7 Replies Latest reply on Sep 23, 2017 10:26 PM by colesdav

    MultiGPU (Crossfire DX11) Benchmark - Final Fantasy XIV Stormblood Benchmark. RX Vega/FE Results please?

    colesdav

      Hello,

       

      I was recently investigating purchase of two RX Vega LE / FE Cards for a new AMD Threadripper Build.
      I have decided against purchase of the RX Vega LE/FE as I need Windows 8.1 64bit OS Support.

      MultiGPU support (two cards) has just been added for Vega Cards with the recent Crimson ReLive 17.9.2 Drivers.
      I thought I may as well share the data I generated whilst investigating Vega purchase.

       

      As you will see from my data below, I get great scaling at 4K UHD with a pair of R9 Nanos on my i7-4790K based system.
      I see no Crossfire related flicker issues at all in the benchmark. It runs really well.

       

      The release notes for the new Vega MultiGPU driver can be found here: Radeon-Software-Crimson-ReLive-Edition-17.9.2-Release-Notes

      Note that the term MultiGPU is AMD new marketing term to cover Modern MultiGPU (DX12) technology and older Crossfire (DX11).
      More information can be found in the following article: Radeon RX Vega graphics cards get 2-way multi-GPU support in Radeon Software 17.9.2 | PCWorld


      Today I will post the Crossfire (DX11)  data I generated running the Final Fantasy XIV Stormblood benchmark
      I discovered the Final Fantasy XIV Stormblood benchmark thanks to  whiskey-foxtrot , he used it to test the performance of his  Watercooled RX Vega 64 Cards on his outstanding Oscar Mike  build.

       

      I tested Crossfire Performance running on the following System:

       

      CPU = Intel i7-4790K CPU. Running at a 9% overclock from Stock. (4.6, 4.6, 4.5, 4.5 GHz for 1,2,3,4 cores active).
      GPU1 =  Powercolor AMD R9 Nano (Fury X) with 4GB HBM.
      GPU 2 = XFX AMD R9 Nano (Fury X) with 4GB HBM.
           I ran the tests using AMD Crimson ReLive 17.9.1 Driver.
           Both R9 Nanos were running with Max Fan, Temp and Power Targets, and a 5% GPU CLK Overclock, to give 1050MHz max.
           I was unable to Under or Overclock the HBM memory in these tests.

           No third party OC Software (Such as Sapphire Trixx or MSI Afterburner) were installed on my System during the tests.

      Motherboard = Asus Z97 Deluxe Motherboard with NFC and Wireless.
      RAM = 32GB Corsair Ballistix Tactical LP DDR3 1600MHz.

       

      I ran the tests on Windows 10 64bit OS with latest Patches as of 15 September 2017.

      The Windows 10 64bit  OS installation was checked, and passed all tests  following fresh AMD Crimson ReLive 19.9.1 Installation.

       

      I would be grateful if people with RX Vega / Vega FE running in Crossfire could run the Final Fantasy XIV Stormblood Benchmark and post their results to this thread please.
      It would give me some idea how Crossfire DX11 on RX Vega / FE cards compares to my existing Dual R9 Nanos.

        • Re: MultiGPU (Crossfire DX11) Benchmark - Final Fantasy XIV Stormblood Benchmark. RX Vega/FE Results please?
          colesdav

          Here is the summary of my Testing Results.

          I tested  four different monitor resolutions: FHD, QHD, QHD+ and 4KUHD.

          The benchmark was run at Maximum Quality Preset and no other benchmark Graphics setting was changed.

          I ran 10 independent tests for each resolution.

           

          P = X*Y = number of Pixels. 
          Scaling = (Score XF) / (Score).

           

          Horizontal Axis = P = Number of Pixels = Resolution.
          Vertical Axis      = Scaling.

           

           

           

          I hope RX Vega/FE Users may use the above data to compare how well their cards and system perform versus my system above.

          Thanks.

