Hacker News Reader

Hacker News

Comment by baq | Hacker News Reader

Parent Full thread

baq

Is there anyone to pick up the pieces?

Apple: obviously not. but then, ARM charges an arm and a leg and they have a history of architecture swaps...

Amazon: with Graviton, very doubtful

Intel: um

AMD: see Intel

Microsoft/Meta/Alphabet: ...maybe? If they started today, they could have a reasonable product by 2030

ARM: 1% probability of a skunkworks product to have the best risc-v CPU once they're something people actually want

NVidia: wanted to buy ARM, announced ARM CPUs like yesterday - but why not, except that people don't care?

Qualcom: most likely option...?

View on HN

Alibaba's already taking the lead on RISC-V, they've even got quad core 2.6 GHz RISC-V laptops: https://www.notebookcheck.net/New-spec-details-and-pricing-f...

By the time Roma shipped it had a 1.5 GHz quad core Chinese SoC with SiFive cores (JH7110), not the Alibaba THead TH1520 SoC which turned out at only 1.85 GHz anyway.

You can buy the same JH7110 SoC in a tablet with docking keyboard&trackpad cover (so pretty much a laptop) for $160 from Pine64 (PineTab-V).

$1499??

I would imagine the target market for these is "people who want a RISC-V development machine" at which point the price to performance ratio doesn't need to be as good for it to still be useful for the potential customers.

That or they don't expect to sell a meaningful number and this one has been productised as a public proof of capability exercise rather than primarily for the sake of the product itself.

It's dropped significantly since then.

Has anyone actually got one of those?

Intel seems to have quite a bit of interest in RISC-V.

They even have a pipecleaner chip using a SiFive core with all the rest being Intel IP.

RISC-V could be their ticket into the phone space.

Finally, Intel knows x86 sucks. They've tried FOUR different times to move off of x86 (iAPX 432, i960, i860, and EPIC). Every one has failed spectacularly.

RISC-V offers an offramp. Patent all the best methods of combining x86 and RISC-V on a chip the release your backward compatible RISC-V + x86 chip. Businesses will buy that chip for decades to remain backward compatible and your patents will guarantee you another 20 years of profits.

EPYC was AMD, Intel's x86 successor was called EPIC.

I have no idea why it came out as EPYC. I fixed it. Thanks!

> Apple: obviously not. but then, ARM charges an arm and a leg and they have a history of architecture swaps...

Given Apple’s very special relationship with ARM, and their culture of control, and them literally being a flagship standard bearer on the performance segment I’d be shocked if they paid anywhere near market rate (not that their licensing is even market-available but you get what I’m saying).

I imagine Apple has an ironclad contract with ARM that likes of which everyone else would be jelly. Even some of the internal architecture changes have long been forbidden by ARM licensing.

Not to mention, ARM is about to destroy the market by trying to backtrack on all their licensing. They now want per device royalties paid by the device manufacturer instead of chip maker to get a larger slice of the pie. And they are basically moving to kill Qualcomm's licenses and IP out of spite in a lawsuit.

Do apple even pay licencing fees?

I thought they could basically do whatever they wanted with the instruction set.

They have a lot of freedom which nobody else has access to, but according to arm’s SEC fillings they are a paying client

Apple pays nothing, because they do not license ARM cores in any way, instead having a full royalty-free license for ARM ISA (and afaik covering updates too) since before ARM got big. Samsung used to have one, no idea if they still have one, but the mismanagement of the unit that built their custom ARM designs might have killed the whole venture.

Apple does not pay nothing. Read recent pre-IPO Arm SEC filing (not a lot of details, but enough to reveal they clearly aren’t paying nothing).

The rumor is that Apple owns part of AArch64 itself, and it's relationship with ARM is more like the Intel/AMD relationship.

That would mean that at a minimum they still pay for some of the Cortex M cores that are in their SoCs.

Ah, good source to check, thanks.

This isn’t true. Apple paid licence fees for the very first ARM Ltd processors - ARM would have had no income if they hadn’t.

And there is no way that any company gets a ‘forever’ and ‘all future IP’ license just because they had an early shareholding.

The "nothing" probably refers to royalties which is true for architectural licensees like Apple. They instead pay for that license which is not sold per-CPU.

It was very clear that they meant nothing at all.

> Apple pays nothing, because they do not license ARM cores in any way, instead having a full royalty-free license for ARM ISA (and afaik covering updates too) since before ARM got big.

I will admit to being a bit hyperbolic there. Compared to normal ARM licensing, it's way less.

Meanwhile I recall start of RISC-V hype being, among other things, Western Digital dropping a ton of investment money into it just to escape ARM license costs.

We all get a bit hyperbolic at times! Yes, I'm sure it's a lot less than most customers pay.

I remember the WD announcement. They have open sourced the cores now I think. If you're shipping millions of drives those fees will add up.

For values of "now" equal to 2019

https://github.com/westerndigitalcorporation/swerv_eh1

> ARM charges an arm and a leg and they have a history of architecture swaps...

Do they? I don't think the details of their contract with ARM are public so they might have a very good deal. Also licensing fees are probably a much smaller issue for companies which have very high margins.

But what would they be buying? The ISA is free. A few designs without the engineering teams that built them? Doesn’t seem that attractive.

There's more to a chip than the ISA.

I think Nvidia. they wanted to buy arm because they didn't want to depend on external people for their architecture. once that is blocked going to riscv makes sense for them (recent arm chip announcements aren't that relevant to this analysis because their design was before the arm deal was blocked)

> Is there anyone to pick up the pieces?

I do not know if it matters to the RISC-V ecosystem if someone picks up the pieces, but for the employees involved it does.

Google is bringing Android to RISC-V. They're quite capable of designing their own chips and even have their own phones to put them in, and chromebooks too. Not that they're going to change the world by doing so.

How about TI? Microchip? Even Siemens or Bosch?

Microchip seem like a likely buyer to me. They have been buying up various silicon semi-specialty companies like mad over the past 10ish years and they already have some RISC-V SOCs for sale.

If Microchip could offer a semi-custom chip design business leveraging what SiFive have done, with reasonable minimum order quantities, I think that'd be quite interesting. But I doubt Microchip will pay a billion dollars for such a purchase, they could also probably just go hire away the key engineers for a lot less.

MicroSemi (which was later bought by MicroChip) had a relationship with SiFive and you could buy a MicroSemi PolarFire FPGA dev board with a preconfigured SiFive U74 soft-core in it. DAMN if it wasn't difficult to configure.

Which is my way of saying... yeah... Microchip buying SiFive? Stranger things have happened.

Reply on news.ycombinator.com