NEW Benchmark comparison between AMD & Nvidia

Discussion in 'General Discussion' started by John, May 2, 2014.

  1. DrR1pper

    DrR1pper Registered

    Joined:
    Apr 29, 2012
    Messages:
    3,286
    Likes Received:
    23
    OH MY ****ING GOD!......I'VE FIGURED IT OUT!!!!!!!!!!!!

    Sincere apologies for the expletives, but it's just such a ****ing relief to finally know what's the problem after all the hours of trying everything short of hopping on one leg during the benchmark to see if that would fix it.

    Ok.....are you ready?.....*drum roll*.......it's the civic + silverstone replay file we've been using. Somehow it's corrupt.

    How do i know this? I went and tested the old Clio with Loch Drummond benchmark and guess what.....exact performance as i measured over a year ago with same settings!!!!!!

    I then go back to the civic + silverstone replay and benchmark comes back with higher score again suddenly. I then quit out of the replay and run it again and then suddenly 16% fps loss (hint-hint same as with Saabjock that one time he went from 3.8Ghz to 4.0Ghz and got a 16% fps boost and assumed it was the cpu!) I guess around something like 1 in every 20 times i suddenly regain the 16% fps loss with a fresh reload of the replay and maybe 1 in every 50 times get an insane 35% fps boost. Yeah, that replay file we've been using is extremely screwy for some reason.

    How i figured this out though was by finally decided to try the old clio + loch drummond replay again (oh how i wish i did this first). 5 times in a row reloading from fresh and 100% consistency in results and matching how i scored over a year ago with that replay file.

    **** ME! What a giant pain in the bottom but just glad to have finally figured it the **** out.

    This also explains why some people with the same cards are getting wildly different results!
     
    Last edited by a moderator: Sep 28, 2014
  2. Spinelli

    Spinelli Banned

    Joined:
    Jan 28, 2012
    Messages:
    5,290
    Likes Received:
    28
    Jesus Christ, I only said it 62 times that a new benchmark has to be started with the new build. Replay files - or any sort of file, bit, or byte - from an old build?? That's a no, no.

    Brand new replay, brand new player PLR file, brand new shaders and CBash files (just delete all the files in those folders), etc. etc

    Anyways you found it now, and that's all that matters, just remember for the future. :)
     
    Last edited by a moderator: Sep 28, 2014
  3. DrR1pper

    DrR1pper Registered

    Joined:
    Apr 29, 2012
    Messages:
    3,286
    Likes Received:
    23
    No, the even much older Clio + Loch Drummond replay works absolutely fine. Something just went wrong with the replay file John made.

    We cannot rely on any of the results any of us have received with the Civic + Silverstone replay.
     
    Last edited by a moderator: Sep 28, 2014
  4. Marek Lesniak

    Marek Lesniak Car Team Staff Member

    Joined:
    Oct 6, 2010
    Messages:
    1,583
    Likes Received:
    101
    DrR1pper, I told you already to do Clio @ Loch, as Civic @ Silver seemed weird ;)

    His test was done on Build 660. We cannot compare results from different builds, like you just did.

    So, I did the benchmark on B660. Here are the results (with older and the newest driver)

    Gigabyte GTX 780 OC
    core: 954MHz; Boost (effective) 1084MHz
    Mem 1500MHz (6000MHz)

    nV driver: 331.65
    Avg: 68.474 - Min: 53 - Max: 92

    nV Driver: 344.11
    Avg: 69.120 - Min: 53 - Max: 94

    As you can see, my average score is only ~1fps lower than what you have predicted. I think we can say, my system is fine :)

    ------------------------------------

    Nuno's score (for reference)

    GPU = Nvidia GTX 780 Ti
    CPU = Intel I7 3770K @ 4.5ghz
    RAM = 4x4GB DDR3 2133mhz
    HDD = OCZ Agility 3 120GB
    Driver = 337.50

    1920x1080
    Core: 876MHz, boost effective probably ~970-980MHz
    Mem: 1750MHz (7000MHz)

    Avg: 79.407 - Min: 64 - Max: 103
    ____________

    1920x1080 (GPU + 140mhz core / +340mhz mems)
    Core: 1016MHz, boost effective probably ~1140-1150MHz
    Mem: 1920MHz (7680MHz) ??

