DirectX 9 CPU Benchmark Thread

Discussion in 'General Discussion' started by Thomas Jansen, Jul 22, 2019.

  1. Peter Kerenyi

    Peter Kerenyi Member

    Joined:
    Jan 30, 2015
    Ratings:
    +19 / 0 / -0
    Hello guys sorry if this does not fully belong here. I have a Ryzen 5 2600 (Stock) + RX 5700 XT. I am struggling with FPS in R3E. Would someone with a similar setup run a quick race start with the WTCR cars (27 AI) at Silverstone and tell me their performance? I am getting around 48-50 FPS for some reason. The game used to run a lot better on my i5 4460. And the Ryzen should have a better single core performance if R3E is limited to a single core. I don't know what changed... With DTM 1992 I get above 60 FPS. Both tested on high graphics preset but that should not make a difference.
     
  2. Balrog

    Balrog Well-Known Member

    Joined:
    Apr 10, 2015
    Ratings:
    +466 / 0 / -0
    A quick test with AGESA 1.0.0.4 on defaults (PBO + AutoOC +75 MHz):
    [​IMG]
    Finally passed the 9000 mark, but looks like I won't get more out of it without manual finetuning.
     
    • Like Like x 2
    • Winner Winner x 1
    Last edited: Nov 25, 2019
  3. Shodan

    Shodan Member

    Joined:
    Feb 28, 2017
    Ratings:
    +6 / 0 / -0
    Hi Sebastien, I would like to try this on my 2700x. I would think that you set different affinities for oculus server than for raceroom itself? Otherwise they would bite each other?
     
  4. Sebastien Brunier

    Sebastien Brunier Active Member Beta tester

    Joined:
    Dec 17, 2017
    Ratings:
    +38 / 0 / -0
    Oculus server is the only process I originaly keep the same affinity than RRRE64.exe I've done this because it is as important as the game itself anyway and by default Windows makes all processes compete against each other, and there was more drawbacks during my tests to set a differrent affinity for this one, so I kept same priority and same affinity.

    wmic process where name="OVRServer_x64.exe" CALL setpriority "above normal"
    Powershell "ForEach($PROCESS in GET-PROCESS OVRServer_x64) { $PROCESS.ProcessorAffinity=21760}"


    I've also tried this affinities works pretty well (the one I use atm)
    wmic process where name="OVRServer_x64.exe" CALL setpriority "above normal"
    Powershell "ForEach($PROCESS in GET-PROCESS OVRServer_x64) { $PROCESS.ProcessorAffinity=43520}"


    That last one is using SMT only for Oculus server on the cores used by the game (8,10,12,14 for R3E, 9,11,13,15 for Oculus Server) The idea behind is that as I set all my other programs I use during racing (Crewchief, Discord, Chrome, etc...) on the first 4 cores (threads 1 to 7 ProcessorAffinity=254) only the game and Oculus server are using the last 4 cores, and as the multi-threaded ratio in CPU-Z benchmark and Cinebench R15 is arround 10.6 for 8 cores I thought for a quick and dirty calculation that for 4 cores you get 5 so to speak. As you can see there is a lot of room for tests and creativity ;-)
     
    • Useful Useful x 1
  5. Shodan

    Shodan Member

    Joined:
    Feb 28, 2017
    Ratings:
    +6 / 0 / -0
    Thanks for the Writeup! I'll give it a try.
     
  6. Neawoulf

    Neawoulf Member

    Joined:
    Jan 25, 2017
    Ratings:
    +15 / 0 / -0
    Inspired by my recent performance issues in VR (and on the monitor as well) here are my benchmark results:

    Unigine Benchmark DX9low 3600X.jpg
    CPUZ Ram timings.jpg

    • CPU: AMD Ryzen 5 3600X
    • CPU OC: Stock (around 4300mhz max)
    • Memory: 16GB DDR4-3200 MHz Dual (G.Skill 3200mhz Ripjaw V) @2933mhz *edit* tested 3200 mhz now, the benchmark results are very similar */edit*
    • Memory timings: CL16-18-18-38
    • GPU: Nvidia Geforce GTX 2080
     
    • Like Like x 1
    Last edited: Dec 13, 2019
  7. Sebastien Brunier

    Sebastien Brunier Active Member Beta tester

    Joined:
    Dec 17, 2017
    Ratings:
    +38 / 0 / -0
    What is your mobo ? It's a bit odd to have a limit @2933 it could be that it needs a slight bump in voltage to work absolutely fine at the advertised speed of 3200.

