4 ms·

Not really sure what’s the advantage of switching from whatever you have today (ARM?), to a qualcomm IP. Qualcomm is historycally a quite shitty partner to have

Not really sure what’s the advantage of switching from whatever you have today (ARM?), to a qualcomm IP. Qualcomm is historycally a quite shitty partner to have. Nobody is going to give the secret sauce for free. Also these AI/ML use cases are just some marketing bs, not really grounded by anything really. There are plenty of ISAs you can find freely available today (Open sparc anyone?). The problem is fragmentation…

ndsipa_pomu 3y ago
RISC-V is not tied to Qualcomm, so you can either choose a different partner or do it all yourself if you've got the necessary skills and resources. > This open ISA provides designers and developers with greater control over their computing environments, allowing them to fine-tune their systems without relying on third parties or incurring additional license fees associated with proprietary architectures.
- squarefoot 3y ago
  Also, having competition in the field benefits everyone, including those who wouldn't switch anyway.
- hashtag-til 3y ago
  That’s very aspirational, but hardly backed by any reference these days.
  - Moto7451 3y ago
    Agreed, especially given their position in ARM, a ubiquitous and easy to license IP. The lock in will come from the GPU, accelerator/coprocessor, and other IPs that will lead to Qualcomm/other large vendor lock in.
  - ndsipa_pomu 3y ago
    I don't follow what you mean. You can purchase SiFive RISC-V chips.
    - themiddleupper 3y ago
      Are they open source?
      - ndsipa_pomu 3y ago
        I believe so. https://www.sifive.com/press/sifive-launches-industrys-first-open-source-risc-v-soc https://www.sifive.com/press/sifive-launches-industrys-first... https://github.com/sifive https://github.com/sifive
      - jlokier 3y ago
        I believe SiFive's CPUs and chip designs are not open source generally, although they do open source a lot of related things such as software, specifications, tools. I looked at their GitHub and didn't find any CPUs in there, though there are a lot of repos. Here's an interesting take on what someone thinks "open source" means in this context: gjsman-1000 wrote: Open source refers solely to the software side of things in these groups and types of projects. Thus, all of the drivers and software you need to use the SiFive is open source and thus you can say it is an open source design. However, it is not an "open hardware" design in that the IP used to design the chip is not released. https://news.ycombinator.com/item?id=25762114 https://news.ycombinator.com/item?id=25762114 I personally don't think of closed chip designs with open software and specifications as open source chips. That would be like saying Intel and AMD x86s are open source. Though to be fair to those companies, their CPU programming models are much more open than their FPGA programming models, which requires proprietary software and secret specifications to program them at all.
- Geezus_42 3y ago
  Yes, but how long until Qualcomm, and whoever else, builds proprietary add-on instruction sets that become SOB and lock people in?
  - rwmj 3y ago
    It's entirely the point of the RISC-V custom extension space that Qualcomm are allowed to do this. It's certainly a risk that software could get locked in to using these custom extensions, but it's also very easy to tell if software is using those extensions from simple examination of the opcodes. For those interested, check the bottom byte[*] of each instruction for: x000 10xx custom-0 x010 10xx custom-1 x101 10xx custom-2 x111 10xx custom-3 [*] Where "bottom byte" is actually the first byte since instructions are always stored little endian. Edit: To be clear and avoid confusion, RISC-V has two sorts of extensions, the custom vendor-specific extension space that I'm talking about here, and the RVI standard extensions which I'm not talking about.
    - znpy 3y ago
      > software could get locked in to using these custom extensions the question arises: can such extensions be patented? we've seen already that APIs cannot get patented (see the Oracle vs Google litigation). If Qualcomm was to develop a set of extension, but other players in the risc-v space could implement some equivalent that's ISA/binary compatible... I wouldn't see any issue with that. Am I being too optimistic ?
      - colejohnson66 3y ago
        IANAL, but yes, ISA extensions can be patented. For example, the x86-64 extension ("Long Mode") was created by AMD, and Intel had to pay to license it.[0] Those patents expired a few years ago, allowing Microsoft and Apple to write x86-64 to ARM translators freely. [0]: https://en.wikipedia.org/wiki/X86-64#Licensing https://en.wikipedia.org/wiki/X86-64#Licensing
      - rwmj 3y ago
