3 ms·

Next week, Qualcomm will likely announce a 64 core RISC-V RVA23. ARM really shouldn't pursue an aggressive posture with lines outside iOS or Win11 ecosystems.

Next week, Qualcomm will likely announce a 64 core RISC-V RVA23. ARM really shouldn't pursue an aggressive posture with lines outside iOS or Win11 ecosystems. The leverage won't hold a position already fractured off a legacy market. =3

chx 2y ago
1. I am fairly sure games and other performance sensitive apps are using the Android NDK which is not available for RISC-V. 2. I am fairly sure a competitive RISC-V CPU is not days or weeks but years away.
- svnt 2y ago
  > 2. I am fairly sure a competitive RISC-V CPU is not days or weeks but years away. And chasing a moving target fueled by the largest technology companies on the planet.
  - Joel_Mckay 2y ago
    Tying products to Googles ecosystem is usually financially risky. Not a good long-term strategy for startups. =3
- Joel_Mckay 2y ago
  I think it is more of a "chicken and egg" ordering problem. 1. The RISC-V design standard fragmentation issue has been addressed. 2. A reasonable mobile class level SoC will be available for integration after any large production run of the chips. If ARM forces #2 out of silliness, than it also accelerates #1 in the market. In general, there is plenty of use-cases even if a chip is not cutting edge. =3
SG- 2y ago
you think they can just flip a RISC-V switch and keep all the performance instantly? I can't really understand the logic from some people here.
- Joel_Mckay 2y ago
  "keep all the performance instantly" Depends what you mean by performance (most vendors ARM accelerated features are never used for compatibility reasons), as upping the core count with simpler architecture is a fair trade on wafer space. i.e. if ARM is using anti-trust tactics to extort more revenue, that budget is 100% fungible with extra resources. Note, silicon is usually much cheaper than IP licenses. One can ask polity to explain things without being rude to people. Have a wonderful day =3
- mrweasel 2y ago
  People also seem to forget that everything needs to be ported. If you're an Android manufacturer, you're not going to stop shipping phones, waiting for the Android RISC-V to catch-up to ARM, or for RISC-V to get the speed and features of current ARM CPUs. You're going to buy ARM processors from another vendor. The Windows ARM port is going to take even longer, I doubt that Microsoft has that working at anything beyond research stage, if that. Getting the RISC-V ecosystem up to par with ARM is going to take years. If you want to spin this in RISC-Vs favor, then yes, forcing a company like Qualcomm to switch would speed things up, but it might also give them a bit of a stranglehold on the platform, in the sense that they'd be the dominant platform, with all of their own customisations.
  - Joel_Mckay 2y ago
    In theory, the Raspberry Pi foundation could easily move 3 million 1.8GHz RVA23 in 1 quarter... with 64 cores + DSP ops it wouldn't necessarily need a GPU initially. =3
    - mrweasel 2y ago
      The Raspberry Pi community would probably jump on a RISC-V board, but that doesn't help Qualcomm or it's customers.
      - Joel_Mckay 2y ago
        Manufacturers adapt quickly to most architectural changes. If you are running in a posix environment, than porting a build is measured in days with a working bootstrap compiler. RISC-V already has the full gcc and OS available for deployment. We also purchase several vendors ARM products for deployment. Note, there was a time in history, when purchasing even a few million in chips would open custom silicon options. Given how glitched/proprietary/nondeterministic ARM ops are outside the core compatibility set, it is amazing it was as popular as the current market demonstrates. Engineers solve problems, and ARM corporation has made themselves a problem. =3
kmeisthax 2y ago
Funnily enough Qualcomm tried to persuade RISC-V to let them drop compressed instructions. Presumably because they're trying to crowbar a RISC-V decoder onto the Nuvia design and compressed instructions are breaking it somehow.
- Joel_Mckay 2y ago
  They should buy the intel alumni founded RISC-V startup, and pour resources into a RVA23 based chip with dual on-chip SDR asic sections (they have the IP). i.e. create a single open-chip solution for mid-tier mobile communication platforms. They won't do this due to their cellular chip line interests. However, even if it just ran a bare bones Linux OS... it would open up entire markets. =3
  - delfinom 2y ago
    With how poorly Intel is doing, not sure "Intel alumni" is a plus. Lmao.
    - Joel_Mckay 2y ago
      Indeed, the Intel installed base has enough market inertia to last a business cycle or two with AMD. Even with the recent silicon defects... People will tolerate the garbage as they want the NVIDIA+Intel performance. Architecturally speaking, there were better options available... just never the equivalent price over performance of consumer grade hardware. =)

chx 2y ago

1. I am fairly sure games and other performance sensitive apps are using the Android NDK which is not available for RISC-V. 2. I am fairly sure a competitive RISC-V CPU is not days or weeks but years away.