    If your mobo is 1st gen X370 B350 or second gen X470 B450 it could be the mobo manufacturer only communicated on the standard CPU IMC speeds, the speed is mainly dictated by the IMC that is on the CPU itself unless your mobo have really bad traces (leads to crosstalk if PCB doesn't have enough layers) but that is unlikelly.

    Ryzen 3000 series standard memory speed is 3200.

    FYI my Trident Z kit refused to start at 3200 XMP profile (1.35v) always defaulting to 2133 JDEC profile, I had to bump voltage to 1.4 for the XMP profile to work, and now it is running at 3666c14 manual timings at 1.455v without a single memory error on a B450 (stopped the memorry test @1500%)
     
  8. Neawoulf

    Neawoulf Member

    Joined:
    Jan 25, 2017
    Ratings:
    +15 / 0 / -0
    I'm using an "old" Asus Prime B350 Plus because before my current CPU, which i bought about two months ago, i had a Ryzen 1600X. And since i was curious ... i just tried setting the RAM speed to 3200 mhz @ 1.35v and it seems to work.
    Either it's one of the new Bios updates or the new CPU (everything above 2933 Mhz was crashing all the time with the old 1600X) ... it looks very stable so far. I did that benchmark with the same settings again with the RAM on 3200 mhz, but the result is very similar.
     
  9. Sebastien Brunier

    Sebastien Brunier Active Member Beta tester

    Joined:
    Dec 17, 2017
    Ratings:
    +38 / 0 / -0
    Ryzen 1000 series had a 2666 memory speed support iirc, 2000 series was 2933 by design and 3000 series is 3200, and you're right agesa updates improved memory speed support.

    Now you have XMP support I think you should play with Ryzen Timing calculator from 1usmus it is available to download on

    https://www.techpowerup.com/download/ryzen-dram-calculator/

    Default XMP timings are meant for Intel processors, as it is an Intel tech and certification.

    Mobos usually do memory training to adapt the memory subtimings, but they are often lenient and meant to run not to achieve best performance.

    Manual timings gives best results because they are optimized for Ryzen, gains can be very significant.

    You need to select the memory type (You first have to do a bit od research on your specific kit (there are 3 different RipJaws kit at 3200 spec with different timings, so there are BDies for the 3200c14 but it could be SK Hynix for the lower specs kits) each ram chip manufacturer have his own chip specs and different OC potential, timing rules), the topology (single rank, dual rank, number of sticks), select the speed you want to achieve, click FAST profile buton and you get a list of timing to fix in your mobo BIOS.

    Always save first a profile of your existing configuration
     
    • Useful Useful x 1
  10. Neawoulf

    Neawoulf Member

    Joined:
    Jan 25, 2017
    Ratings:
    +15 / 0 / -0
    Thanks! This looks really interesting, also very complex. I think i'll take a closer look at this later. RAM timings etc. were always a bit of a mystery for me, but i'd be great if it helps.
     
  11. cortmarshal

    cortmarshal Member

    Joined:
    May 1, 2019
    Ratings:
    +8 / 0 / -0
    Finally upgraded my GPU from the 580 to a 1070Ti Duke, and as expected didn't notice any improvements in this heaven test but it's nice in-game.

    What really made the difference for me was another BIOS update, this time AGESA 1.0.0.4B. Boosts on the 3600 are finally where they should be, running about .075 faster across the board with a very slight power draw increase. It all inspired me to do some more memory tuning too. When I ran the benchmark again I was shocked.