        RISC-V has two sorts of extensions. The custom extensions I mentioned in my post above are vendor-specific and they can do what they want, patent them, keep them secret, whatever. However the standard extensions which go through ratification by RISC-V International and are meant to be implemented by anyone must have a royalty-free patent grant (see: https://riscv.org/wp-content/uploads/2022/03/RISC-V-Amended-Internal-Regulations-APPROVED-2022-Feb-17.pdf https://riscv.org/wp-content/uploads/2022/03/RISC-V-Amended-...)
- fidotron 3y ago
  In fairness using Qualcomm for purposes other than mobile modem integration is a niche pursuit for ARM windows machines and Meta Quest VR. Their CPU implementations never recovered from the 64 bit transition that wrong footed them completely. As the parent alluded to, if there is a problem in the Android ecosystem you will almost certainly find either Qualcomm or Google in the middle of it.
- ilyt 3y ago
  > RISC-V is not tied to Qualcomm, so you can either choose a different partner or do it all yourself if you've got the necessary skills and resources. ISA is the least important part in CPU. The hard part is having all drivers for the peripherals given chip has and knowing most of the chinese vendors it's straight to landfill after few years unless something becomes so popular community RE it.
  - ndsipa_pomu 3y ago
    > ISA is the least important part in CPU It is essential though, so it's an important step towards having entirely open CPUs
    - ilyt 3y ago
      Proprietary ISA with open peripherals would be FAR better than other way around. Frankly the ISA part is easiest to change from OS kernel perspective. If "open" ISA comes with a bunch of blobs it's essentially exactly same situation as using ARM chip.
      - ndsipa_pomu 3y ago
        There's a HN link to https://pineapple-one.github.io/ https://pineapple-one.github.io/ on the front-page now. Is that of interest?
TradingPlaces 3y ago
The advantage is that ARM is trying to change their license and royalty structure and are currently suing Qualcomm. https://www.courtlistener.com/docket/64938776/arm-ltd-v-qualcomm-inc/ https://www.courtlistener.com/docket/64938776/arm-ltd-v-qual...
- hashtag-til 3y ago
  You mean advantage for who? That their quarrel, not a trend.
  - rwmj 3y ago
    If ARM are going to attack Qualcomm, who else will they go for? Particularly since they now need to justify their very high stock price. With RISC-V you have a clear license (BSD) which lets you do what you want, now and in the future.
    - klelatti 3y ago
      Qualcomm took the first step in this case by trying to sell Nuvia derived designs under their own license agreement. Whoever wins it’s not really Arm going on the ‘attack’ if there is a reasonable case that the license has been breached.
      - TradingPlaces 3y ago
        We have no visibility into either contract at the center of this case. I’m holding off judgement until we understand Qualcomm’s architecture license, and the assignment clause in Nuvia's. I will only say that suing a 10% customer is, generally speaking, a terrible idea.
        klelatti 3y ago
        > a terrible idea Well you’re not really holding off judgment then! It’s also a terrible idea to let big customers breach their contracts but as you say we haven’t seen the contracts concerned.
phkahler 3y ago
>> There are plenty of ISAs you can find freely available today (Open sparc anyone?). The problem is fragmentation… If by "fragmentation" you mean lack of industry-wide support, then you've answered your own question as to why RISC-V. Most Debian packages can run on it. There is even an Android port to RISC-V. Outside x86 and ARM, the next best supported ISA is RISC-V. Unrelated: why do they call the instruction set an architecture or ISA?
- acdha 3y ago
  Strictly speaking an instruction set doesn’t include things like the memory model which you need to know to be able to port code. For example, Apple had to add an extension to their ARM chips which makes stronger guarantees about the order in which memory is written because x86 Mac apps might have dependencies on that which their developers hadn’t been aware of.
  - snvzz 3y ago
    >Strictly speaking an instruction set doesn’t include things like the memory model which you need to know to be able to port code. Very strictly speaking, to the point where it's true but irrelevant. RISC-V does have specifications for memory models: RVWMO and Ztso. They are both ratified.
    - acdha 3y ago
      Oh, apologies if it sounded like I was trying to say RISC-V doesn't - I was only trying to explain why we use the term “instruction set architecture” as opposed to just “instruction set”.
      - snvzz 3y ago
        Because the parent was talking about fragmentation, it does not hurt to ensure it is understood that the memory model _is_ standardized in RISC-V.
- brucehoult 3y ago
dkjaudyeqooe 3y ago
ARM is no saint either. The problem is that when you have total control over the key IP at some point you become prone to extracting more money out of your customers through various forms of coercion. The advantage of moving to RISC-V is that no one owns it and you can always leave your IP vendor for another one. Because RISC-V is an attractive ISA it's having an anti-fragmentation effect, creating a popular standard and an industry around it.

