Announcement

Collapse
No announcement yet.

LibreSOC Still Striving To Produce An Open-Source Hybrid CPU/GPU Built On OpenPOWER

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #81
    Originally posted by pal666 View Post
    zink's gl compliance isn't complete either. writing vulkan driver requires writing driver, llvmpipe is already available. and you can write different gallium backend no harder than write different vulkan driver. also there's lavapipe
    I don't think lavapipe is much use outside compliance/reference testing. A single command buffer isn't the design you want. https://airlied.blogspot.com/2020/08...layer-faq.html

    Zink is aiming for compliance, while srw is aiming to support a specific visualization toolkit. (also 4.2 vs 3.2currently)

    And while a specific gallium backend might be nice, that I suspect will be a later problem and not needed to produce the minimal viable product. But given that even desktop environments are suporting vulkan, it may not be necessary as OpenGL becomes legacy. (And raytracing will accelerate that trend rather further)

    Comment


    • #82
      Originally posted by lkcl View Post
      Imagination let their patents expire: the announcement of the new GPU by apple was timed just after that. Imgtec's CEO went mental, and had to be told by his lawyers, "yeahh you screwed up by not keeping an eye on the patents".
      yehyeh. Jeff Bush's approach is the "guide" here - if you read his paper he created a metric "pixels / clock" and goes to some lengths to work out which stages of a Shader spend the most cycles. he was then able to try (i think) 7 different design modifications to improve on that: some of them software, some in hardware (but limited to adding general-purpose compute instructions only).
      we'll replicate that approach... but include hardware 3D custom opcodes. trouble is, right now, everyone's asking, "what's the performance what's the performance" and until we have a first iteration in place (until we have replicated what Jeff did) we simply can't answer.
      l.
      sounds very interesting... bridgman from AMD also did write they want to look into OoO for radeon gpus.
      believe it or not right now "performance" is not yet a critial point in your mission.

      i personal y do not believe in mobile chips for such a project because anyone who care about power consumtion will buy 5nm apple or 7nm google ARM hardware...

      best first market will be high security desktop/workstation/server chip on the mainboard or PCIe card
      in this power consumsion is not the main problem because it just costs money in the electrix bill and does not suck out the battery of your mobile device.

      whatever performance it maybe be... you can fokus on create better performance later.

      if your chip can handle gnome desktop at 4K it will be fine for a start.
      Phantom circuit Sequence Reducer Dyslexia

      Comment


      • #83
        Originally posted by Qaridarium View Post

        sounds very interesting... bridgman from AMD also did write they want to look into OoO for radeon gpus.
        believe it or not right now "performance" is not yet a critial point in your mission.

        i personal y do not believe in mobile chips for such a project because anyone who care about power consumtion will buy 5nm apple or 7nm google ARM hardware...
        luckily we have a potential customer (two) who are interested in the open-ness and transparency more than anything else. they *need* a full audit trail for their high security customers and we're the only SoC company that's prepared to say, "yep we're happy for you to do a full audit: most code is online anyway. anything beyond that you pay a service charge but we're still perfectly happy to let you do a full audit, knock yourself out".

        best first market will be high security desktop/workstation/server chip on the mainboard or PCIe card
        in this power consumsion is not the main problem because it just costs money in the electrix bill and does not suck out the battery of your mobile device.

        whatever performance it maybe be... you can fokus on create better performance later.

        if your chip can handle gnome desktop at 4K it will be fine for a start.
        this should be easily achievable, especially if we include DDR4 ECC RAM and allow at least 16 GB or 32 GB.

        one thing we are planning (actually, been requested by the customer) is including both PCIe Host *and* PCIe Client. this allows the processor to actually become a Graphics Card *and* to be able to plug in NVMe SSDs, SATA PCIe Cards etc. etc.

        it's multiple market, basically. this is how you amortise the massive cost of the NREs. it's an industry-standard practice for SoCs. you don't go after one market (too risky), you go after several.

        Comment


        • #84
          Originally posted by lkcl View Post
          luckily we have a potential customer (two) who are interested in the open-ness and transparency more than anything else. they *need* a full audit trail for their high security customers and we're the only SoC company that's prepared to say, "yep we're happy for you to do a full audit: most code is online anyway. anything beyond that you pay a service charge but we're still perfectly happy to let you do a full audit, knock yourself out".
          Yes NSA want to buy IBM/redhat servers/workstations with a full audit and all opensource.
          also i think google has similar customers like NSA who also want such a GPU.

          yes full audit and all open and opensource is really a great feature.

          Originally posted by lkcl View Post
          this should be easily achievable, especially if we include DDR4 ECC RAM and allow at least 16 GB or 32 GB.
          one thing we are planning (actually, been requested by the customer) is including both PCIe Host *and* PCIe Client. this allows the processor to actually become a Graphics Card *and* to be able to plug in NVMe SSDs, SATA PCIe Cards etc. etc.
          it's multiple market, basically. this is how you amortise the massive cost of the NREs. it's an industry-standard practice for SoCs. you don't go after one market (too risky), you go after several.
          yes go for 32GB DDR4 ECC ram is really a epic win.
          and also your plan to support PCIe Host and Clint is also a great win.

          yes this is all very good. i really believe in you and your doing.
          Phantom circuit Sequence Reducer Dyslexia

          Comment

          Working...
          X