1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Mantle Benchmark Thread

Discussion in 'Graphics Cards' started by LtMatt, Feb 3, 2014.

  1. LtMatt

    Caporegime

    Joined: Jul 12, 2007

    Posts: 31,863

    Location: United Kingdom

    Please post your Mantle Benchmarks here. I will add scores to the third post. Please no fanboy talk, no arguments about Mantle bugs or image quality issues etc. Please use the other thread for that. I will ask Mods/Dons to delete any off topic posts. This is for posting benchmark scores and discussion about said scores only.


    AMD Catalyst™ 14.1 Beta Driver for Windows®

    Download Link
    http://support.amd.com/en-us/kb-articles/Pages/latest-catalyst-windows-beta.aspx



    Please see the second post for the Battlefield 4 benchmark sequence ive created for us all to follow.
     
    Last edited: Feb 3, 2014
  2. LtMatt

    Caporegime

    Joined: Jul 12, 2007

    Posts: 31,863

    Location: United Kingdom

    Battlefield 4 Benchmark Video





    Instructions

    Mission 3 - South China Sea

    Reach the 'Escape The Titan Sequence' and jump onto the ship (wait for autosave to kick in) and then restart checkpoint to begin as i do. The benchmarking starts as soon as the checkpoint is loaded.

    1. Copy the video as closely as possible.
    2. At the start, get ammo from the crate then backup to use the plane wreckage as cover.
    3. Use one clip shooting the first wave of enemies.
    4. Spend the second clip shooting the guy to the left in the burnt out chopper. He has a grenade launcher and killing him quickly makes the other troops retreat.
    5. Once your squad mate says they're retreating move up to the ammo crate.
    6. Pick off any left over enemies and pick up more ammo before moving on to the next cover.
    7. Pick off a couple more enemies far away in the distance.
    8. Throw a grenade at the explosive barrel and slowly walk past it.
    9. Shoot any explosive barrels, once or twice till they catch light and continue on.
    10. Kill the last enemies as i do. Don't forget the final guy by the three explosives barrells. Do not shoot those.
    11. Proceed to the end of the Titan and look down at the boat. End the benchmark here.


    Benchmark Posting Instructions

    1. Please state your cpu clock info, clock speed.
    2. Please state your gpu info, clock speed and drivers used.
    3. Please include minimum and average fps.
    4. Please state your resolution & settings used.
    5. If posting a Mantle and DX11 bench please ensure the settings are matched across both API's.


    Useful In Game Commands

    FPS Monitor type - PerfOverlay.DrawFps 1

    Cpu+Gpu performance overlay type - PerfOverlay.DrawGraph 1

    To take a screenshot type - Render.Screenshot

    To enable a FPS log type at start of bench - PerfOverlay.FrameFileLogEnable 1

    To stop FPS log at end of bench type - PerfOverlay.FrameFileLogEnable 0

    A little tip. Whilst the level/map is loading you can type in PerfOverlay.FrameFileLogEnable 0 and press enter. Then when you need to start logging just bring up the console, press down arrow key on your keyboard or S and that last written command comes up. Saves you retyping it each time. Just change the 0 to 1 and hit enter to start recording.

    How To Work Out Your FPS

    Using the command PerfOverlay.FrameFileLogEnable above you can start recording fps figures. To read the output which is placed in your documents BF4 folder use the following tool below.

    Link to the app here guys (203.5KB):
    http://www.sendspace.com/file/c7qmr3

    Probably easiest to place the .exe in the same folder as the CSV file.
    Then open a console window in said folder and type:
    bf4parse <filename>

    e.g.
    bf4parse FrameTimeLog001.csv

    All credit to Cambofrog for this handy little tool which works out all the fps for you quickly via command prompt.

