Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Arm is canceling Qualcomm's chip design license

bloomberg.com

Post reply on HN
Print view View original

341–350 of 480 posts

Re: Arm is canceling Qualcomm's chip design license

#341

Post by lizknope »

haerwu wrote: post #192
lizknope wrote: post #69

Earlier quoted context omitted.

Snapdragon 805 had a 32-bit Krait designed by Qualcomm https://www.qualcomm.com/products/mobile/snapdragon/smartpho... 810 had a 64-bit core designed by ARM https://www.qualcomm.com/products/mobile/snapdragon/smartpho... 820/821 had a 64-bit Kryo custom core designed by Qualcomm https://www.qualcomm.com/products/mobile/snapdragon/smartpho... After that it was all cores from ARM. The custom CPU team worked on their se…

When you look at https://gpages.juszkiewicz.com.pl/arm-socs-table/arm-socs.ht... table then you can see that in Snapdragon 8xx series the first "ARM cores out of shelf" was 888 in 2020. 865 (2019) has Cortex-A77 + Kryo 4xx Silver 888 (2002) uses Cortex-X1 + Cortex-A78 + Cortex-A55 cores

That table doesn't have the Snapdragon 810

https://www.qualcomm.com/products/mobile/snapdragon/smartpho...

I worked on it in 2014. The table does have 808 listed. That may have been a lower end version.

Qualcomm got caught being late. They were continuing development of custom 32-bit cores and Apple came out with a 64-bit ARM core in the iPhone. The Chief Marketing Officer of Qualcomm called it a gimmick but Apple was a huge customer of Qualcomm's modems. Qualcomm shoved him off to the side for a while.

https://www.cnet.com/tech/mobile/qualcomm-gambit-apple-64-bi...

Because Q's custom 64-bit CPU was not ready the stop gap plan was to license a 64-bit RTL design from ARM and use that in the 810. It also had overheating problems but that's different issue. There were a lot of internal politics going on at Q over the custom cores and server chips that ended up in layoffs.

Re: Arm is canceling Qualcomm's chip design license

#342

Post by adrian_b »

dagmx wrote: post #306
adrian_b wrote: post #233

Earlier quoted context omitted.

Now Qualcomm does not want to continue to use any Arm cores, both because their own cores are better and because that would save them the cost for royalties. Obviously, Arm tries to prevent Qualcomm from using their own cores, because this time Arm would lose a major source of their revenue if Qualcomm stopped licensing cores. When Arm has given architectural licenses to Qualcomm and Nuvia, they were not worried abou…

There’s a LOT of conjecture in your comment. Your very first line is one to begin with. ARM also doesn’t seem to care if QC design their own cores. They just care that they renegotiate the royalty agreement. This is clear if you actually read their statements.

Because Arm wants to increase the royalties for the cores designed by Qualcomm, that is a pretty certain indication that these royalties have been smaller than for the cores licensed from Arm.

Therefore Arm cares a lot if Qualcomm designs their own cores, because that would cause a smaller revenue for Arm.

If Arm had not cared whether Qualcomm designs their own cores, they would have never sued Qualcomm.

The official reason why Arm has sued Qualcomm, is not for increasing the royalties, because that has no legal basis.

It is obvious that the lawsuit is just a blackmail instrument to force Qualcomm to pay higher royalties for the cores designed by them, but the official object of the lawsuit is to forbid Qualcomm to design their own cores, by claiming that the Oryon cores used in the new Qualcomm chipsets for laptops, smartphones and automotive applications have been designed by violating the conditions of the architectural licenses granted by Arm to Qualcomm and Nuvia, so Arm requests that Qualcomm must stop making any products with these Arm-compatible cores and they must destroy all their existing core designs.

Re: Arm is canceling Qualcomm's chip design license

#343

Post by monocasa »

bhouston wrote: post #289
snvzz wrote: post #236

Earlier quoted context omitted.

The high end requires specifications that were not available until RVA22 and Vector 1.0 were ratified. First chips implementing these are starting to show up as seen in e.g. MILK-V Jupiter, which is one of the newest development boards in the market. With the ISA developed in the open, the base specs microcontrollers can target would naturally tend to be be ratified first, and thus microcontrollers would show up firs…

Isn't the problem the lack of advanced features for executing the current ISA with speed? I thought RISC-V chips seen in the wild do not pipelining, out-of-order, register renaming, multiple int/float/logic units, speculation, branch predictors, multi-tier caching, etc. The lack of speed isn't really related to a few missing vector instructions.

There's a lot of cores that do all of that.

Most cores are pipelined; it is RISC after all.

There are quite a few superscalar cores, even a c906 is superscalar.

The c910/c920 is an OoO, renaming core, with speculation.

What they're lacking is area and power. A ROB with six entries is not going to compete with a ROB of six hundred entries.

Re: Arm is canceling Qualcomm's chip design license

#344

Post by shash »

torginus wrote: post #295
shash wrote: post #243

Earlier quoted context omitted.

Well, Tenstorrent, Andes and others have their respective designs... On the in-order side, I can see on-par performance with the ARM A5x series quite easily.

After a bit of digging, I found that the SiFive P670 has performance equivalent to the Galaxy S21, or the desktop Ryzen 7 2700, which is not too bad and definitely usable in a smartphone/laptop form, so competitive with the 2021 era designs. Definitely usable. It's not clear what's the power level is.

The P670 is a core, not a chip, so you can't really get to power numbers (or indeed, raw performance as opposed to performance / watt or performance / GHz) without actually making a practical chip out of it in some given process node. You're better off comparing it to a core, such as the ARM Cortex series, rather than the S21.

