13 Replies Latest reply on Aug 17, 2017 9:47 AM by colesdav

    Infinity Fabric performance directly proportional to RAM speed equating to increase in FPS

    goldstone77

      Can anyone offer validation to these claims? If so this would be fairly important news to put out for the RED TEAM! In some cases it looks to have beat the 7700k@5GHz in some 1080p games.

       

       

      RYZEN + 3600MHZ RAM - closing the 7700K gap in gaming? - YouTube

       

       

       

      Please add your thoughts!

        • Re: Infinity Fabric performance directly proportional to RAM speed equating to increase in FPS
          ajlueke

          I guess a lot of the games I play currently didn't show the latency effect.  Some games, like Gears of War 4 and Deus Ex, already run faster on the 1800x than their Intel counterparts.  Possibly due to those games DX12 implementations properly assigning work loads to all the cores.

           

          I'm not currently playing any games that have the lower than expected frame rates on Ryzen processors, so it is difficult to test if faster memory alleviates this issue.  I can take a look at it this weekend.  I don't think my ASRock board has a BIOS that allows for 3600 MHz settings yet, but I have gotten 3200 MHz working at 14-16-16-36 timings and 1.35V with two 8GB DIMMs installed.  When I install four DIMMs the best I have been able to do is 2667 MHz at the same 14-16-16-36 timings.

            • Re: Infinity Fabric performance directly proportional to RAM speed equating to increase in FPS
              goldstone77

              Just doing a benchmark with ram running at 2100 or 2400 then 3200, and post your FPS difference! Any game works! I also, wonder if lower latency ram has an effect?

                • Re: Infinity Fabric performance directly proportional to RAM speed equating to increase in FPS
                  savagebeastzero

                  Hey goldstone77,

                   

                  Ram timings directly correlate to better performance with Ryzen as well. Using the CPU-Z benchmark, with the CPU at 3.966 Mhz and memory at 2666Mhz 16-16-16-39-2T I scored a 20133, but by simply tightening up my timings to 15-16-16-34-2T at 2666Mhz, I scored 20681. This increased FPS by roughly an average of 3-5% in 3DMark and a few gaming benchmarks as well. Those are fairly substantial gains. So, make sure that users are also including their ram timings. At 2666/15-16-16-34, I actually out performed my 3000Mhz CPU-Z benchmark score at the same CPU multiplier by 15 points. So, definitely something that should be included with peoples results.

                  2 of 2 people found this helpful
                  • Re: Infinity Fabric performance directly proportional to RAM speed equating to increase in FPS
                    ajlueke

                    Hello goldstone77,

                     

                    I did some testing on the RAM per your request.  I ran the Gears of War 4 benchmark at two steps of 533MHz, giving me clocks of 2133Mhz, 2667MHz, and 3200 MHz.  I ran the benchmark three times at each frequency and used the final result of each.  The CPU is a R7 1800x at 4.1 GHz, and I did not alter RAM timings during testing.  They remained at 14-16-16-16-36.  Below are the final runs for each RAM frequency.

                     

                    First, 2133Mhz.

                    Gears of War RAM 2133.png

                    Next, 2667MHz

                    RAM 2666.png

                    Finally, 3200MHz.

                    GEars of War RAM 3200.png

                    I have compiled the results below in a table.  Overall I gained 15% in my fps by taking the memory from 2133MHz to 3200MHz.  Interestingly, the gains from 2667MHz to 3200MHz were approximately double those from 2133MHz to 2667MHz, despite an identical 533MHz increase.  Does this mean increasing returns as the frequency gets higher?  The plot thickens.

                    Bench.PNG

                    1 of 1 people found this helpful
                      • Re: Infinity Fabric performance directly proportional to RAM speed equating to increase in FPS
                        goldstone77

                        Ajlueke that's great! 3200MHz clock frequency seems to have a profound affect on FPS! This suggest if stability, becomes possible at higher clock frequencies much more performance might be achieved. Thank you very much for spending this time benchmark for all of us in great detail! 3200MHz at 14-16-16-16-36 are really great numbers! Could you test to see how much different the performance is with higher latency for this same resolution. Change the 14 to 15-16-16-16-36 and also 16-16-16-16-36. Most people seem to use CAS 16 RAM. Let's see if there is a notable difference in performance vs. latency at those frequency 2133, 2667, and 3200.  Also, would the use of 1080p resolution be possible as well. Again thank you for your time in helping me collect qualitative and quantifiable data to help answer my questions!

                          • Re: Infinity Fabric performance directly proportional to RAM speed equating to increase in FPS
                            ajlueke

                            It is interesting data.  I didn't really expect to see a 15% fps increase in a game benchmark, nor the more pronounced increase in the latter frequency boost over the former.  I chose the Gears of War 4 benchmark as it always displays both the CPU and GPU data.  CPU benchmarks are typically done at 1080p with all settings on low to ensure the CPU is the system bottleneck and the CPU FPS are what is being monitored.  With this benchmark, I can monitor changes to CPU FPS despite the fact that my R9 Fury X was by far the bottleneck in every scenario.

                             

                            I will run a couple different timings on my system when I get a moment and see how those impact the scores.  The Corsair Dominator kit I have doesn't seem to like odd values in the CL timing.  15 was wildly unstable, but both 14 and 16 have been rock solid through Firestrike stress test, a few hours of gaming, and days of uptime.  I can go back through and run the benchmark at 16 CL with those same frequencies.