    With PBO +100 I'm getting these results, up 8% from my previous score. I've retaken fastest minimum by quite a bit :p
    upload_2020-1-3_1-37-17.png
    upload_2020-1-3_1-39-42.png
     
    • Like Like x 1
    Last edited: Jan 3, 2020
  12. The Iron Wolf

    The Iron Wolf Well-Known Member

    Joined:
    Mar 18, 2015
    Ratings:
    +130 / 0 / -0
    • CPU: i7 8700K
    • CPU OC: 5.0Ghz
    • Memory: 16GB DDR4-3200 MHz Dual
    • Memory timings: CL16-18-18-38-560-2T
    • GPU: Nvidia Geforce 2080ti

      upload_2020-1-25_12-53-53.png
    upload_2020-1-25_12-54-3.png
     
    • Like Like x 1
  13. Quicksilver

    Quicksilver New Member

    Joined:
    Nov 20, 2016
    Ratings:
    +3 / 0 / -0
    [​IMG]

    9700K@5GHz
    3600 15/15/15/35 (overclocked from 3200 CL14)

    Still need to tighten secondary and tertiary timings
     
    • Like Like x 1
  14. Andy81

    Andy81 Member

    Joined:
    Jan 11, 2018
    Ratings:
    +5 / 0 / -0
    • CPU: AMD Ryzen 9 3900x
    • CPU OC: stock
    • Memory: 32GB DDR4 3600
    • Memory timings: CL16 19 19 39
    • GPU: Nvidia Geforce GTX 1080ti

    upload_2020-1-27_22-47-23.png

    upload_2020-1-27_22-48-9.png
     
    • Like Like x 1
  15. R.Noctua

    R.Noctua Well-Known Member

    Joined:
    Dec 12, 2019
    Ratings:
    +99 / 0 / -0
    This may sound ridiculous, but the best performance boost I got after reinstalling the W10. I was experiencing severe FPS sagging, trying to change settings, turning off secondary tasks... it's useless.
    I was tired of it and I just reinstalled Windows. Now I have stable 60 FPS with maximum settings (almost, only track animation removed).
    Raceroom is perfectly optimized for the DX9, but so there may be a problem with a trashed system.
     
  16. Larry Foster

    Larry Foster Member

    Joined:
    Apr 14, 2019
    Ratings:
    +13 / 0 / -0
    Does anyone know if the developers are considering moving to Directx 11 or 12
    Or just going to another game engine
     
  17. Bull Shark

    Bull Shark Well-Known Member

    Joined:
    Mar 14, 2019
    Ratings:
    +162 / 0 / -0
    Little to no chance they will go to directx 11 or 12. As far as I understand, it is almost not possible without rewriting the engine. But I could be wrong. :D
     
  18. Rmbonk86

    Rmbonk86 Active Member

    Joined:
    Feb 1, 2019
    Ratings:
    +25 / 0 / -0
    Upgraded a few weeks ago .. its been a long work in progress but here are my results.
    I still need to upgrade my GPU.. but budget has been the biggest issue.

    cpu has been OC'd from 3.1 to 3.6 btw


    upload_2020-1-28_19-58-13.png
    upload_2020-1-28_19-57-46.png
     
    • Like Like x 1
  19. Rmbonk86

    Rmbonk86 Active Member

    Joined:
    Feb 1, 2019
    Ratings:
    +25 / 0 / -0
    So on my last post above I had Voltage in bios on auto and also didnt have XMP profiles enabled and got the score of 4851.. Tonight I went into bios and changed the voltage to 1.35 and enabled XMP to profile 2 which is the 3200mhz. I ran the same Heaven benchmark again.. and got a 15 percent increase in score. I wasnt expecting that much of an improvement. [​IMG] upload_2020-1-31_0-40-47.png
     
  20. Sebastien Brunier

    Sebastien Brunier Active Member Beta tester

    Joined:
    Dec 17, 2017
    Ratings:
    +38 / 0 / -0
    The benchmark is very light on GPU and heavily CPU bound, so an improvement on RAM speed is giving more headroom limiting the CPU bottleneck, it is especially true on Ryzen that is very very sensitive to RAM speed.
    If you run default XMP profile, you can still get some improvements using the 1Usmus Ryzen Timing Calculator and fixing the timings yourself instead of auto timings. An extra 5% isn't out of reach.

    few explanations here:
    https://forum.sector3studios.com/in...pu-benchmark-thread.13473/page-15#post-191387