Skip to content

Comment on Arm Predicts Stagnation If Nvidia Deal Failsparent

Comments

Not now, but the ISA is so superficial for a modern superscaler out-of-order design I don't see why existing designs couldn't easily be "reskinned", say if Nvidia-ARM jacked up fees.

I think most of the effort is getting GCC, LLVM, Linux, etc. support, and some working design to put RISC-V on top of. In other words, the difficult is in the number of cats to be herded.

In a way, I hope the deal does go through just to accelerate the transition away from ARM.

All of the RISC-V software ecosystem exists, and you can buy high performance out-of-order RISC-V implementations now from SiFive, as well as low power implementations that are very competitive with low power ARM.

The only problem right now is that high performance in the RISC-V world is about 4-5 generations behind ARM, so the fastest SoCs you can buy have cores that are roughly equivalent to a Cortex A73. SiFive is catching up a little each generation but they are a way off yet.

Yeah I think that "4-5 generations" is not about RISC-V intrinsically needing to catch up, but about the current RISC-V manufacturers needing to catch up.

I see 0 reason an apple silicon or amazon graviton couldn't be reskinned in a few months time, loosing no speed.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.