HD 6970 & HD 6950 Reviews

Discussion in '3D Hardware, Software & Output Devices' started by onethreehill, Dec 15, 2010.

  1. Unknown Soldier

    Veteran

    Joined:
    Jul 28, 2002
    Messages:
    2,238
    Likes Received:
    33
    I don't understand why Techpowerup reviews Riddick but doesn't review MW2 or COD BOps.

    Riddick is a really, really old DX9 game and looks crap. Hell, review COD WOW, it looks miles better than Riddick.

    Review games people play on a daily bases.
     
  2. AlphaWolf

    AlphaWolf Specious Misanthrope
    Legend

    Joined:
    May 28, 2003
    Messages:
    8,414
    Likes Received:
    269
    Location:
    Treading Water
    Judging by the techpowerup crossfire review, antilles could be a beast.

    I'm sure you can find a dozen other sites out there that will review COD and MW2. It's actually not bad to include older titles, its not like they only reviewed with a few games they reviewed with more than a dozen.
     
  3. eastmen

    Legend Subscriber

    Joined:
    Mar 17, 2008
    Messages:
    9,975
    Likes Received:
    1,491
    hardware canucks seems to be the only one with 10.12 betas

    From what i can tell it seems to do much better there than on other sites. its one of the only sites where the 6950 consistantly out performs the 570
     
  4. Rangers

    Legend

    Joined:
    Aug 4, 2006
    Messages:
    12,322
    Likes Received:
    1,120
    Here is an "analysis" i did of Anandtech's review, game benchmarks.

    Code:
    1920X1200 resolution, % vs 6970 vs: 
    
    
                             580                570
    
    Warhead            +10.5%          -5.8%
    Battleforge         +41.3%          +23.7%
    Metro 2033        +7.7%            -6.6%
    Hawx                 +47.0%         +23.9%
    Civ 5                  +18.9%         +18.9%
    Battlefield BC2     +19.9%         +3.1%
    Stalker COP         +6.7%           -7.7%
    Dirt 2                  +43.3%         +24.7%
    Mass Effect 2       +15.9%         -3.2%
    Wolfenstein         +2.7%          -6.7%
    
    Total                                       +6.4%
    Total minus 3 outliers                 -8%
    Hmm didn't format exactly right but you get the idea. I have several thoughts.

    First, Anand uses some off games. Hawx is oldass, and has a sequel. The reviewer basically seemed to state he uses it because it's a flight game with a built in benchmark. Guess he was too lazy to do some fraps runs or something but wanted to include a flight game. Wolfenstein is another oldass game. No doubt it's in there to give a look at Open GL but IMO just replace it with something newer.

    Next we notice how AMD gets killed in certain games, the "outliers", and it will make them look bad and kill any averages. Here the outliers are Dirt 2, Hawx, and Battleforge. We see 580 crushes 6970 by over 40% in all those. I think it would behoove AMD to clean those up if possible. You could probably make an argument Nvidia's more consistent performance alone is a positive. For example 570 is not going to get killed in any game, it will perform decently seemingly in all, which could factor in a buying decision.

    However one can debate the value of those outlier benchmarks. In Dirt 2 the 6970 is getting 77 FPS, in Hawx 117, and in Battleforge, 64. Are higher FPS really of much value here? Maybe one could argue in Battleforge it's getting close. In 2560 6970 gets 48 FPS, well below 60, but then again 580 is only getting 62 at 2560. These games all scale well, so at least one could say perhaps they could be future indicators that certain other games may exhibit similar discrepancies, I suppose.

    Then we get games that are odd and dont scale well. Civ 5 prime example, as you see the 580 and 570 and even 480 are exactly the same speed there. It seems to run into several scaling plateaus. About 42 FPs for single cards and lower 50's for CF/SLI. Nonetheless 6970 only gets 35 FPS there vs 41.5 for Nvidia so it is a tangible difference.

    If you simply add up all the % and average, you get a disappointing result, 6970 is 6.4% slower than even GTX570. But again that seems a bit misleading perhaps, throw out the three outliers and now 6970 is a solid 8% faster than 570. If you added up the percentages for 580, the outliers would cause an even much larger swing, since they are even more "outlying" for 580.

    If you want to be kind to AMD, at least in the AT review, the most difficult games, Stalker, Metro, and Crysis, are the most positive for AMD.

    Also, 6970 is faster than 570 in 5 games and slower in 5. It loses all 10 to 580.

    You really kind of have to get an overall picture. That's just the AT review, I imagine every review could present it's own peculiarities.
     
    #24 Rangers, Dec 15, 2010
    Last edited by a moderator: Dec 15, 2010
  5. Rangers

    Legend

    Joined:
    Aug 4, 2006
    Messages:
    12,322
    Likes Received:
    1,120
    I could not see any major difference with AT in some of the games I glanced at. For example, still losing to 58- by ~50% in Dirt 2 at Hardware Canucks, etc. The 10.12 drivers, dont expect major improvements I'd guess.
     