    Avg: 89.859 - Min: 71 - Max: 126

    -------------------------------------------------------------

    Did that test and here are my scores:
    8763
    Graphics 10449

    8700 is affected by CPU, so don't be so attatched to it. Graphics score is the most important and if you look at various results, stock 780 scores about 9000-9300 points, so mine, beeing slightly overclocked, performs as expected. Case closed, OK? We did enough OT in this thread already :)
     
  5. Prodigy

    Prodigy Registered

    Joined:
    Nov 2, 2012
    Messages:
    938
    Likes Received:
    53
    Ok then it's time to do a new benchmark replay, with current build and maybe another car+track combo. I guess civic+silverstone is like default since it's in demo too. But we could do something else also, as long as it is ISI official.
     
  6. DrR1pper

    DrR1pper Registered

    Joined:
    Apr 29, 2012
    Messages:
    3,286
    Likes Received:
    23
    Well, i started new benchmark thread and chosen to use the clio with loch drummond replay. I gave my reason in the OP but i didn't think about the fact it might be best to use what's available in the freely and readily available demo. Hmmm. But then wondering is it really needed? This benchmark isn't designed to find out live performance but simply relative performances between cards.

    http://isiforums.net/f/showthread.php/21976-NEW-Benchmark-comparison-between-AMD-Nvidia-2-0
     
  7. Saabjock

    Saabjock Registered

    Joined:
    Nov 13, 2011
    Messages:
    898
    Likes Received:
    14
    Yes. The PCI-e 3.0 fix was applied right AFTER the first test...which was done at 3.8GHz and PCI-e 2.0.
    To apply the fix:

    HKEY_LOCAL_MACHINE/SYSTEM/ CurrentControlSet/ Control/Video

    The folder associated with a graphics card will have three or more subfolders (depending on how many PCI-E slots available on the motherboard).
    The values listed for each subfolder will be 0000, 0001, 0002, 0003, 0004, and Video.
    Review only the subfolders labeled as "0000". You will know you have selected the correct "0000" subfolder when you see a registry labeled “DriverDesc” with a value that matches the graphics card you have installed. Example, the value in my “DriverDesc” registry value reads “NVIDIA Geforce GTX 680”.

    Right click on the folder labeled “0000”. Select “New”, then select “DWORD (32-bit) Value“, then enter “RMPcieLinkSpeed” for the name of the registry.

    Right click the “RMPcieLinkSpeed” registry you just created, then select “Modify”, then enter “4” as the data value and verify that the “Hexadecimal” option is checked under “Base”, and then select “OK”.

    Repeat steps 9 and 10 for each graphics card associated folder (named “0000”)

    Once you have completed creating the RMPcieLinkSpeed registry for each card, close the Registry Editor window and restart your computer.
     
  8. DrR1pper

    DrR1pper Registered

    Joined:
    Apr 29, 2012
    Messages:
    3,286
    Likes Received:
    23
    Ok but since we now know the replay file is corrupt and you suffered the same random 16% fps loss as me, I think it was just a coincidence that you applied the pci-e fix at the same time.
     
  9. speed1

    speed1 Banned

    Joined:
    Jul 26, 2012
    Messages:
    1,858
    Likes Received:
    0
    How we can know it was corrupted and how to compare against a different track like Loch D., maybe Silverstone is just more performance hungry and not really corrupted, and how to explain the bad solo live perfromance of mine on Silverstone with triples homie ?
     
  10. DrR1pper

    DrR1pper Registered

    Joined:
    Apr 29, 2012
    Messages:
    3,286
    Likes Received:
    23
    Is that 10449 gpu score in firestrike?
     
  11. speed1

    speed1 Banned

    Joined:
    Jul 26, 2012
    Messages:
    1,858
    Likes Received:
    0
    It is lower than the 290 score stock and just with 10% OC of mine. Would be nice to use those scores for comparison as well.
     
  12. Saabjock

    Saabjock Registered

    Joined:
    Nov 13, 2011
    Messages:
    898
    Likes Received:
    14
    You guys are way too funny. lol
    How come you didn't just do a new benchmark with 860 to begin with?
     
  13. Marek Lesniak

    Marek Lesniak Car Team Staff Member

    Joined:
    Oct 6, 2010
    Messages:
    1,583
    Likes Received:
    101
    Because it is good to verify scores against an already existing database. If, with a new benchmark, we would notice weird scores, in the end we probably end up doing old benchmarks for verification anyway.
     
