Announcement

Collapse
No announcement yet.

AMD Raven 2 Support Might Land In Time For Mesa 18.3

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

  • AMD Raven 2 Support Might Land In Time For Mesa 18.3

    Phoronix: AMD Raven 2 Support Might Land In Time For Mesa 18.3

    Prolific open-source AMD hacker Marek Olšák has sent out a revised patch enabling support for the yet-to-be-released Raven 2 APUs within the RadeonSI Gallium3D...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    It will be interesting to see how much of an improvement Raven 2 is. I'd like ot see the chip is a NUC sized computer as my laptop experience with Raven 1 hasn't been that bad.

    Comment


    • #3
      APUs are the future of computing. All that is holding them back is RAM bandwidth. As process nodes get better, and since mhz can only go up so much while most people don't really need that many cpu cores, the option is either to just drop the TDP of cpus and apus or enlarge the igpu part and add on-chip L4 cache/HBM RAM. Assuming HSA and libraries improve they can provide a great option for high performance computing/gaming on an APU.

      The question is, if AMD is willing to put HBM3 RAM on an 7nm APU when it becomes available. Rumours say HBM3 is going to be considerably cheaper than HBM2, and if the igpu gets large enough on an 7nm process performance around 3-4 Teraflops is not impossible... With 2-4gb of HBM3 acting as a buffer for the igpu, this part could easily eliminate the under 100-200$ dGPU market. Since gpus are really expensive these days, not to mention energy hungry and noisy, many many people would opt for such a gaming APU. I sure hope AMD has plans for something like this. I would buy such an APU in an instant, for both gaming and ROCm reasons...

      Comment


      • #4
        Originally posted by TemplarGR View Post
        APUs are the future of computing. All that is holding them back is RAM bandwidth.
        That and thermals.

        Comment


        • #5
          Originally posted by LinAGKar View Post

          That and thermals.
          Thermals are no real issue actually. Not for the long term, at least. In the short term, there is the issue that the market probably isn't ready to support larger apu dies instead of the traditional and wasteful big cpu + big gpu combo. APUs are significantly more efficient due to unified RAM. Take for example a 65W cpu + 135W dgpu combo vs a 200W APU: The APU should be considerably more efficient and faster, given similar RAM bandwidth. The problem is that you need a new socket and higher end motherboards to support such an APU, and since APUs are currently considered the budget choice the market is not ready to support such a product. But if they keep advancing APUs by adding HBM3 and beefing the iGPU up, eventually they will be in a position to support larger APUs and even dual socket motherboards for larger APUs, and it is game over for dGPUs...

          Comment


          • #6
            Originally posted by TemplarGR View Post
            APUs are the future of computing. All that is holding them back is RAM bandwidth. As process nodes get better, and since mhz can only go up so much while most people don't really need that many cpu cores, the option is either to just drop the TDP of cpus and apus or enlarge the igpu part and add on-chip L4 cache/HBM RAM. Assuming HSA and libraries improve they can provide a great option for high performance computing/gaming on an APU.

            The question is, if AMD is willing to put HBM3 RAM on an 7nm APU when it becomes available. Rumours say HBM3 is going to be considerably cheaper than HBM2, and if the igpu gets large enough on an 7nm process performance around 3-4 Teraflops is not impossible... With 2-4gb of HBM3 acting as a buffer for the igpu, this part could easily eliminate the under 100-200$ dGPU market. Since gpus are really expensive these days, not to mention energy hungry and noisy, many many people would opt for such a gaming APU. I sure hope AMD has plans for something like this. I would buy such an APU in an instant, for both gaming and ROCm reasons...
            8/16 cores with vega 25 sort of performance would make a very nice machine

            Comment


            • #7
              Originally posted by TemplarGR View Post

              Thermals are no real issue actually. Not for the long term, at least. In the short term, there is the issue that the market probably isn't ready to support larger apu dies instead of the traditional and wasteful big cpu + big gpu combo. APUs are significantly more efficient due to unified RAM. Take for example a 65W cpu + 135W dgpu combo vs a 200W APU: The APU should be considerably more efficient and faster, given similar RAM bandwidth. The problem is that you need a new socket and higher end motherboards to support such an APU, and since APUs are currently considered the budget choice the market is not ready to support such a product. But if they keep advancing APUs by adding HBM3 and beefing the iGPU up, eventually they will be in a position to support larger APUs and even dual socket motherboards for larger APUs, and it is game over for dGPUs...
              That is assuming that dGPUs do not get more and more interesting abilities. ATM we have Ray Tracing making its way into games, but that is a short term fix for the better Path Tracing tech. Path tracing is going to require a large amount of dedicated silicon.

              We will continue seeing better and better APU graphics (a decade ago IGP's were a joke, now you can build a low end 1080p gaming machine).

              Comment


              • #8
                Originally posted by TemplarGR View Post
                APUs are the future of computing. All that is holding them back is RAM bandwidth. As process nodes get better, and since mhz can only go up so much while most people don't really need that many cpu cores, the option is either to just drop the TDP of cpus and apus or enlarge the igpu part and add on-chip L4 cache/HBM RAM. Assuming HSA and libraries improve they can provide a great option for high performance computing/gaming on an APU.

                The question is, if AMD is willing to put HBM3 RAM on an 7nm APU when it becomes available. Rumours say HBM3 is going to be considerably cheaper than HBM2, and if the igpu gets large enough on an 7nm process performance around 3-4 Teraflops is not impossible... With 2-4gb of HBM3 acting as a buffer for the igpu, this part could easily eliminate the under 100-200$ dGPU market. Since gpus are really expensive these days, not to mention energy hungry and noisy, many many people would opt for such a gaming APU. I sure hope AMD has plans for something like this. I would buy such an APU in an instant, for both gaming and ROCm reasons...
                Fully integrated monolithic APUs can't scale to high performance configurations and will always be limited to low-midrange. The only reason consoles do this is their very long lifespans as products... meaning the yield issues have a long time to be worked out. Rather than only having a 6mo-1year lifespan.

                To scale you need multiple chips due to using recent processes with iffy yields. So the only way you can get a large fast GPU/CPU combo is to do it on some sort of interposer with multiple chips + infinity fabric.

                Comment


                • #9
                  Originally posted by cb88 View Post

                  Fully integrated monolithic APUs can't scale to high performance configurations and will always be limited to low-midrange. The only reason consoles do this is their very long lifespans as products... meaning the yield issues have a long time to be worked out. Rather than only having a 6mo-1year lifespan.

                  To scale you need multiple chips due to using recent processes with iffy yields. So the only way you can get a large fast GPU/CPU combo is to do it on some sort of interposer with multiple chips + infinity fabric.
                  I feel that AMDs MCM approach will be the winner over the next decade - they just need to figure out how to split things up logically

                  Comment


                  • #10
                    Glad to know that there is support for notebooks we'll probably never see.

                    Comment

                    Working...
                    X