Announcement

Collapse
No announcement yet.

NVIDIA Loses Huge GPU Order Due To Linux Blob

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

  • #11
    It's not that big a win if it's a one-time affair. Those ten million GPUs will need support and currently neither AMD nor Nvidia have any dedicated resources for MIPS or ARM.
    Also, keep in mind this is China we're talking about. Most likely they don't care much about the card, they just want the specs of the GPU so they can make a chinese version. But that's just speculation on my part.

    Comment


    • #12
      Well for basic 3d radeon oss is ok, but maybe they need to add a dedicated h264 decoder chip in addition.

      Comment


      • #13
        My God it might finally be time to track down AdBlock for Firefox and put it to use.

        Where do these ridiculous articles come from?

        First, the "source" is mostly hearsay and there's no details that could even be confirmed.

        Second, what kind of a MORON would actually release source code to a Chinese "company"? lol. WhyTF do people think NVIDIA refuses to open any of their tech to begin with? They're not worried about IBM here.

        Comment


        • #14
          Originally posted by bug77 View Post
          It's not that big a win if it's a one-time affair. Those ten million GPUs will need support and currently neither AMD nor Nvidia have any dedicated resources for MIPS or ARM.
          Also, keep in mind this is China we're talking about. Most likely they don't care much about the card, they just want the specs of the GPU so they can make a chinese version. But that's just speculation on my part.
          The article was about an order for 10 million hardware units, not just blueprints or specifications.

          Also, the open drivers are grossly inadequate in terms of providing anything that China would need to create duplicates of AMD GPUs. You still need:

          1. Fabrication processes that can fab at a competitively small size (only a few foundries in the world have hit 32nm and smaller, and those companies wouldn't give up their secrets for all the tea in China, literally)

          2. The hardware / IC design blueprints at many different levels of abstraction, with understanding of how it all fits together.... way more complicated than I can even express in words.

          3. The proprietary IP that isn't disclosed in the open source driver -- a lot of the bits of AMD cards will never be unlocked by the open drivers unless reverse engineered without AMD's consent, because AMD won't publish specs on them. UVD I'm looking at you, but there are others too... VCE, could be others.

          4. The proprietary software algorithms in fglrx that AMD doesn't want to release due to the potential of giving Nvidia a competitive advantage... while some of the LLVM stuff is pretty hairy and computer science-y in the open drivers, the general point that fglrx is using "secret" micro-optimizations and algorithms which are not being disclosed to the open driver developers is something I highly suspect, despite protestations from Bridgman otherwise. I mean, if this isn't the case, then explain why the open drivers will never even approach the performance of fglrx except on passively cooled / lowest-end cards and APUs. You've released all the "specs" for the hardware that does the 3d rendering, right, so WHAT makes up the performance difference if HiZ, 2D tiling, PCI-E 2.0 and the other obvious gaping-hole missing features don't make up for the difference?

          I bet China (the government, not any particular corp) could create a GPU about as fast as an Intel 965G integrated chip from ~2008, and that's just going on knowledge that is either publicly available or easily obtainable from high-tech Chinese corporations that cooperate with the government. But I doubt that they have enough information to create something truly competitive with an AMD or Nvidia chip without going to AMD/Nvidia and asking them to produce cards for them, on their (well, TSMC's) foundry process, using their IC designs, not China's.

          Comment


          • #15
            Originally posted by allquixotic View Post
            WHAT makes up the performance difference if HiZ, 2D tiling, PCI-E 2.0 and the other obvious gaping-hole missing features don't make up for the difference?
            Probably per-application hacks and optimizations.
            Last edited by RussianNeuroMancer; 22 June 2012, 08:02 PM.

            Comment


            • #16
              Originally posted by johnc View Post
              Second, what kind of a MORON would actually release source code to a Chinese "company"? lol. WhyTF do people think NVIDIA refuses to open any of their tech to begin with? They're not worried about IBM here.
              This "company" is a government institution doing main R&D for the Chinese science & defense departments, builds many of China's supercomputers (which they use HyperTransport for—something AMD designed with input from Cray), provides hardware to the Chinese military, and has been using AMD chipsets & IGPs for a couple of years, at least.

              And AMD has the radeon driver, which might be enough for most intended purposes.

              I'd say the most interesting part of that article is that they considered switching to Nvidia...
              Last edited by JanC; 22 June 2012, 07:57 PM.

              Comment


              • #17
                hmmm big fail for nvidia... i guess, i do own a nvidia gfx card and i'm pretty much happy about it. with this 10 million deal AMD will get things changed, hope fully the OSS drivers could be in parity feature/support with the closed source windows drivers. though i'm gona stick with my current gfx card for now until i see that AMD gets its oss drivers, as i said previously, in parity with its win-drivers

                Comment


                • #18
                  What surprises me is nobody yet has paid attention to why this company is doing this. Think about it, these are ARM and MIPS processors that are likely using pci-e 16x based GPUs. First of all, I'm not aware of any processors of those kinds doing such a thing, but I don't see it likely that this company is also doing 1 GPU per CPU - I'm guessing they're going to look into Crossfire. This strategy makes sense, because x86 processors are a little overkill if you're just trying to use GPUs for massively parallel tasks.

                  On a side note, it is absolutely fantastic that AMD is getting this attention - they need the revenue and not only will AMD use this money to supply more open source driver developers (they're obligated to at this point) but maybe this company will also help make contributions. I guess the bad news is these driver changes are likely only going to focus on openCL and the likes, not so much graphics.

                  As for comments about there not being radeon driver for ARM or MIPS, why do you think these companies wanted the open source in the first place? If nvidia supplied an ARM based binary of their current blob driver, then I'm sure this company wouldn't have been so quick to dismiss them. The only thing that really confuses me is why didn't they just use the nouveau driver if they care about open source so much? Although that driver doesn't have dedicated support, its still better overall.

                  Comment


                  • #19
                    I think people are missing the point

                    Quite a few people here seem to be making the assumption that the Chinese chose AMD because they have open source drivers (to be fair, this is what Michael's article says). Having actually read the original article...

                    After years of focusing on building infrastructure, one of priorities for the Chinese government is focusing on building the knowledge infrastructure as well. In terms of computers, this is a real battlefield. While the western media tends to go by Blue’s company line about GPUs being a “thing of the past” and frantically releasing the […]


                    ... it seems that AMD is giving the Chinese access to the source code of their proprietary driver, presumably under NDA, which they will then port to the Longsoon architecture, and, I assume, keep closed.

                    I'm not sure this is a win for open source at all. It *does* seem to be a win for "the customer is always right" when that customer has a lot of money, which presumably most corporations already realise. Except possibly Nvidia, if the original article is to be believed.

                    I would like to be wrong about this, and have it demonstrated that this large GPU order was due to the existence of an open driver, but right now I don't see it.

                    Comment


                    • #20
                      So much for the argument that there is no good financial reasons to open source your drivers.

                      If there's one thing I think AMD does better than its competitors (Intel and nVidia), it is the attempt to offer the industry what it wants. They left AMD64 open, and basically laid down the framework for efficient and cheap multi-CPU x86 computing with the HT link. Those 2 things alone helped stand the market on its ear and fundamentally changed Intel for the better. Itanium just didn't seem like a positive future, no matter how many billions went into it.
                      Last edited by MonkeyPaw; 22 June 2012, 09:44 PM.

                      Comment

                      Working...
                      X