Hacker News Reader

Comment by frankchn | Hacker News Reader

> Apple has already laid the architectural groundwork, a chip with similar performance numbers can probably be commissioned with ~50 million.

I think you might be two orders of magnitude off here. A class leading general purpose compute chip on a cutting edge process? I don’t think $5 billion to produce working chips is unreasonable. If it were only $50M to produce A14/M1 level chips, Qualcomm would already have done so.

Cerebras is cool, but they are not on TSMC 5nm and don’t have to design general purpose silicon, where you have pesky things like huge reorder buffers or 8-wide decode blocks, which are not seen in any other production CPU right now.

View on HN

Agreed. Apple has acquired multiple companies and been building custom CPUs for 10 years with a significant sized design team. They paid about 280M for P.A. Semi in 2008. They reportedly also hired a number of DEC StrongARM refugees as well as several other small design teams. Apple was also an original ARM investor and have been using ARM cores for various uses for decades.

A competitor also needs 10 years experience designing higher and higher performance ARM cpus in tremendously heat/power constrained systems (phones). There are a lot of long term threads that have come together in the M1 chip.

Building a useful 8-wide decode block on x86 might be extremely expensive in terms of area and power.

For a long time, we’ve been told that the nasty x86 ISA is no big deal and that a good CPU can cache decoded instructions and play other tricks to make the ISA irrelevant. But x86 really does have fundamental problems. The length of an instruction can’t be determined until the instruction is almost fully decoded, which makes deciding in parallel quite nasty. And the x86 memory model is fundamentally more expensive than ARM’s.

M1 is the first serious attempt to make a competitive high-performance non-x86 CPU. Perhaps the real lesson from M1 will be that x86 can’t compete.

You are right, I am mostly estimating based off upgrading existing architectures with insights obtained from M1. 50 million may be enough for the raw manufacturing cost for one batch but research and development from scratch would not be cheap. I am basing the cost off the assumption that there are already experienced chip designers at Intel/Qualcomm who are willing to take the opportunity to build a startup with a 50 mil budget.

Reply on news.ycombinator.com