            • Re: MultiGPU (Crossfire DX11) Benchmark - Final Fantasy XIV Stormblood Benchmark. RX Vega/FE Results please?
              colesdav

              Here is the information required to download the Final Fantasy XIV Stormblood Benchmark.
              You may download the benchmark from the following location:

               

              FINAL FANTASY XIV: Stormblood Official Benchmark

               

              Cheers.

              • Re: MultiGPU (Crossfire DX11) Benchmark - Final Fantasy XIV Stormblood Benchmark. RX Vega/FE Results please?
                whiskey-foxtrot

                Credit for finding the Stormblood Benchmark actually goes all to darkwolfca, I would never have known about it had he not mentioned it!

                 

                Is "Score FX", your Crossfire score? If so you've pretty much beat my current settings at the same resolutions (on the "Balanced" profile):

                FFXIV-1920x1080.png

                FFXIV-2560x1440.png

                FFXIV-3200-1800.png

                FFXIV-3840x2160.png

                1 of 1 people found this helpful
                  • Re: MultiGPU (Crossfire DX11) Benchmark - Final Fantasy XIV Stormblood Benchmark. RX Vega/FE Results please?
                    colesdav

                    Hi,

                     

                    Yes, Score XF is the Score with Crossfire on. I used default AMD Crimson  ReLive Global Settings for both cards, and as I stated above I ran the R9 Nanos with Max fan, Temp and Power target set and a mild +5% OC. Note both of my R9 Nanos are running air cooled, with the reference fans and shrouds, not on water.
                    I took screenshots of all settings I used if you need any more information.

                     

                    I would have expected the Vega 64 to be around 30% faster than an R9 Nano for a single card.
                    I do not know how well your Vega 64 cards are scaling in Crossfire though.
                    Since your cards are running on water you should really be hammering my results I would have thought.

                     

                    I did not purchae RX Vega LE/FE because no Windows 8.1 64bit support, however I did need some more OpenCL GPU's

                    I managed to pick up another R9 Nano (Fury X) for 250 in another sale ... people are selling them off cheap now that RX Vega 56/64 is here.

                     

                    I will do a few more runs with the 19.7.2 Driver, and post those results later.

                     

                    Cheers.

                      • Re: MultiGPU (Crossfire DX11) Benchmark - Final Fantasy XIV Stormblood Benchmark. RX Vega/FE Results please?
                        whiskey-foxtrot

                        In DX11 results can be all over the place and it's not really the best benchmark to scale RX Vega (or Vega FE). In modern API's it blows everything else out of the water without really flinching. OpenCL/rendering/Compute/etc)? No question about it, Vega in any form would be my pick over previous generations.

                         

                        With the DX11 benchmarks there's of course a lot of fine tuning that can happen like with the results I posted earlier, I simply ran on Balanced mode - both cards remained under 32C, at 70% capacity on one, less than 40% on the other - with zero noise and no stress on the system. I wouldn't be able to say the same with any older card(s).

                          • Re: MultiGPU (Crossfire DX11) Benchmark - Final Fantasy XIV Stormblood Benchmark. RX Vega/FE Results please?
                            colesdav

                            Hi,

                             

                            I absolutely agree  that the RX Vega and Vega FE are much better in OpenCL / Rendering than the R9 Nano. I do have those numbers  also. I just did not get around to posting those last week. I have been investigating some Gaming App's in my spare time over the past two weeks.

                             

                            The R9 Nanos were both throttling at the end of the benchmark for sure and they were getting up to around 65-70'C at the very end but they are very compact cards with small air coolers. Running the R9 Nanos  on EK Waterblocks would solve the temperature and maybe throttling issues, but the R9 Nanos still are hampered by lower power limit in VBIOS, inability to under/overclock the HBM  and the Voltage Regulator on the R9 Nano is a much cut down version of the Voltage Regulator on an R9 Fury X, for example.

                             

                            The Room Temperature was 26'C during my testing and the R9 Nano cards were both at 30'C before each benchmark run started.

                             

                            It would be interesting to see how a pair of  R9 Fury X performs on the Final Fantasy Stormblood benchmark.