svnt 2y ago
> 2. I am fairly sure a competitive RISC-V CPU is not days or weeks but years away. And chasing a moving target fueled by the largest technology companies on the planet.
- Joel_Mckay 2y ago
  Tying products to Googles ecosystem is usually financially risky. Not a good long-term strategy for startups. =3
Joel_Mckay 2y ago
I think it is more of a "chicken and egg" ordering problem. 1. The RISC-V design standard fragmentation issue has been addressed. 2. A reasonable mobile class level SoC will be available for integration after any large production run of the chips. If ARM forces #2 out of silliness, than it also accelerates #1 in the market. In general, there is plenty of use-cases even if a chip is not cutting edge. =3

SG- 2y ago

you think they can just flip a RISC-V switch and keep all the performance instantly? I can't really understand the logic from some people here.

Joel_Mckay 2y ago
"keep all the performance instantly" Depends what you mean by performance (most vendors ARM accelerated features are never used for compatibility reasons), as upping the core count with simpler architecture is a fair trade on wafer space. i.e. if ARM is using anti-trust tactics to extort more revenue, that budget is 100% fungible with extra resources. Note, silicon is usually much cheaper than IP licenses. One can ask polity to explain things without being rude to people. Have a wonderful day =3
mrweasel 2y ago
People also seem to forget that everything needs to be ported. If you're an Android manufacturer, you're not going to stop shipping phones, waiting for the Android RISC-V to catch-up to ARM, or for RISC-V to get the speed and features of current ARM CPUs. You're going to buy ARM processors from another vendor. The Windows ARM port is going to take even longer, I doubt that Microsoft has that working at anything beyond research stage, if that. Getting the RISC-V ecosystem up to par with ARM is going to take years. If you want to spin this in RISC-Vs favor, then yes, forcing a company like Qualcomm to switch would speed things up, but it might also give them a bit of a stranglehold on the platform, in the sense that they'd be the dominant platform, with all of their own customisations.
- Joel_Mckay 2y ago
  In theory, the Raspberry Pi foundation could easily move 3 million 1.8GHz RVA23 in 1 quarter... with 64 cores + DSP ops it wouldn't necessarily need a GPU initially. =3
  - mrweasel 2y ago
    The Raspberry Pi community would probably jump on a RISC-V board, but that doesn't help Qualcomm or it's customers.
    - Joel_Mckay 2y ago
      Manufacturers adapt quickly to most architectural changes. If you are running in a posix environment, than porting a build is measured in days with a working bootstrap compiler. RISC-V already has the full gcc and OS available for deployment. We also purchase several vendors ARM products for deployment. Note, there was a time in history, when purchasing even a few million in chips would open custom silicon options. Given how glitched/proprietary/nondeterministic ARM ops are outside the core compatibility set, it is amazing it was as popular as the current market demonstrates. Engineers solve problems, and ARM corporation has made themselves a problem. =3

kmeisthax 2y ago

Funnily enough Qualcomm tried to persuade RISC-V to let them drop compressed instructions. Presumably because they're trying to crowbar a RISC-V decoder onto the Nuvia design and compressed instructions are breaking it somehow.

Joel_Mckay 2y ago
They should buy the intel alumni founded RISC-V startup, and pour resources into a RVA23 based chip with dual on-chip SDR asic sections (they have the IP). i.e. create a single open-chip solution for mid-tier mobile communication platforms. They won't do this due to their cellular chip line interests. However, even if it just ran a bare bones Linux OS... it would open up entire markets. =3
- delfinom 2y ago
  With how poorly Intel is doing, not sure "Intel alumni" is a plus. Lmao.
  - Joel_Mckay 2y ago
    Indeed, the Intel installed base has enough market inertia to last a business cycle or two with AMD. Even with the recent silicon defects... People will tolerate the garbage as they want the NVIDIA+Intel performance. Architecturally speaking, there were better options available... just never the equivalent price over performance of consumer grade hardware. =)