    Source
    http://forums.overclockers.co.uk/member.php?u=60932

    Example

    [​IMG]
     
    Last edited: Feb 3, 2014
  3. LtMatt

    Caporegime

    Joined: Jul 12, 2007

    Posts: 31,863

    Location: United Kingdom

    Mantle API Results

    1080P

    166 AVG. 4.7 MIN. CPU FX8320 @ 4Ghz - 7850 @ ???/???? - 14.1 Drivers sockky - Link
    110.4 AVG. 3.3 MIN. CPU 2500K @ 4.8ghz - 7970+7950 Xfire @ 1000/1400 - 14.1 Drivers sockky - Link
    94 AVG. 6.6 MIN. CPU 3570K @ 4.2ghz - 290 Xfire @ 1000/12500 - 14.1 Drivers EL_Jock - Link
    73 AVG. 25 MIN. CPU 2500K @ 4.0ghz - 290X @ 1000/1300 - 14.1 Drivers Sieb - Link



    1440P

    54.1 AVG. 38.2 MIN. CPU 2700K @ 4.8ghz - 290 @ 975/1250 - 14.1 Drivers LtMatt - Link
     
    Last edited: Feb 5, 2014
  4. LtMatt

    Caporegime

    Joined: Jul 12, 2007

    Posts: 31,863

    Location: United Kingdom

    DirectX 11 Results

    1080P

    156 AVG. 25.2 MIN. CPU FX8320 @ 4Ghz - 7850 @ ???/???? - 14.1 Drivers sockky - Link
    72.3 AVG. 11 MIN. CPU 2500K @ 4.8ghz - 7970+7950 Xfire @ 1000/1400 - 14.1 Drivers sockky - Link
    65 AVG. 25 MIN. CPU 2500K @ 4.0ghz - 290X @ 1000/1300 - 14.1 Drivers Sieb - Link
    63.1 AVG. 28.9 MIN. CPU 3570K @ 4.2ghz - 290 Xfire @ 1000/1250 - 14.1 Drivers EL_Jock - Link

    1440P

    44.8 AVG. 38.2 MIN. CPU 2700K @ 4.8ghz - 290 @ 975/1250 - 14.1 Drivers LtMatt - Link
     
    Last edited: Feb 5, 2014
  5. LtMatt

    Caporegime

    Joined: Jul 12, 2007

    Posts: 31,863

    Location: United Kingdom

    Ok guys post away. I'll let everyone decide how we do this. Do we stick to any specific settings or just make it a free for all? Stick to Ultra/Medium preset? Open to suggestions.
     
  6. SS-89

    Sgarrista

    Joined: May 13, 2007

    Posts: 7,837

    Location: London

    It's not going to be a ranked bench so it doesn't really matter does it? As long as Mantle vs Direct X are done at the same settings then we can get an idea of performance differences on different setups.

    Might be worth having a list with our usernames and setups linked to our benches? Then people with comparable setups can go straight to the relevant benches to see performance differences.

    i.e.

    SS-89 - 4770k@4.6, 290X Trifire (Linked to post)
     
    Last edited: Feb 3, 2014
  7. LtMatt

    Caporegime

    Joined: Jul 12, 2007

    Posts: 31,863

    Location: United Kingdom

    That's what i was thinking. Could actually have a DX11 leaderboard as well to compare. Ok so any settings allowed so long as they're matched between API's.
     
    Last edited: Feb 3, 2014
  8. EL_JOCK

    Hitman

    Joined: Aug 26, 2005

    Posts: 728

    Location: I can see OcUK from here

    Hi LtMatt,

    got ready and done the first run, however the output file only has frame times, cpu frame & gpu frame times not the fps is there a way to get that as well? this was on mantle

    Been using that file you linked to get the overlay on
     
  9. LtMatt

    Caporegime

    Joined: Jul 12, 2007

    Posts: 31,863

    Location: United Kingdom

    Yes there is, im trying to figure it out myself lol. I saw this post and this explains it, but ive yet to test it myself. Check it out and see if you can make sense of it. I'm about to run my own bench then ill test.

    http://www.rage3d.com/board/showpost.php?p=1337433810&postcount=684
     
  10. EL_JOCK

    Hitman

    Joined: Aug 26, 2005

    Posts: 728

    Location: I can see OcUK from here

    yeah i get it,

    you can remove the non game frames i.e cutscene section, if you use the cpu frame time as the key to when you where in game, will do a run work out my score and get back
     
  11. shankly1985

    Capodecina

    Joined: Nov 25, 2011

    Posts: 19,228

    Location: The KOP

    What we using to record the frame's?
     