  6. eastmen

    Legend Subscriber

    Joined:
    Mar 17, 2008
    Messages:
    9,975
    Likes Received:
    1,491
  7. onethreehill

    Newcomer

    Joined:
    Mar 10, 2010
    Messages:
    182
    Likes Received:
    0
  8. Unknown Soldier

    Veteran

    Joined:
    Jul 28, 2002
    Messages:
    2,238
    Likes Received:
    33
  9. onethreehill

    Newcomer

    Joined:
    Mar 10, 2010
    Messages:
    182
    Likes Received:
    0
    #30 onethreehill, Dec 15, 2010
    Last edited by a moderator: Dec 15, 2010
  10. Kaotik

    Kaotik Drunk Member
    Legend

    Joined:
    Apr 16, 2003
    Messages:
    8,166
    Likes Received:
    1,836
    Location:
    Finland
    So, which sites give average results nowadays?
    Techpowerup, Computerbase - any others?
    Oh, and hardwarecanucks have some sort of average difference chart in the end
     
  11. CarstenS

    Veteran Subscriber

    Joined:
    May 31, 2002
    Messages:
    4,797
    Likes Received:
    2,056
    Location:
    Germany
    Damien at Hardware.fr also has a „rating” posted. Dunno what all this fuss about ratings is, though. 10 and 90 also averages 50, as does 40 and 60, whereas in the former, you'll definitely won't have any fun playing 50 percent of the games. ;)
     
  12. Kaotik

    Kaotik Drunk Member
    Legend

    Joined:
    Apr 16, 2003
    Messages:
    8,166
    Likes Received:
    1,836
    Location:
    Finland
    It gives one better view on the overall situation than single benchmark, even if there's that 10 and 90 case somewhere screwing things up.
     
  13. CarstenS

    Veteran Subscriber

    Joined:
    May 31, 2002
    Messages:
    4,797
    Likes Received:
    2,056
    Location:
    Germany
    Yes I know that many people feel that way. But personally, I think performance (and that's not just benchmark bars) of a modern graphics card just cannot be represented truthfully by just one value. (Sorry for being OT, will stop that now)
     
  14. Evildeus

    Veteran

    Joined:
    May 24, 2002
    Messages:
    2,657
    Likes Received:
    2
    I think the average result of HFR is good to have an idea on the difference of performance.
     
  15. Kaotik

    Kaotik Drunk Member
    Legend

    Joined:
    Apr 16, 2003
    Messages:
    8,166
    Likes Received:
    1,836
    Location:
    Finland
    Well, it's not completely OT, benhcmarking methods, measurements etc are kinda part of reviews.

    Of course seeing a FPS graph of each test would be the best (and "realtime FPS graph" even better), but at least personally I see the usual min/avg/max useless excluding the avg value, as min can be caused by single spike or hundred spikes and we can't know which it is - I rather take 1 bigger spike vs 100 (slightly) smaller ones
     
  16. DavidGraham

    Veteran

    Joined:
    Dec 22, 2009
    Messages:
    2,749
    Likes Received:
    2,515
    What is up with the rage3d review ? they say HD 6970 beats GTX 570 by 15% , however in their own benchmark suit , GTX 570 consistently beat HD 6970 !
     
  17. Jawed

    Legend

    Joined:
    Oct 2, 2004
    Messages:
    10,873
    Likes Received:
    767
    Location:
    London
    http://www.techreport.com/articles.x/20126/2

    So that would appear to mean that the overhead of 24 SIMDs, in comparison with 20 SIMDs in HD5870, for "the same performance" is not good for graphics.

    Other stuff on that page: ECC for off-die memory only, task-parallel compute kernels and decent DMA.

    Of course all these changes were planned before AMD had sight of Fermi. Well, maybe not the ECC - I expect that's a software solution, particularly as there's no internal ECC...

    Oh and the epic stupidity of the Bit-tech review bears special attention. They can't even get the price right. Fuckwits.
     
  18. Mize

    Mize 3dfx Fan
    Moderator Legend Veteran

    Joined:
    Feb 6, 2002
    Messages:
    5,048
    Likes Received:
    1,097
    Location:
    Cincinnati, Ohio USA
    No doubt there are some really poorly executed reviews out there. I think the results are what most of us expected as of about 5 days ago. Ultimately it's a good product at the right price, but it doesn't have me diving for my wallet. With a 5870 now the only really appealing upgrades (as in "wow" upgrades) out there are the 580 GTX or CF 6950s and both just don't seem worth it at >$500.

    How long till Antilles? Can we get a $499 on that one? :)
     
  19. no-X

    Veteran

    Joined:
    May 28, 2005
    Messages:
    2,297
    Likes Received:
    247
    I think the drivers are quite bad. It would be logical to expect more consistent performance (compared to Fermi) than HD5870 offered. Arithmetic performance about the same, so where Cypress shined, Cayman should be at least as good. Texturing is 24% faster and tessellation, where Cypress falled significantly behind Fermi, was improved mostly.

    But in fact Cayman is more incosistent in performance, than Cypress. Significant performance drops with MSAA in some games, performance drops at 2560*1600 in others, sometimes it falls behind Cypress... but in other scenarios performance is very close (±5%) to GTX580 (Stalker, Metro, Just Cause 2, Crysis Warhead, F1...), so I'm quite convinced, that we'll see some improvements in drivers (hopefully soon)
     
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...