Announcement

Collapse
No announcement yet.

Mumblings Of A "Big New" Open-Source GPU Driver Coming...

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

  • Now guys don't get excited over what I'm about to say. In any event Think Apple here and the M1 processor. I know many here think that Apple is the worse thing to happen to open source but they have gone that route in the past when it suits their needs. The would do this for a number of reasons but one of them is to show wider industry support for their technology. It may not mean much to the Apple faithful but the more thoughtful users might see the ability to run Linux on M1 as a huge benefit. If you could run that Linux with the same efficiency as on Apple Mac OS, and had an excellent GPU drive, the current M1 machines would take the Linux world by storm. At least for the portable segment of the market as you simply can't get better performance per watt numbers anywhere.

    The other thing here is that WWDC is real soon now and that is a very good time to let something like this slip into the public view. It might also explain why Apple has hired so many Linux engineers.

    As for the other mentions so far:
    NVidia:
    Not a chance!

    Imagination:
    Actually this could be very interesting because lets face it they screwed up not getting their GPU's accepted in a wider way. With new management and a look at the broader markets it would make a lot of sense for them to put themselves in a better position to leverage the general move to ARM GPU's. If you are interested n ARM the only two real players you have are Apple systems and Linux systems. I don't see Microsoft surviving the move to ARM for general compute.

    Comment


    • Originally posted by wizard69 View Post
      I don't see Microsoft surviving the move to ARM for general compute.
      It's interesting how bad Microsoft's record is on anything outside of x86. Over the years, they've tried to gain a foothold in SPARC, MIPS, Alpha, Itanium, and ARM, and failed miserably every single time. If the industry follows Apple's lead and shifts away from x86 in the coming years, its going to be real interesting to watch, that's for sure!

      Comment


      • Originally posted by wizard69 View Post
        Imagination:
        Actually this could be very interesting because lets face it they screwed up not getting their GPU's accepted in a wider way. With new management and a look at the broader markets it would make a lot of sense for them to put themselves in a better position to leverage the general move to ARM GPU's. If you are interested n ARM the only two real players you have are Apple systems and Linux systems. I don't see Microsoft surviving the move to ARM for general compute.
        They had a vibrant development board ecosystem with the beagleboard and the pandaboard, both based on OMAP and the SGX.

        All ARM vendors screwed up gpu drivers, trying to replicate the IMG model of selling IP and then selling development services on top. SoC and device makers would get to pay these vendors to fix the bugs in their drivers... So when I opened up the ARM GPU space, i was a threat to each and every one of them (just like i was a threat to ATI in the decade before).

        ARM was politically against open source from the getgo, ARM also gained no traction with big open source players (especially with samsung too afraid to hire yours truly) for them to properly push for open source drivers. ARM backed linaro, which was supposed to solve all issues linux with ARM,but they too went very silent on open source gpu drivers very quickly.

        Vivante was probably too small, and has fallen to the wayside, only the freescale coup of slapping 19y hw support on the imx6 meant that some development was paid in the end.

        If Rob Clark had not managed to get under the protective umbrella of red hat, and its XX% of time allowed for personal projects, this too probably would not have happened. What was also fortunate is the radeon pedigree, and the lack of a development services that Qualcomm could sell for just graphics, as in this case, the SoC maker was the IP owner.

        Broadcom/Videocore never had a market, and the skewed promise of being fully open and pro-education of RPi (which still is a very fundamental falsehood almost a decade later), meant that it was the logical thing to do when everyone demanded open source gpu drivers back in 2012. And even then they pretty much came out with a lie initially.

        IMG lost a lot of marketshare, as it no longer was the ARM space monopolist. It had a hard time competing with ARM, who could sell all IP under one umbrella, and the probably much lower cost and not very competitive Vivante. OMAP pretty much vanished from the scene when Ti decided to no longer compete in the mobile space. Renesas remained, but still is a very niche player, even in the automotive market that it tried to relegate itself to. All that was left was Apple revenue, which was massive, but even that vanished and required some legal action. They tried to rebrand themselves slightly to help salvage their acquisition of MIPS, but the actions they took with GPUs achieved nothing. Plus, i was around, and i had worked with IMG hw at Nokia, and knew its driver/firmware structure, and i kept everyone from attempting to this fundamentally unstable driver structure. My logic from march 2011, when i started looking into freeing ARM GPUs: If we free the others, IMG will be forced to free itself. IMG now tries to move in on the RISC-V space, and they apparently learned from MIPS petering out; if they want to sell PowerVR IP to the RISC-V ecosystem they will have to have at least an open source driver.

        Despite being the nexus of open source drivers for ARM GPUs, despite having convinced everyone to stay away from PowerVR, and despite having put out the dumpster fire of the SGX code leak back in 2014, IMG of course did not bother to contact me.

        Comment


        • Originally posted by kpedersen View Post

          Haha, the top 500 super computers are *all* Linux. And over 50% of them utilize NVIDIA hardware.
          The cost of these machines actually dwarfs every Windows 10 consumer device ever made and sold.

          But of course, they give the source to those "big guys". It is really just plebs like you and I that miss out. And yet we represent the 99% of... well, y'know. Humans.
          Having worked on one of the top 1,000 machines I doubt it. They take input/design very seriously with vendors that size but the risk of their secret sauce getting out is very large. Best you get is documentation of undocumented features. HPC has a surprisingly open ethos beyond the lesser competent admins who practice "security by obscurity" as a religion.

          Comment


          • Originally posted by Qaridarium View Post
            right but if you are clever you use 1-2 year old amd cards... my vega64 right now is rock solid...
            yeah, same

            Comment


            • Originally posted by Linuxxx View Post
              I guess you didn't read down to the part where the OP noted....

              The flavor of Ubuntu we use is purpose built by Nvidia. I have a copy of the source if we wanted to play with it more. Which we have a team of software engineers working on a scheduler that will be made available to the public to submit for compute time on.
              I'm gonna have to say that would qualify as "not your typical Ubuntu distro.", so instead of refuting the claim you appear to have confirmed it.

              Comment


              • So nothing came of this...

                Comment


                • It won't be NVIDIA. Even ignoring the mining limitation stuff, they have (for a long time) had clauses in the license for their binary drivers (Linux, Windows, whatever) that prohibit the use of the drivers in conjunction with a consumer GPU (and in some cases even have driver code to enforce it) for certain purposes (not sure the details but IIRC it's some specific GPU compute workloads and I think certain VM things). They wont even release the things necessary for Noveau to turn up the clock/fan/power stuff from the lowest level (where it boots up at) for any vaguely modern GPU in part for this very driver restriction so they certainly aren't going to go whole-hog open source.

                  Could be ARM with Mali (not sure if the open source drivers for that are officially open sourced by ARM or just a RE project like Noveau is) or Qualcomm with their GPU (again not sure if the open source drivers are officially open sourced by Qualcomm or not)

                  PowerVR also makes sense, they used to have a decent amount of market share with their MBX and SGX GPUs (my old Nokia N900 had a TI OMAP chip in it with a PowerVR SGX GPU) but now they lost marketshare to ARM with Mali and Qualcomm with whatever they have in Snapdragon and such so going open could help them gain market share again.

                  Maybe its Broadcomm (or whoever) with a full FOSS GPU driver for all the graphics stuff in the various Raspberry Pi SoCs (so no more binary blobs for the Pi at all. (I know they published some stuff but what they published was not everything and not for the specific SoCs in the Pi)

                  Comment

                  Working...
                  X