3Dmark11 low combined score fix ?

Discussion in 'Benchmark Mayhem' started by Gripen90, Apr 25, 2012.

  1. Gripen90

    Gripen90 Guest

    Messages:
    869
    Likes Received:
    21
    GPU:
    2x RTX 2080Ti SLi
    My "problem" is still that while I get a normal graphics, cpu, physics score my overall combined score is too low.

    I know this issue has been addressed before but still I haven't found or heard of a solution or what is the actual cause of the low performance score.

    I know that on the Futuremark site it has been stated that problem is:
    The issue is that...
    - The "official new way of detecting SLI setups" fails on far too many SLI systems. Original SystemInfo that shipped with 3DMark 11 does it like this and SLI configurations just hung before Test 6. So we had to revert it back to the "legacy" method.
    - The "legacy way of detecting SLI setups" works, but it can't tell normal SLI from SLI that is in Single GPU mode (from NVIDIA control panel) or Multi-GPU config that has dedicated PhysX card.
    - Test 6 (combined) load changes with the number of GPUs. In order to properly render the scene with multi-GPU, each GPU must have a copy of the GPU physics simulation running on it (using Bullet/DirectCompute) - otherwise the physics code would have to shuffle data between GPUs and that would slow things down horribly. Hence the benchmark needs to know how many GPUs you have.
    - If Test 6 gets wrong number of GPUs (in your case, I guess it gets "3"), the number of DirectCompute simulations running is larger than the number of GPUs actually rendering the scene, causing notable performance loss.


    With my setup I originally got a combined score of around 5600 points (old forceware 263,14 driver) which seemingly is too low also according to 3Dmark own health-meter.

    Now several months have passed and I decided to once again fire up 3Dmark11 with the latest patch.
    Before I used Forceware 280,26 which got me this result: http://3dmark.com/3dm11/3253056
    With the latest WHQL Forceware 296,10 I got this result: http://3dmark.com/3dm11/3253056

    Still even with newer drivers and the latest patch for 3Dmark11 the low Combined score issue isn't solved with 3 cards.
    Well I managed to get a bit higher combined score though - but still not getting 6000points there as 3Dmark11 lists as the target score.
    http://3dmark.com/healthcheck/3dm11/3253056

    Is there any fix to this yet ? patch and newer drivers haven't help much since mid summer 2011.
     
    Last edited: Apr 25, 2012
  2. Gripen90

    Gripen90 Guest

    Messages:
    869
    Likes Received:
    21
    GPU:
    2x RTX 2080Ti SLi
    no one with a fix or ?
     
  3. ---TK---

    ---TK--- Guest

    Messages:
    22,104
    Likes Received:
    3
    GPU:
    2x 980Ti Gaming 1430/7296
  4. sykozis

    sykozis Ancient Guru

    Messages:
    22,492
    Likes Received:
    1,537
    GPU:
    Asus RX6700XT
    At stock clocks, i7 870 isn't even fast enough for GTX580 SLI....nevermind Tri-SLI...
     

  5. Gripen90

    Gripen90 Guest

    Messages:
    869
    Likes Received:
    21
    GPU:
    2x RTX 2080Ti SLi
    The CPU speed isn't the issue since it's in the normal range of the CPU type without overclocking.
    (Haven't got the time or lust/interest in overclocking hardware any longer)

    As addressed in the first post it's some sort of driver/futuremark card recognition issue when test 6 is initializing. This "bug" has existed for a year (almost) now, at I haven't seen any fix for it yet, but I don't know if I'm looking in the wrong place for the fix :)

    The combined score remains the same with 2 and3 cards.
     
    Last edited: Apr 26, 2012
  6. Lane

    Lane Guest

    Messages:
    6,361
    Likes Received:
    3
    GPU:
    2x HD7970 - EK Waterblock
    Last edited: Apr 26, 2012
  7. Gripen90

    Gripen90 Guest

    Messages:
    869
    Likes Received:
    21
    GPU:
    2x RTX 2080Ti SLi
    1) Is checked- it runs in normal (admin mode).

    2) That my be an issue. As stated in from futuremark there can be a problem with the 3 card detection in test 6.

    3) Yea I see. The only problem is that everything is overclocked so you can't see what the original "base" score is.
    I so often see people "overclock out of their problems" - if their score is low then then OC it up and assume it's fixed...but with the removal of the OC the problem still lies dorment.

    Still thanks for the hint, I'll look around at their forum :)
     
  8. Lane

    Lane Guest

    Messages:
    6,361
    Likes Received:
    3
    GPU:
    2x HD7970 - EK Waterblock

    Yeah, but if someone in the overclockers league have face this particular problem, they know how to fix for sure.. cause it will run them crazy to not have the score they should have.
     
  9. sykozis

    sykozis Ancient Guru

    Messages:
    22,492
    Likes Received:
    1,537
    GPU:
    Asus RX6700XT
    The problem is that the 3 GTX580's are being heavily bottlenecked by the processor....which even Hilbert experienced on his OC'd test rig during his review of the cards. This is why others are "overclocking their way out of problems". Whether you like it or not, that is the solution.
     