                             

                            It is actually unfortunate that the gap between 2667MHz and 3200MHz was so wide.  I purchased a four DIMM kit, but have not been able to get beyond 2667MHz with all four DIMMs installed.  Your query has piqued my curiosity now though.  I am now more eagerly awaiting the UEFI update for my board that allows for frequencies beyond 3200MHz, it will be interesting to see what a stable 3400MHz could bring.

                            • Re: Infinity Fabric performance directly proportional to RAM speed equating to increase in FPS
                              ajlueke

                              I ran the Gears4 benchmark at the three frequencies again, but with timings of 16-18-18-18-36.  I will display the results in a table below.  On average, going from 16-18-18-18-36 to 14-16-16-16-36 increased fps by 5% (max 8% at 2133MHz) regardless of frequency.

                               

                              Overall, going from CL16 at 2133Mhz to C14 at 3200 MHz increased my fps in Gears 4 by 25%

                               

                                  

                              CPU Game (fps)CPU Render (fps)
                              RAM 2133 CL16182.6175.2
                              RAM 2133 CL14198.1189.2
                              RAM 2666 CL16206195.9
                              RAM 2666 CL14209.2197
                              RAM 3200 CL16216.9209
                              RAM 3200 CL14229.5217.3
                              Total FPS Gained (MHz+CL)26%24%

                               

                              1 of 1 people found this helpful
                      • Re: Infinity Fabric performance directly proportional to RAM speed equating to increase in FPS
                        ajlueke

                        Also, it is worth noting that the recent Ashes of the Singularity update fixed the performance issues associated with that game and Ryzen.

                         

                        AMD Ryzen Performance Update Released For Ashes of the Singularity - Legit ReviewsAMD Ryzen 7 1700 Test System

                         

                        The take away is that all developers can likely update their engines to scale appropriately with Ryzen.  Now whether they choose to do so is another story.  Most modern engines will likely be designed to access all the resources on Ryzen CPUs appropriately, especially if they use DX12 or Vulkan.  AMD also signed a deal with Bethesda to develop games specifically with Ryzen in mind.

                         

                        Bethesda partners with AMD to optimize games for Ryzen and Vega | PC Gamer

                         

                        Couple that with the fact that new Playstation and Xbox systems will launch this winter based off Ryzen APUs, it is a pretty safe bet that less than expected performance of Ryzen CPUs in games will fade away.

                        • Re: Infinity Fabric performance directly proportional to RAM speed equating to increase in FPS
                          savagebeastzero

                          Hey goldstone77,

                           

                          O.k, so I've compiled some more results since my last post and once again the data shows continued multi-threaded performance gains when tightening ram timings further. Yet, this time my adjustments weren't nearly as aggressive, but even still, the results were incredibly positive showing ever further continued multi-threaded performance gains in these ongoing tests, simply due to a few more timing adjustments.

                           

                          3975 ram timings 1515153263 cpuz benchmark.png

                          Newest Score!!!

                          15-15-15-32-63-2T Multi-Threaded Benchmark Score: 20711

                          Previous Score

                          15-16-16-34-63-2T Multi-Threaded Benchmark Score: 20681

                          Original Score

                          16-16-16-39-63-2T Multi-Threaded Benchmark Score: 20133

                           

                          I will continue working on these ram timing adjustments, attempting to tighten timings further, until I simply can't maintain a perfectly stable system regardless of bios adjustments made. Below I will list the exact memory and specifications I'm working with, along with my DIMM configuration for those interested in the results specific to these kits, in the case you either own them, or are attempting to determine the results to see if they're worth considering for your specific use case scenario with the Ryzen platform.

                           

                          CPU: Ryzen R7 1700X @39.75 multiplier = 3966.96Mhz

                          Motherboard: Crosshair VI Hero X370 / Bios Rev. (902)

                          Memory Information

                          Model: EVGA  DDR4-3000 (PC4-24000) CL15 Desktop Memory Kit

                          Rank: SR-D (Single Rank DIMM)

                          Manufacturer Part Number #: 8GX-D4-3000-MR

                          Channel Configuration: 4x8GB in dual channel mode = 32GB

                          Actual 100% Stable Memory Frequency Achieved Thus Far: DDR4-2666Mhz @ 4 DIMMS DC SR

                          Most Efficient Memory Timings Achieved as of 4/2/2017:  15-15-15-32-63-2T

                          Other Achievable Timings (#2): 15-16-16-34-63-2T

                          Other Achievable Timings (#3): 16-16-16-36-63-2T

                          Memory Voltage Used (All Configs): Manual Voltage 1.35 / Peak Voltage 1.35

                          Notable Memory Bios Settings: Extreme Phase Mode and 120%+ available power access

                          Note: 120% additional power access is for my testing only and is not required for the end users daily memory stability using these timings / Use 100% for proper daily Bios config.

                           

                          The above data is a work in progress and in future edits, I will provide more detailed information in regards to the adjustments made the Crosshair VI Hero bios and other data I feel important to note.

                           

                          Once I obtain all 100% stable memory timings @2666Mhz and am finished submitting this data to Asus for their data bank, I will move onto higher memory frequencies and repeat the process once again. I will likely move this data to its own thread, so as to not detract from the thread creators original vision. I will post the new threads link location in a response once (if) I move the data to a new location, in the case that anyone is searching for it. I just wanted this information to be available and published somewhere for the end users reference.

                           

                          Expect the next results within 24 hours of this reply. Have a nice day everyone and enjoy your time in the AMD the community.

                           

                           

                           

                           

                           

                          2 of 2 people found this helpful