ndsipa_pomu 3y ago

RISC-V is not tied to Qualcomm, so you can either choose a different partner or do it all yourself if you've got the necessary skills and resources. > This open ISA provides designers and developers with greater control over their computing environments, allowing them to fine-tune their systems without relying on third parties or incurring additional license fees associated with proprietary architectures.

squarefoot 3y ago
Also, having competition in the field benefits everyone, including those who wouldn't switch anyway.
hashtag-til 3y ago
That’s very aspirational, but hardly backed by any reference these days.
- Moto7451 3y ago
  Agreed, especially given their position in ARM, a ubiquitous and easy to license IP. The lock in will come from the GPU, accelerator/coprocessor, and other IPs that will lead to Qualcomm/other large vendor lock in.
- ndsipa_pomu 3y ago
  I don't follow what you mean. You can purchase SiFive RISC-V chips.
  - themiddleupper 3y ago
    Are they open source?
    - ndsipa_pomu 3y ago
      I believe so. https://www.sifive.com/press/sifive-launches-industrys-first-open-source-risc-v-soc https://www.sifive.com/press/sifive-launches-industrys-first... https://github.com/sifive https://github.com/sifive
    - jlokier 3y ago
      I believe SiFive's CPUs and chip designs are not open source generally, although they do open source a lot of related things such as software, specifications, tools. I looked at their GitHub and didn't find any CPUs in there, though there are a lot of repos. Here's an interesting take on what someone thinks "open source" means in this context: gjsman-1000 wrote: Open source refers solely to the software side of things in these groups and types of projects. Thus, all of the drivers and software you need to use the SiFive is open source and thus you can say it is an open source design. However, it is not an "open hardware" design in that the IP used to design the chip is not released. https://news.ycombinator.com/item?id=25762114 https://news.ycombinator.com/item?id=25762114 I personally don't think of closed chip designs with open software and specifications as open source chips. That would be like saying Intel and AMD x86s are open source. Though to be fair to those companies, their CPU programming models are much more open than their FPGA programming models, which requires proprietary software and secret specifications to program them at all.
Geezus_42 3y ago
Yes, but how long until Qualcomm, and whoever else, builds proprietary add-on instruction sets that become SOB and lock people in?
- rwmj 3y ago
  It's entirely the point of the RISC-V custom extension space that Qualcomm are allowed to do this. It's certainly a risk that software could get locked in to using these custom extensions, but it's also very easy to tell if software is using those extensions from simple examination of the opcodes. For those interested, check the bottom byte[*] of each instruction for: x000 10xx custom-0 x010 10xx custom-1 x101 10xx custom-2 x111 10xx custom-3 [*] Where "bottom byte" is actually the first byte since instructions are always stored little endian. Edit: To be clear and avoid confusion, RISC-V has two sorts of extensions, the custom vendor-specific extension space that I'm talking about here, and the RVI standard extensions which I'm not talking about.
  - znpy 3y ago
    > software could get locked in to using these custom extensions the question arises: can such extensions be patented? we've seen already that APIs cannot get patented (see the Oracle vs Google litigation). If Qualcomm was to develop a set of extension, but other players in the risc-v space could implement some equivalent that's ISA/binary compatible... I wouldn't see any issue with that. Am I being too optimistic ?
    - colejohnson66 3y ago
      IANAL, but yes, ISA extensions can be patented. For example, the x86-64 extension ("Long Mode") was created by AMD, and Intel had to pay to license it.[0] Those patents expired a few years ago, allowing Microsoft and Apple to write x86-64 to ARM translators freely. [0]: https://en.wikipedia.org/wiki/X86-64#Licensing https://en.wikipedia.org/wiki/X86-64#Licensing
    - rwmj 3y ago
      RISC-V has two sorts of extensions. The custom extensions I mentioned in my post above are vendor-specific and they can do what they want, patent them, keep them secret, whatever. However the standard extensions which go through ratification by RISC-V International and are meant to be implemented by anyone must have a royalty-free patent grant (see: https://riscv.org/wp-content/uploads/2022/03/RISC-V-Amended-Internal-Regulations-APPROVED-2022-Feb-17.pdf https://riscv.org/wp-content/uploads/2022/03/RISC-V-Amended-...)
fidotron 3y ago
In fairness using Qualcomm for purposes other than mobile modem integration is a niche pursuit for ARM windows machines and Meta Quest VR. Their CPU implementations never recovered from the 64 bit transition that wrong footed them completely. As the parent alluded to, if there is a problem in the Android ecosystem you will almost certainly find either Qualcomm or Google in the middle of it.
ilyt 3y ago
> RISC-V is not tied to Qualcomm, so you can either choose a different partner or do it all yourself if you've got the necessary skills and resources. ISA is the least important part in CPU. The hard part is having all drivers for the peripherals given chip has and knowing most of the chinese vendors it's straight to landfill after few years unless something becomes so popular community RE it.
- ndsipa_pomu 3y ago
  > ISA is the least important part in CPU It is essential though, so it's an important step towards having entirely open CPUs
  - ilyt 3y ago
    Proprietary ISA with open peripherals would be FAR better than other way around. Frankly the ISA part is easiest to change from OS kernel perspective. If "open" ISA comes with a bunch of blobs it's essentially exactly same situation as using ARM chip.
    - ndsipa_pomu 3y ago
      There's a HN link to https://pineapple-one.github.io/ https://pineapple-one.github.io/ on the front-page now. Is that of interest?

