions.
- TheMode 3y ago
  Some ISAs are definitely more complex than they should, but RISC-V/x86 is the least of my problem if the goal is to interpret some windows/macos/linux binary. The complexity goes beyond a list of instructions. As for licensing I agree that it can be seen as a step in the right direction, but at the same time it seems like we absolutely refuse to learn. All standards inevitably deprecate, I'd prefer every hardware manufacturers to come up with their own one-off ISA and an easy way to run/interpret existing applications. Software ecosystems are more about locking you to a specific vendor/standard, they prevent the emergence of better systems.
panick21_ 3y ago
There is a question of time. How long does it take to implement a chip of X speed with RISC-V vs x86. With x86 there is so much more to implement and even more stuff to verify. > Our software is order of magnitudes slower than it could be, RISC-V or x86 doesn't change anything. That's not really true, the most critical bits of the stack or often implemented pretty fast. This is what DEC learned: "The conclusion was clear: for any given amount of investment, the RISC designs would outperform a VAX by 2-to-1." - DEC is dead, long live DEC I'm not suggesting that it is 2-to-1 now but its still relevant.

snvzz 3y ago

>I don't really see how the ISA is relevant There's definitely better and worse ISAs e.g. some are unjustifiably complex. Yet the key advantage of RISC-V is not at all technical, and lies in its licensing scheme. There is no need to ask for permission to implement RISC-V and thus be able to leverage the best ecosystem out there, which RISC-V is rapidly building. There is also no need to ask for permission to extend it with custom extensions, to suit your specialized needs. Using the RISC-V trademarks is only a little more complicated, as it requires proving standard compliance. Yet this aligns well with most implementations; Compliance is desirable as it enables tapping into existing software and tooling, which is much cheaper than building and maintaining bespoke solutions.

TheMode 3y ago
Some ISAs are definitely more complex than they should, but RISC-V/x86 is the least of my problem if the goal is to interpret some windows/macos/linux binary. The complexity goes beyond a list of instructions. As for licensing I agree that it can be seen as a step in the right direction, but at the same time it seems like we absolutely refuse to learn. All standards inevitably deprecate, I'd prefer every hardware manufacturers to come up with their own one-off ISA and an easy way to run/interpret existing applications. Software ecosystems are more about locking you to a specific vendor/standard, they prevent the emergence of better systems.

panick21_ 3y ago

There is a question of time. How long does it take to implement a chip of X speed with RISC-V vs x86. With x86 there is so much more to implement and even more stuff to verify. > Our software is order of magnitudes slower than it could be, RISC-V or x86 doesn't change anything. That's not really true, the most critical bits of the stack or often implemented pretty fast. This is what DEC learned: "The conclusion was clear: for any given amount of investment, the RISC designs would outperform a VAX by 2-to-1." - DEC is dead, long live DEC I'm not suggesting that it is 2-to-1 now but its still relevant.