  14. speed1

    speed1 Banned

    Joined:
    Jul 26, 2012
    Messages:
    1,858
    Likes Received:
    0
    DrR1pper here you can see what i meant in the context of that a GPU can't driven by itself. You can see very well the difference between the CPU frequenz of 3800 ( previous tests ) and 4000Mhz. It is just a plus of 200Mhz but still has an influence on the GPU score with and without OC of the GPU, and i'm sure it would increase with a more Powerfull CPU/Base until the GPU reaches the limit of it's cababilities.

    Standard - GPU 10326
    View attachment 14397

    GPU OC 10% - GPU 11146
    View attachment 14396


    I agree with you and rogue22 that this has nothing to do with the overall performance of rf2 in realtion to AMD's as it seems ( see the other thread why i think so ), but i just would like to show you that it always is a combination of hardware and it's total performance. :)

    And compared with this graph, the 290 is still very powerfull, even if it is just a non X and driven by an outdated sys, but rf2 let it look like scrap.

    [​IMG]
     
    Last edited by a moderator: Sep 28, 2014
  15. DrR1pper

    DrR1pper Registered

    Joined:
    Apr 29, 2012
    Messages:
    3,286
    Likes Received:
    23
    That can't be right Speed1.

    You showing at 3.8Ghz:

    GPU score: 10326
    CPU score: 5859

    Then with the CPU at 4.0Ghz (a 5.3% increase) with your graphics card held constant:

    GPU score: 11146 (an 8% increase)
    CPU score: 5880 (a 0.4% increase) <- Essentially no statistically relevant change...i.e. can be put down to the natural tiny variances with repeat runs

    Non of this makes any sense and tripped my bs alarm. How is it that when you overclocked the cpu your cpu score didn't even change (which it should have and in a near linear fashion...i.e. cpu should have gone up around 5.3%) and yet instead your GPU score somehow goes up 8%.

    Without making accusations, the only way i can see that you cpu score remained constant and yet your gpu score raised 8% is if you raised only your graphics card clock speeds (maybe around 10-12%).

    If however what you are showing me to be true, you are the only example disproving the rule that i know of. I'm not convinced this is the case.

    Can you repeat those with cpu-z graph of clock speed and afterburner core clock and mem clock graph whilst repeating the benchmarks please. What you're showing defies belief.
     
    Last edited by a moderator: Sep 28, 2014
  16. Spinelli

    Spinelli Banned

    Joined:
    Jan 28, 2012
    Messages:
    5,290
    Likes Received:
    28
    Speed1, you need to use some sort of monitor program to monitor and log your CPU clocks, and for each core/thread while doing these tests. As DrR1pper said, something seems really messed up with your CPU as even the overclock didn't do anything in the Firestrike CPU score...

    Also, what version of Windows are you running? Set your windows power mode to "High performance", and if using Windows 7, then please download a program which will allow you to edit and disable "core parking" in the power settings for the high performance profile. You don't need to do the "core parking" fix with Windows 8 or 8.1, those operating systems are much, much smarter when it comes core parking.
     
  17. speed1

    speed1 Banned

    Joined:
    Jul 26, 2012
    Messages:
    1,858
    Likes Received:
    0
    What you guys are seeing there is true, how and why is that exactly i don't know.
     
  18. DrR1pper

    DrR1pper Registered

    Joined:
    Apr 29, 2012
    Messages:
    3,286
    Likes Received:
    23
    Well please repeat the results Speed1 and take additional screenshots with the CPU and GPU clock speed graphs from the firestrike benchmark please.

    As i showed you here (http://isiforums.net/f/showthread.p...AMD-Nvidia-2-0?p=309890&viewfull=1#post309890), your 8% rise in Firestrike is too coincidental with the 8% rise you got with the solo live benchmark with 10% GPU overclock.

    I think you made a sincere mistake during firestrike benching.
     
  19. speed1

    speed1 Banned

    Joined:
    Jul 26, 2012
    Messages:
    1,858
    Likes Received:
    0
    Ok i'm going to confuse you one more again because the GPU score is higher on standard than previously, but the GPU scores are quite equal with both CPU frequencies :p The only thing i changed today was the Auto CPU clock setting in the bios to manual because i tought it could be a reason for that rf2 don't recognizes my CPU and the indicated CPU frequency varies with each restart of rf2, but it has not changed the fact.

    GPU standard - CPU @3800Mhz

    View attachment 14398

    GPU standard - CPU @4000Mhz

    View attachment 14399
     
  20. DrR1pper

    DrR1pper Registered

    Joined:
    Apr 29, 2012
    Messages:
    3,286
    Likes Received:
    23
    Good. That laws of physics have not changed. :p
     

Share This Page