Announcement

Collapse
No announcement yet.

Zink Lands Threaded Context Support For A Big Speed Boost With OpenGL Over Vulkan

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

  • #31
    Originally posted by oleid View Post

    And you forget that Zink is re-using the already fine-tuned gallium stack.
    Gallium works, but I wouldn't exactly call it "fine tuned". There is much work to be done at every level of this stack. :+ )

    Comment


    • #32
      Originally posted by bridgman View Post
      That might be optimistic. We are no longer seeing the kind of drops in cost per transistor that drove most of the historical price/performance gains.

      That said, the crazy retail/scalper pricing we are seeing today should go away with the next mining bust for sure.
      There have been a few studies into where the cards are going. We were heading into a GPU/Silicon supply shortage before the mining boom. The mining boom has not helped.

      The world growth in silicon usage has exceed forecasts of TSMC and other fabs and also due to fabs being cut throat with each other on margins there is not enough rainy day funds put away to build new Fabs to cover the up coming demand as well as not starting the fabs construction before the demand exceed ability to supply.

      https://www.techpowerup.com/276029/t...duct-prices-up

      We started seeing the fab pressure problem at the end of 2020 with may different fabs dropping their discounts on big volume orders. We are now seeing TSMC and others planning to slow down price reduction for older nm production.

      The reality is we could be going into 5 years freeze on silicon production price reduction this will also mean newer nodes will be higher price. We are going into a abnormal time in silicon pricing that is going to last a while because its going to take 5-8 years to build the fabs get them running and catch up. Silicon fabs cost billions of dollars to build and that money has to come from somewhere.

      TemplarGR is being insanely optimistic from what I am seeing. Mining bust may fix the scalper pricing but from my point of view with the current supply of silicon production problems and the higher prices that will have to be paid due to the silicon production supply problems expect for the next 5 years for GPU to possible go up by 100-200 USD for the same performance to cover the production problems.

      Do note things like car "automotive engine management" chips that are not needing the latest silicon that don't come from the fabs that make CPU or GPUs are also having production short falls.

      From what I have seen there was a general problem with silicon production before the mining boom. The mining boom is just at the worst possible time. Even without the mining boom gpu price reductions for performance due to lack of fab supply was going to stall.

      Comment


      • #33
        Originally posted by oiaohm View Post
        There have been a few studies into where the cards are going. We were heading into a GPU/Silicon supply shortage before the mining boom. The mining boom has not helped.
        Agreed - however IMO there is an important distinction between "demand running near the supply limit" (which results in temporary shortages like the one we experienced during the latest console refresh launch) and "demand hopelessly outstripping supply" which is what we got with the addition of a mining boom.

        Originally posted by oiaohm View Post
        We started seeing the fab pressure problem at the end of 2020 with may different fabs dropping their discounts on big volume orders. We are now seeing TSMC and others planning to slow down price reduction for older nm production.
        Fair point - I don't know for sure how much of the reduction in historical price-per-transistor cost reduction trends is due to increased production costs and how much is due to reduced competition between fab vendors due to lack of surplus capacity. Certainly both appear to be a factor.

        Comment


        • #34
          Originally posted by bridgman View Post
          Fair point - I don't know for sure how much of the reduction in historical price-per-transistor cost reduction trends is due to increased production costs and how much is due to reduced competition between fab vendors due to lack of surplus capacity. Certainly both appear to be a factor.
          Those two are not independent. As the cost of making new fabs increases that is part of the increased production cost this reduced the number of fab vendors with the funds to produce make new fabs that results in reduced surplus capacity and competition between fab vendors in time.

          The number of vendors making the highest end silicon has been reducing being AMD you saw how global foundry dropped out being top end silicon production but they were not the only one.

          The writing on the wall that this crunch was coming was really their years ago. The mining boom is just the straw that broke the back of all ready well and truly overloads camel,

          The current mess has been a perfect storm.
          1) Increasing fab creation cost.
          2) reducing vendors making top end silicon due to the increase amount of funds need to make more advanced fabs.
          3) 1 and 2 above reducing capacity growth.
          4) Those buying silicon making the remaining vendors of silicon be really tight on mark-up.
          5) Demand progressively increasing faster than new capacity coming on line.

          These 5 points were the case before the mining boom. So silicon production was on the path of running out of capacity. Mining boom just made it happen sooner instead of latter.

          Of course the consumer buying the products wanting year on year better price on performance is one of the driving facts of point 4.

          The first 1-4 are linked and cause 5. If cost to buy silicon had properly gone up relative to the cost to make new fabs so that the current generation of silicon being made was bringing in the money to make the next generation of fab without requiring massive production lot of smaller vendors of silicon would still be around and the capacity would be larger right now so the mining boom would not have been as big of a problem.

          The hard point we were going to have a silicon production crunch sooner or latter. Sorting out the crunch is most likely going to see a price correction in the upwards direction for a while to move the price closer to what it should have been. Maybe allowing some of the fab companies that dropped out of high end silicon production to reenter the game with means to make a profit..

          Comment


          • #35
            I'm imagining that in a year or two, most new software, from toolkits, to game engines, and desktop environments will be outputting to Vulkan. In that world, Zink plays a big role allowing OpenGL to work where it's still invoked.

            People seem fixated on OpenGL as if the universe isn't going to shift to Vulkan. Same for X11. I don't think these things are really up for debate; X11 is dying and Wayland is replacing it. OpenGL 4.6 is the terminal version of OpenGL and Vulkan is replacing it. X11 and OpenGL will still -work-, but it's going to be OK if they're a bit slower or quirkier than they are today, because they're going to be legacy. I don't think getting 100% native performance on OpenGL is going to matter when OpenGL is only being invoked to run legacy software.

            Comment


            • #36
              Originally posted by mangeek View Post
              OpenGL 4.6 is the terminal version of OpenGL
              This is not quite right. There might be a 4.7 or 4.8 ...yet from what Khronos has said. But those will not have any new extensions and will happen if multi vendors new GPU are shipping with the same extensions. There is not going to be a Opengl 5.x. So Opengl 4.X is the end of the line with 4.6 being the end of new feature development.

              Opengl is not quite in as bad of position as x org X11 server as Opengl still has a maintainer in the Khronos group. Note I said not quite but its very close to as bad.

              Other thing to consider is like some of your open source legacy software that is currently opengl based could end up over time with ports to Vulkan as well. So for users of open source software Zink may be only a stop gap.

              Comment

              Working...
              X