  10. Agent-A01

    Agent-A01 Ancient Guru

    Messages:
    11,630
    Likes Received:
    1,121
    GPU:
    4090 FE H20
    lol 3 580s on a 870.. bro my i7 2600k at 4.8 bottlenecks my cards slightly at these clocks. stock 580s are fine though. your graphics score is 1000 points more than mine(16000), either new cpu or take out the 3rd card
     

  11. ---TK---

    ---TK--- Guest

    Messages:
    22,104
    Likes Received:
    3
    GPU:
    2x 980Ti Gaming 1430/7296
    dont blame futurmark. its your cpu like I said before. oc your cpu and take out that 3rd card. look at the fps in the tests 2cards me 3 cards you. your 3rd card looks like its barely used. doesnt p55 m/b have something like half speed pci e slots or something like that?
     
    Last edited: Apr 26, 2012
  12. MM10X

    MM10X Guest

    Messages:
    4,240
    Likes Received:
    1
    GPU:
    3080 FE
    yes, p55 has a maximum of 24 pci-e lanes.
     
  13. Agent-A01

    Agent-A01 Ancient Guru

    Messages:
    11,630
    Likes Received:
    1,121
    GPU:
    4090 FE H20
    p55 has PCIe lanes that are rated 2.5GT/s compared to full speed lanes in sandy bridge 5GT/s. so yeah p55 x16 slot =s 8x sandy bridge slot.

    no it has 20 pci-e lanes. 16 lanes for the CPU alone and the other 4 dedicated to the chipset
     
  14. Gripen90

    Gripen90 Guest

    Messages:
    869
    Likes Received:
    21
    GPU:
    2x RTX 2080Ti SLi
    Yes at 1280x720 resolution where the Performance benchmark is running the CPU is bottlenecking a bit. But in every day gaming I use 1920x1200 with max amount anti-aliasing this minimizing any bottleneck heavily.

    This guy is running at ci7 870 stock clocks too but with 2-way SLi and still he gets a higher combined score than me 6600points vs 5900points.
    http://3dmark.com/3dm11/1644271
    That's why I think something is odd.

    Oh well I lost the 3Dmark logic when going from 3dm05/06.
    Unigine Heaven still scales like it should.


    Autumn 2011 I had ordered a P67 setup with core i7 2700K, but upon closer investigation I found out that I'd merely get 5-8% better performance than with the exsisting system at the settings I usually play at. That wasn't worth a $800 upgrade at the time.

    A new Ci7 3820 with a matching 3-way- SLi capable motherboard is something I'd like in a near future... but isn't likely to happen since I'm pretty broke.
     
    Last edited: Apr 27, 2012
  15. Agent-A01

    Agent-A01 Ancient Guru

    Messages:
    11,630
    Likes Received:
    1,121
    GPU:
    4090 FE H20
    Sorry but a 5-8% figure is way off. Not sure how you figured that in
     

  16. Gripen90

    Gripen90 Guest

    Messages:
    869
    Likes Received:
    21
    GPU:
    2x RTX 2080Ti SLi
    It would be in 1920x1200 and 16xmsaa/32csaa according to most reviews of the ci7 2700k vs 870.
     
  17. Agent-A01

    Agent-A01 Ancient Guru

    Messages:
    11,630
    Likes Received:
    1,121
    GPU:
    4090 FE H20
    So you are saying you compared a review with 3 580s with a 2700k vs 870? You cant do guess work with multiple cards. CPU bottlenecks become very apparent when adding another card, especially a 3rd. Couple that with the low bandwidth PCIe lanes of the P55 chipset and you are looking at at least a 30-40% gain in FPS all around.
     
  18. Gripen90

    Gripen90 Guest

    Messages:
    869
    Likes Received:
    21
    GPU:
    2x RTX 2080Ti SLi
    The p67 nor z68 does anything usefull in that regard. Also by increasing the resolution and maximizing AA the cpu bottleneck gets considerably lower.
     
  19. ---TK---

    ---TK--- Guest

    Messages:
    22,104
    Likes Received:
    3
    GPU:
    2x 980Ti Gaming 1430/7296
    OP thought you were here for help? Gl on your futuremark problem I'm out.
     
  20. Gripen90

    Gripen90 Guest

    Messages:
    869
    Likes Received:
    21
    GPU:
    2x RTX 2080Ti SLi
    The other aspect wasn't about the futuremark problem at all - it's about when you hit at bottleneck, and that bottleneck really depends on a lot of settings. I know that at the 3dm11 Performance settings the CPU is bottlenecking somewhat.... but at extreme setting at 1920x1080 the bottleneck is much smaller comparing CPU OC vs Non-OC. :)

    This guy is running at ci7 870 stock clocks too but with 2-way SLi and still he gets a higher combined score than me 6600points vs 5900points.
    http://3dmark.com/3dm11/1644271
    That's why I think something is still odd.:3eyes:

    I'm gonna pull one card out and test with 2-way SLi running stock.
     
    Last edited: Apr 27, 2012

Share This Page