Hacker News Reader

Comment by ywvcbk | Hacker News Reader

> RISC-V

We’d just get a bunch of proprietary cores which might not even be compatible with each other due to extensions. Companies like Qualcomm would have zero incentives to share their designs with anyone.

ARM is not perfect but it at least guarantees some minimal equal playing field.

> Afaik, ARM also licenses ready-to-go cores

Which is the core of Qualcomm’s business. All their phone chips are based on Cortex. Of course ARM has a lot of incentives to keep it that way, hence this while thing.

View on HN

> We’d just get a bunch of proprietary cores which might not even be compatible with each other due to extensions.

No different than ARM. Apple has matrix extensions that others don't, for example.

The ecosystem (e.g., Linux and OSS) pressure will strongly encourage compatible subsets however. There is some concern about RISCV fragmentation hell that ARM used to suffer from, but a least in the Linux-capable CPU space (e.g., not highly embedded or tiny), a plethora of incompatible cores will probably not happen.

> Companies like Qualcomm would have zero incentives to share their designs with anyone.

ARM cores are also proprietary. All ARM cores actually, you can't get an architectural license from ARM to create an open source core. With RISCV at least you can make open cores and there are some out there.

But opening the ISA is attacking a different level of the stack than the logic and silicon.

> RISCV fragmentation hell that ARM used to suffer from

Does "fragmentation hell" refer to Age Old problem of massive incompatibility in the Linux codebase, or the more "modern" problem people refer to which is the problem of device trees and old kernel support for peripheral drivers? Because you can rest assured that this majestic process will not be changed at all for RISC-V devices. You will still be fed plenty of junkware that will require an unsupported kernels with blobs. The ISA isn't going to change the strategy of the hardware manufacturers.

Neither. It refers to proliferation of incompatible extensions that plagued ARM. Most well known one was hf incompatibility, but there were others too.

> No different than ARM

Everyone has more or less the same access to relatively competitive Cortex and Neoverse cores. As ARM’s finances show that’s not a very good business model so it’s unlikely anyone would do that with RISC-V.

You can make opensource cores, but nobody investing massive amounts of money/resources required to design high end CPUs will make them open source. The situation with ARM is of course not ideal but at least the playing field is somewhat more even.

> No different than ARM. Apple has matrix extensions that others don't, for example.

Not anymore, M4 supports ARM SME instead.

> Companies like Qualcomm would have zero incentives to share their designs with anyone.

And yet, that's what linux did in 1991- they shared the code, lowering the cost of buying an operating system. I wouldn't say there is zero incentive, but it certainly lowers the incentive without a profitable complementary hardware implementation that can be sold for less than the proprietary isa when there is a royalty free license allows the manufacturer/fab designer "mask rights" to get a small margin within the difference of the foundry/ISA proprietary core competitor.

Hardware is not software. Almost nothing alike. Surr it might happen for low-end/old chips with low margins but nothing cutting edge or competitive on the high-end

software used to be hardware limited, and therefore efficient. Today Software relies on many more transistors per joule to compute a number of operations in high-level languages. I'd agree 22nm isn't leading edge, but foundries like Skywater could theoretically provide more options at 65nm and 90nm in coming years that are fully open source, except for the cost of the foundry technique perhaps: https://riscv.org/wp-content/uploads/2018/07/1130-19.07.18-L...

Yes, I think we might be talking about slightly different things. I don’t really see OS model working for higher-end / leading-edge chips.

In segments where chips are almost an interchangeable commodity and the R&D cost might be insignificant relative to manufacturing/foundry it would make a lot more sense.

Even the base integer instructions are Turing complete in RISC-V. Only instruction extensions that could be a point of contention are Matrix operations, as T-Head and Tenstorrent already have their own. Even then, I can't find a reason how this "clash" is any different than those in the x64 or ARM space.

Even if Qualcomm makes their own RISC-V chips that are somehow incompatible with everyone else's, they can't advertise that it's RISC-V due to the branding guidelines. They should know them because they are on the board as a founding top tier member.

> they can't advertise that it's RISC-V due to the branding guidelines

Unless it’s a superset of RISC-V. They can still have proprietary extensions

Reply on news.ycombinator.com