A threat to ARM
But since ARM isn't British any more, who cares?
The RISC-V open instruction set architecture got a boost today after it emerged that five chip giants are coming together to jointly invest in a company to develop reference architectures based on the standard. The new entity will be formed in Germany with investment from Infineon Technologies, Qualcomm, NXP Semiconductors, …
I'm predicting that this new organization won't publicly release their designs but will instead license them.
Only the founding members will have free access to the designs. Everyone else will have to pay, just like they do with ARM. Maybe the designs will be slightly cheaper than ARM, but not much.
These companies could fund the designs themselves and give them away.
Maybe they'll publish some low-end cores to get the ball rolling, but the high-end stuff will remain proprietary. Also, I believe governments are prodding them not to release their wares to prevent China or Russia from using the designs to technologically stay on par with the West.
Why would you think the designs would be cheaper than ARMs? For the last time before I give up explaining this to people outside the industry: *the license cost is an irrelevant fraction of the actual cost*.
The license cost is 2%. To get a feel for what this means in practice, suppose you were a manufacturer of engine control units, and you had the contract for the whole of Volkswagen That means you are selling 5 million units per year. The actual ARM chip in those is costing you 30 cents each. $1.5million annually. The license you pay to ARM is…..$30k. That’s less than the cost of one intern. Nobody changes a whole design to save $30k, on a project worth $500M.
It’s a ridiculous paranoid fantasy by ARM haters.
ARM don't have much in the way of competition now. I'd expect the initial release of these RISC-V designs to be cheaper than the ARM ones to try and grab market share. As you point out, that's pretty much irrelevant in reality. But if you tell Wall Street you're getting your chip designs 2% cheaper, you'll likely find your share price goes up, which seems to be all that matters these days.
Give ARM a few days and they'll be matching, if not undercutting, the RISC-V price anyway.
It's not the licensing costs but the terms and conditions to use the IP. Some companies may want to alter the design to differentiate themselves from the herd. ARM may let you do this if you pay them an exorbitant amount of money. With an open-source RISC-V design it would cost you nothing.
That makes zero sense, for RISCV, as soon as you stop to think about it. Either you mean “add special instructions”….in which case, the ecosystem is instantly fragmented since binary compatibility is broken. It’s “RISCV”, but only in the sense that every CPU on the planet has an ADD, JMP, AND, XOR.
Or, you mean “implement clever techniques to improve power and performance”, ie clock-gating, better branch predictors, memory wormholing, etc etc. For RISCV, this is definitely needed….except…it correctly admits that the inplementation details are where 99.9% of the value and design cost is. The “notional base” in the spec has was 0.01% of the design cost, and therefore zero value. Most decent hardware engineers could have written that spec from scratch in an afternoon, and coded it the next day. Making it “open source” is just laughably pointless. Giving engineers a document that says “your CPU needs an ADD, JMP, XOR” is just an embarassing waste of paper.
That’s not just a theoretical point. At the cheap microcontroller end, the market is saturated by at least a dozen different manufacturers (not ARM) all of whom just wrote their own ISAs containing maybe 20 instructions. They’ve been around for years or decades. Nobody waited for or needed RISCV. Theres no use for binary compatibility.
This post has been deleted by its author
The idea is to come up with extensions that are useful to everyone and publish them as a standard so others may use them as well. You could keep your extensions private but that would only result in them not being supported in the tooling and you'd have the task of doing it yourself.
It makes no sense to willy-nilly add extensions just to be different. That won't work.
Also: maybe you should read why RISC-V was invented in the first place [1]
[1]: https://thechipletter.substack.com/p/risc-v-part-1-origins-and-architecture
Considering that ARM is planning to jack-up the costs for licensing tenfold I'd say it's a certainty that RISC-V will be (much) cheaper. And also a given that RISC-V will quickly takeover the market for low-end MCU's and higher-end stuff eventually too.
ARM was toast the minute they broke with the unwritten law that they should remain an independent and impartial vendor of CPU cores. Alas the shareholders had dollar signs in their eyes and wanted to sell to the highest bidder. That's when things started to go off the rails.
From my reading of it, it appears that Broadcom wishes to join the consortium too:
"AVGO has announced its intention to become a stakeholder in a company aimed at advancing the adoption of RISC-V globally."
https://finance.yahoo.com/news/broadcoms-stock-top-stories-competition-183142189.html