←back to thread

283 points walterbell | 4 comments | | HN request time: 0.835s | source
1. Pet_Ant ◴[] No.45771581[source]
Couldn't you switch up the decoder logic and make it a RISC-V chip and just blow away existing competition that isn't quite Pi yet?
replies(1): >>45773840 #
2. wmf ◴[] No.45773840[source]
The decoder is made by Arm who is never going to do that.
replies(1): >>45774718 #
3. LeFantome ◴[] No.45774718[source]
I think that is the point being made. Replace the ARM decoder with a RISC-V one and make a RISC-V chip with SoundWave performance using the RISC-V ISA.

The fact that you have to argue with ARM about what you are allowed to do is the main reason not to use ARM. RISC-V is not about cost; it is about control. ARM suing Qualcomm to stop Elite X should be everything the industry needs to know to choose RISC-V wherever possible.

If you are going to launch a chip for yourself (like Apple did with Apple Silicon) or Amazon did with Graviton, I would choose RISC-V over ARM if starting today. That is what Tenstorrent did for their platform. I can see NVIDIA releasing their own RISC-V chip.

In the case of AMD, what are their customers asking for? Probably not RISC-V at this point (sadly). So ARM makes a lot of sense for them.

To get back to the original suggestion, replacing the ARM decoder in SoundWave with a RISC-V one, I do not know how feasible that is in practice. The entire chip is designed around the ISA, especially registers and memory model. It is not like compiling Kotlin instead of Java. Or rather, it could be like that if both ARM and RISC-V instructions were designed to compile down to the same micro-architecture (but they are not).

replies(1): >>45775463 #
4. wmf ◴[] No.45775463{3}[source]
My point is that you can't take a core like X925 and modify it because Arm won't allow it. If you want a RISC-V core you have to design it yourself.