                             

                            I agree that DX 11 is an old API and from the CPU side one fast core is much more important than many cores as far as the AMD DX11 Driver is concerned.

                             

                            The DX11 API is however used by the vast majority of games and even new titles, like PREY,  are still using it.

                             

                            I was just looking at the Steam Hardware Survey yesterday.
                            Here is the graph of the market share of Gaming GPU's  for the last year according to Steam:

                             

                            If you look at it, Nvidia are in a very strong position in gaming now.

                            They have increased from 56.4% to 67.62% in the space of 1 year. That is a gain of 11.22% in 1 year.

                            You can also see their rate of market share gain increased since the RX Vega Launch.

                            I do not know if Nvidia reduced GPU prices to increase the rate of market share gain since RX  Vega launch, or if AMD Users waiting for Vega were disappointed by initial Vega Gaming Performance/Power in reviews and simply bought Nvidia cards instead.It might be high prices due to Mining or a AMD fans simplyu cannot get the Vega cards they want at a reasonable price.
                            Maybe some people are still waiting for Vega AIB cards with better coolers to turn up. Maybe dropping Windows 8.1 64bit OS Support is a factor as well. It was in my case.

                             

                            If the current trend for Nvidia gaining Market Share in gaming continues, then, based on what I understand how markets behave, Nvidia will become so dominant that they will dictate which  API is used by developers.

                             

                            I think it is likely that Nvidia are in no hurry to push DX12 or Vulkan, and if they are in no hurry, I do not see many games developers in a hurry to move to DX12 any time soon.

                             

                            Maybe AMD might want to look at working on a better multithreaded DX11 Driver at the moment and worry about newer API's later on  if that is even possible?

                             

                            I think that AMD GPU Reviews need to see AMD GPU's beating Nvidia cards on DX11 API, not just DX12/Vulkan, where there are still very few titles.

                             

                            I really do think that RX Vega/FE has some very interesting and forward thinnking technology. I love to see competition as it is the only way to improve technology and reduce prices.
                            The above Steam Survey Chart is not good for anyone, apart from Nvidia. 
                            Hopefully newer Vega GPUs will turn it around.

                             

                            Cheers.

                            • Re: MultiGPU (Crossfire DX11) Benchmark - Final Fantasy XIV Stormblood Benchmark. RX Vega/FE Results please?
                              colesdav

                              Just a thought.

                               

                              I do have some other DX12 and Vulkan Benchmark numbers if you want to compare those?

                              One title I know of that runs really well with  DX12 and is already optimized to run on AMD Cards is Hitman, and it has an automated benchmark.
                              Not Hitman also runs natively on Linux ;-).


                              I think Doom is well known as ~ the only game really well optimized for Vulkan Interface to date. We could look at that.
                              Wolfenstein II: The New Colossus should be released soon (Mid October 2017?) so that should be a Title that runs vcery well on RX Vega we could compare.
                              I am looking forward to playing that title, not just testing it.

                               

                              Maybe you might want to see how the Vegas run on Ashes of the Singularity as well? It has Vulkan Interface now. I had started a thread on that and I do have some numbers already.

                               

                              I do see Vulkan, Ryzen/Threadripper/Epyc and PCIe Passthrough technology as a serious game changer in the future for AMD.
                              Gaming Developers and Valve must be seriously interested in having only one cross platform API that runs on Linux.
                              Gamers could run Windows inside Linux on a Virtual Machine, allowing Windows to acess PCIe hardware like the GPU and SSD's.
                              Those  AMD CPU's have plenty of spare Cores to run Windows on a Virtual Machine, and still keep the main Linux OS running well.
                              If AMD does a good job on supporting PCIe Passthrough on Linux with Ryzen, Threadripper and Epyc & AMD Partner Motherboard and BIOS then they could use their very strong CPU market and technical position to make the move away from Windows DX11 & 12 and on to Linux + Vulkan for Gamers a real possibility. That will likely take many years still though.