  12. LtMatt

    Caporegime

    Joined: Jul 12, 2007

    Posts: 31,863

    Location: United Kingdom

    See that post i linked above. Once i fully understand it myself ill try to explain better. If anyone else has nailed it please post and ill add it to the OP.
     
  13. KompuKare

    Gangster

    Joined: Dec 31, 2010

    Posts: 257

    Right, that's not totally accurate. You want the reciprocal of that, so:
    1 / (framtimeInMs/1000)
    or if you frametime is in A1:
    =1 / (A1/1000).
    So 7ms would give you: 1 / (7/1000) = 142.85FPS for example.
    That's for FPS. For anything more complex (graphic CPU or GPU spikes etc.), you'd obviously have to more complex spreadsheeting, but converting ms to FSP is the above formula.
    And sorry, don't have BF4 so no benches from me.
     
  14. shankly1985

    Capodecina

    Joined: Nov 25, 2011

    Posts: 19,228

    Location: The KOP

    Cool beans, I'll wait on what you get :D
     
  15. LtMatt

    Caporegime

    Joined: Jul 12, 2007

    Posts: 31,863

    Location: United Kingdom

    Thanks for the explanation. Goes right over my head though lol.
     
  16. EL_JOCK

    Hitman

    Joined: Aug 26, 2005

    Posts: 728

    Location: I can see OcUK from here

    Right o, done a run on mantle and dx, used that rage3d thing you posted, thoughts and results so far,

    If your using your file and not setting it off in game as per the rage 3d, the best way to work out when the level had started is from the CPU frame time.

    You will get a massive run of 5's the 33.3's than back to 5's these are the cut scene, (could be down into the 1000's on excel sheet) before you start to get number around 10ish give or take, then it is the other way round when you go to menu at the end. i took a few extra either side to allow for the black screen just before you load in & load out.

    Using the rage3d you can work it out will update with a few easy tricks to get the numbers in a min

    The results

    Ultra preset with motion blur off & fov @ 90 1080p

    i5 3570k @ 4.2 (H100i) closed water loop
    sapphire Tri-x 290's crossfire (stock for that card)
    Win 7

    Mantle

    Max 229
    Avg 89.7
    Min 15.35

    DX11

    Max 130
    Avg 64
    Min 28

    So by the looks the avg is 39% better, max & min could just be a spike so not sure what we can take from that
     
    Last edited: Feb 3, 2014
  17. KompuKare

    Gangster

    Joined: Dec 31, 2010

    Posts: 257

    lol. Fair enough. Well, then let me try and explain it.

    Frames per second (FPS) is what we want because we are used to it, but what the log file gives us is milliseconds per frame. That is how long the each frame takes in milliseconds. So the time the frame takes. But we want to know now many frames we got in certain amount of time (frametime). So if we device our timeframe by the amount of time a frame took, we get how many frames we get in each timeframe. So 1 divided by 7ms = 0.142 Frames per millisecond. The 1000 is because we want the results in seconds.

    Oh, and actually easier is to divide 1000 by your ms: 1000 / 7ms = 142FPS.

    Edit: and don't forget the mins! From your rage3d link, the bit with "MIN(A1:AXX)". I mean, min frames is far more important. Although with the reports of occasional spikes in the Mantle beta, you want a more complex formula which ignores those spikes.
     
    Last edited: Feb 3, 2014
  18. EL_JOCK

    Hitman

    Joined: Aug 26, 2005

    Posts: 728

    Location: I can see OcUK from here

    i get it, to be honest the result i had the first time was right before i removed it, LtMatt i may be able to help peeps do this easy style give me a min
     
  19. LtMatt

    Caporegime

    Joined: Jul 12, 2007

    Posts: 31,863

    Location: United Kingdom

    Thanks guys. If you can make it any easier that would be great. I wonder if Frafs bench viewer would help? Pretty sure that accepts .csv files.
     
  20. Cambofrog

    Wise Guy

    Joined: Jun 2, 2007

    Posts: 1,180

    Location: Bradford

    I can write a small console app to parse the CSV file and calculate the min, max and average FPS if needs be guys.

    I'll see what EL Jock has in mind first though.