TradingPlaces 3y ago

The advantage is that ARM is trying to change their license and royalty structure and are currently suing Qualcomm. https://www.courtlistener.com/docket/64938776/arm-ltd-v-qualcomm-inc/ https://www.courtlistener.com/docket/64938776/arm-ltd-v-qual...

hashtag-til 3y ago
You mean advantage for who? That their quarrel, not a trend.
- rwmj 3y ago
  If ARM are going to attack Qualcomm, who else will they go for? Particularly since they now need to justify their very high stock price. With RISC-V you have a clear license (BSD) which lets you do what you want, now and in the future.
  - klelatti 3y ago
    Qualcomm took the first step in this case by trying to sell Nuvia derived designs under their own license agreement. Whoever wins it’s not really Arm going on the ‘attack’ if there is a reasonable case that the license has been breached.
    - TradingPlaces 3y ago
      We have no visibility into either contract at the center of this case. I’m holding off judgement until we understand Qualcomm’s architecture license, and the assignment clause in Nuvia's. I will only say that suing a 10% customer is, generally speaking, a terrible idea.
      - klelatti 3y ago
        > a terrible idea Well you’re not really holding off judgment then! It’s also a terrible idea to let big customers breach their contracts but as you say we haven’t seen the contracts concerned.

phkahler 3y ago

>> There are plenty of ISAs you can find freely available today (Open sparc anyone?). The problem is fragmentation… If by "fragmentation" you mean lack of industry-wide support, then you've answered your own question as to why RISC-V. Most Debian packages can run on it. There is even an Android port to RISC-V. Outside x86 and ARM, the next best supported ISA is RISC-V. Unrelated: why do they call the instruction set an architecture or ISA?

acdha 3y ago
Strictly speaking an instruction set doesn’t include things like the memory model which you need to know to be able to port code. For example, Apple had to add an extension to their ARM chips which makes stronger guarantees about the order in which memory is written because x86 Mac apps might have dependencies on that which their developers hadn’t been aware of.
- snvzz 3y ago
  >Strictly speaking an instruction set doesn’t include things like the memory model which you need to know to be able to port code. Very strictly speaking, to the point where it's true but irrelevant. RISC-V does have specifications for memory models: RVWMO and Ztso. They are both ratified.
  - acdha 3y ago
    Oh, apologies if it sounded like I was trying to say RISC-V doesn't - I was only trying to explain why we use the term “instruction set architecture” as opposed to just “instruction set”.
    - snvzz 3y ago
      Because the parent was talking about fragmentation, it does not hurt to ensure it is understood that the memory model _is_ standardized in RISC-V.
brucehoult 3y ago

dkjaudyeqooe 3y ago

ARM is no saint either. The problem is that when you have total control over the key IP at some point you become prone to extracting more money out of your customers through various forms of coercion. The advantage of moving to RISC-V is that no one owns it and you can always leave your IP vendor for another one. Because RISC-V is an attractive ISA it's having an anti-fragmentation effect, creating a popular standard and an industry around it.