SiFive claims a SPECint2006 score of > 12/GHz, meaning that it'll get a performance of about 24 at 2 GHz or ~31 at 2.6 GHz, making it on par with an A76 in terms of raw performance.

Re: Arm is canceling Qualcomm's chip design license

#345

Post by shash »

cesarb wrote: post #301
shash wrote: post #238

Earlier quoted context omitted.

> Virtually all high performance processors these days operate on their own internal “instructions”. The instruction decoder at the very front of the pipeline that actually sees ARM or RISC-V or whatever is a relatively small piece of logic. You're talking essentially about microcode; this has been the case for decades, and isn't some new development. However, as others have pointed out, it's not _as_ simple as just…

> > Virtually all high performance processors these days operate on their own internal “instructions”. The instruction decoder at the very front of the pipeline that actually sees ARM or RISC-V or whatever is a relatively small piece of logic. > You're talking essentially about microcode; this has been the case for decades, and isn't some new development. Microcode is much less used nowadays than in the past. For ins…

> Many details of an ISA extend beyond the instruction decoder. For instance, the RISC-V ISA mandates specific behavior for its integer division instruction, which has to return a specific value on division by zero, unlike most other ISAs which trap on division by zero; and the NaN-boxing scheme it uses for single-precision floating point in double-precision registers can be found AFAIK nowhere else. The x86 ISA is infamous for having a stronger memory ordering than other common ISAs. Many ISAs have a flags register, which can be set by most arithmetic (and some non-arithmetic) instructions. And that's all for the least-privileged mode; the supervisor or hypervisor modes expose many more details which differ greatly depending on the ISA.

All quite true, and to that, add things like cache hints and other hairy bits in an actual processor.

Re: Arm is canceling Qualcomm's chip design license

#346

Post by shash »

adrian_b wrote: post #231
SushiHippie wrote: post #178

Because of all the discussions in the comments about ARM and RISC-V, could someone explain to me the difficulties of designing a chip for a new ISA? I'm wondering because to me as a layman it sounds like it's 'only' a different language, so why is it not that easy to take already existing designs and modify them to 'speak' that language and that's it? Or is an ISA more than just a different 'language'? Or is hardware…

Designing a new clean and better ISA is easy. With the current existing experience that could be done in weeks or a few months at most, to allow for the simulation of how various variants run benchmarks, to determine the best variant. Nevertheless, designing the ISA is the only easy part. Then you have to write high-quality compilers, assemblers, linkers, debuggers and various other tools, and also good documentation…

Interesting to see RISC-V described as an "incumbent" :D

Re: Arm is canceling Qualcomm's chip design license

#347

Post by avvvv »

bhouston wrote: post #255
hajile wrote: post #41

Earlier quoted context omitted.

We're not quite there yet. A bunch of mission critical stuff like SIMD were only added in the last 2-3 years. As it takes 4-5 years to design/ship high-performance chips, we still have a ways to go. Ventana Veyron looks interesting. Tenstorrent's upcoming 8-wide design should perform well. Qualcomm pitched making a bunch of changes to RISC-V that would move it closer to ARM64 and make porting easier, so I think it's…

It is much more than just SIMD. I think RISC-V chips in the wild do not do things like pipelining, out-of-order, register renaming, multiple int/float/logic units, speculation, branch predictors, smart caching. I think all existing RISC-V chips in the wild right now are just simplistic in-order processors.

Could patents be one of the reason why? Genuine question.

Re: Arm is canceling Qualcomm's chip design license

#348

Post by jjtheblunt »

hajile wrote: post #52
mushufasa wrote: post #36

Qualcomm is known for having a particularly aggressive & hardball-style legal department to enforce its patents on core telecom IP. I believe the most likely outcome is they just settle the dispute here. Arm fighting hardball with hardball. Which would not really affect the ecosystem of phones using Qualcomm arm chips, it would just change the margins / market cap of Qualcomm. Yes, longterm Q might invest in their ow…

Qualcomm doesn't have nearly as much to lose as ARM does and they know it. Qualcomm is almost certainly ARM's biggest customer. If ARM loses, Qualcomm doesn't have to pay out. If ARM wins, Qualcomm moves to RISC-V and ARM loses even harder in the long-term. The most likely outcome is that Qualcomm agrees to pay a slight bit more than they are currently paying, but nowhere near what ARM is demanding and in the meantim…

> Qualcomm is almost certainly ARM's biggest customer

what about Apple?

Re: Arm is canceling Qualcomm's chip design license

#350

Post by shash »

bhouston wrote: post #255
hajile wrote: post #41

Earlier quoted context omitted.

We're not quite there yet. A bunch of mission critical stuff like SIMD were only added in the last 2-3 years. As it takes 4-5 years to design/ship high-performance chips, we still have a ways to go. Ventana Veyron looks interesting. Tenstorrent's upcoming 8-wide design should perform well. Qualcomm pitched making a bunch of changes to RISC-V that would move it closer to ARM64 and make porting easier, so I think it's…

It is much more than just SIMD. I think RISC-V chips in the wild do not do things like pipelining, out-of-order, register renaming, multiple int/float/logic units, speculation, branch predictors, smart caching. I think all existing RISC-V chips in the wild right now are just simplistic in-order processors.

Almost all in-order processors will do pipelining, so that's there. Many are even multi-issue. Andes has an out of order core [1] and so does SiFive (though I don't know of many actual chips using these.

[1] https://www.andestech.com/en/2024/01/05/andes-announces-gene... [2] https://www.sifive.com/cores/performance-p650-670

Post reply on HN