3 Replies Latest reply on Feb 20, 2013 9:14 PM by alfa.c

    New FirePro "W"-Series

    Skysnake

      Hi,

       

      because AMD have not updated there hp, i wanted to told you, that there is the new FirePro "W"-Series. It is the first FirePro with GCN architecture.

       

      GL HF

       

      I hope google translate helps with the german news

       

      german:

      (test) http://www.tomshardware.de/firepro-gcn-workstation-grafikkarte,testberichte-241074.html

      they say, that many results are not clear. They think there are some driver problems.

       

      http://www.heise.de/newsticker/meldung/AMD-Workstation-Grafikkarten-mit-28-nm-GPUs-und-erste-FirePro-APU-1660807.html

       

      http://extreme.pcgameshardware.de/user-news/230667-amd-stellt-endlich-neue-firpro-mit-kleiner-uberaschung-vor-update-erster-test.html

       

      englisch:

      http://hothardware.com/News/AMD-Caters-To-Graphic-Pros-With-Blisteringly-Fast-FirePro-GPU-Refresh/

        • Re: New FirePro "W"-Series
          alfa.c

          What I noticed on many sites is how poorly Firepros do against older Fermi-based Quadros like the Quadro 4000 in Specview's CATIA & ProE benchmark (V7800 series, V7900 series, new W7000 series). The ProE issue is less of a concern since ProE is replaced by Creo 2.0 though.

           

          The Quadro 2000 should never be beating the V7900 let alone the W8000.

           

          It really comes down to drivers and the software support of the workstation GPU, because in DX11  the FirePros do simply fantastic. If the W7000 and V7900 can run with the Quadro 5000 and Quadro 4000 then it is good enough but ideally you want to outdo or match the new Kepler Quadro K5000 with the W8000.

           

          I feel like AMD Firepros need work on the OpenGL performance, because they seem tuned for DirectX like Radeons. At the end of the day people don't care about DX11 and 3dmark/passmark benches they care about real software performance not how powerful the GPU itself is. In fact the W5000 has similar performance to mid-highend Radeons in DirectX apps like Solidworks so it's cannibalized by cheaper cards from AMD.

           

          Another thing is the Tom's Hardware results seem all over the place because the V5900 outperforms the V7900.

           

          Right now Firepro is banking on Solidworks and PTC Creo 2.0... which doesn't bode very well when the real money is in Autocad (where Geforce does better than quadros), CATIA (heavily Nvidia biased), Maya, Lightwave, Cinema4D. It's bad when you comparing the Maya benches of the W7000 to your own V5900 (still selling on your site) instead of to the Quadro 4000.

           

          A good read is http://vr-zone.com/articles/why-amd-firepro-still-cannot-compete-against-nvidia-quadro-old-or-new-/17074.html

           

          and http://develop3d.com/workstations/gpu-group-test includes the Quadro K5000

           

          I am posting this today because I heard the AMD Developer summit is soon: http://blogs.amd.com/developer/2013/02/05/amd-2013-developer-summit/

           

          Thanks for your attention

            • Re: New FirePro "W"-Series
              allen.bourgoyne

              Viewperf 11 is not a good indicator of actual graphics performance.  SPEC also has application benchmarks as well as Viewperf.  If you compare results for Viewperf subtests for the various applications against actual application benchmarks, you will see there is no correlation, in many cases, the FirePro cards outperform their NVidia counterparts, in some cases, by a considerable margin. Viewperf uses graphics traces from applications to get data, but you they don't mimic the applications very closely, just grab an OpenGL trace, that in many cases needs to be edited in order to fit into the Viewperf application.  Since Viewperf can't replicate the actual application's rendering logic (this would likely violate patents, etc.), it can't do a great job of imitating actual application performance.  For example, if an application had code that did this: "draw a line; sleep for 1 hour; draw a line", Viewperf traces would just see "draw a line; draw a line" so the resulting Viewperf test would not be representative of the actual application.  Unfortunately most of the online reviewers don't have licenses for all of the applications, so they rely on just Viewperf because it is free.  Like any benchmark, you need to understand how and who developed it (Nvidia did Viewperf 11) so that you can analyze the results.  This is true regardless of card vendor, if you look at Viewperf 11 PTC testing, Nvidia low end card has about the same score as their Q6000.  I think most PTC users would tell you performance of those two cards is not the same.

               

               

              Allen B [AMD]

              This response is provided for informational purposes only, is provided "AS IS" and does not obligate AMD to provide any of the services, technology, or programs described.