NVIDIA G80: General Performance Analysis

Discussion in 'Beyond3D Articles' started by Rys, Mar 26, 2007.

  1. Mark

    Mark aka Ratchet
    Regular

    Joined:
    Apr 12, 2002
    Messages:
    604
    Likes Received:
    33
    Location:
    Newfoundland, Canada
    You're right about that as well, I'm not sure why I reversed the axis or why I hadn't noticed before.

    just so everyone is aware, I am Ratchet, formerly Rage3D's hardware editor and the originator of those Rage3D click-able charts, so no need reporting to Rage3D that Beyond3D is stealing their chart style and presentation ideas :runaway:
     
  2. silent_guy

    Veteran Subscriber

    Joined:
    Mar 7, 2006
    Messages:
    3,754
    Likes Received:
    1,379
    Thank for the article!

    Would it be possible to add page titles to the page navigation drop box? This would avoid the binary search guessing game when trying to locate a specific page. :wink:
     
  3. Tokelil

    Regular

    Joined:
    Mar 27, 2002
    Messages:
    329
    Likes Received:
    0
    Location:
    Denmark
    I miss the page title in the navigation as well. Also the old fill-rate graphs was nice for spotting bottlenecks, so I hope we'll see those back in some form.
     
  4. Jawed

    Legend

    Joined:
    Oct 2, 2004
    Messages:
    10,873
    Likes Received:
    767
    Location:
    London
  5. Mark

    Mark aka Ratchet
    Regular

    Joined:
    Apr 12, 2002
    Messages:
    604
    Likes Received:
    33
    Location:
    Newfoundland, Canada
  6. Neeyik

    Neeyik Homo ergaster
    Veteran

    Joined:
    Feb 6, 2002
    Messages:
    1,231
    Likes Received:
    45
    Location:
    Cumbria, UK
    *Sniff* So long fill-rate graphs (echoing Jawed's comment). Even the one's in the linked R580 aren't as they used to be either, due to not using a linear scale on the "no. of pixels" axis.
     
  7. PeterAce

    Regular

    Joined:
    Sep 15, 2003
    Messages:
    489
    Likes Received:
    6
    Location:
    UK, Bedfordshire
    Nice article (if not very very late!).

    AA scaling graphs are really nice.

    Great game selection chosen :)

    Loved the 'General Shading Performance' page.

    Still no info on theoretical performance tests on Vertex Texture Fetch, Render to vertex buffer, Point sprites, complex vs simple vertex shaders, high-res shadow buffer rendering etc.

    Also no Dynamic Branching tests and no tests showing the difference between using Look-up tables vs PS maths (how balanced is G80?)

    You mention textured fillrates and shader tests using bogobench, Rightmark3D, in-house utility (sounds fun), Marko Dolenc's fillrate tester, but there are no results/graphs to back this up!

    Even just 'bogobench' results would of been nice - comparing the cards blend rates (using FP16 texures/MRTs) is an area that so little has been benched/discussed).

    So much deep analysis seems missing from this Beyond3D article!

    * < PeterAce starts to rock in the corner of the room for the lack of theoretical tests >
     
    #27 PeterAce, Mar 26, 2007
    Last edited by a moderator: Mar 26, 2007
  8. thana

    Newcomer

    Joined:
    Feb 26, 2003
    Messages:
    32
    Likes Received:
    2
    one more vote for the old-style fillrate graphs from me... those were some of the little things that always set b3d apart from every other hw-site out there. combined with the clickable heading concept (and with correct x-axis sampling) it would really stand out in terms of correctness and convenience.

    the rest of the article looks good, i especially like the %-drop graphs (so much nicer to grasp in a quick look than tables).
     
  9. corpsman

    Newcomer

    Joined:
    Mar 27, 2007
    Messages:
    1
    Likes Received:
    0
    No Oblivion :?:

    I am curious why Oblivion was not used. I see other games present in this artivle which seem to be CPU dependant, but i thought this article was about GPU's. I realize not everyone plays Oblivion, but it is known as a true tester of video cards, and since many people may use this article to help them decide if/how they should upgrade, wouldn't Oblivion be better? I am wondering if you can still do these tests using Oblivion, or is it too late? I am curious if i should upgrade my GX2 to an 8800GTX, or wait also for the next line of cards.

    I currently play FEAR and NS.
     
  10. Ailuros

    Ailuros Epsilon plus three
    Legend Subscriber

    Joined:
    Feb 7, 2002
    Messages:
    9,416
    Likes Received:
    178
    Location:
    Chania
    Excellent work. If it wouldn't be too much to ask, I'd prefer to have in the drop down menu underneath indications what each page contains. It helps when I need to link to a specific page and results are easier to find instead of having to browse through a large portion of the article to find it.
     
  11. AlBran

    AlBran Ferro-Fibrous
    Moderator Legend

    Joined:
    Feb 29, 2004
    Messages:
    20,650
    Likes Received:
    5,757
    Location:
    ಠ_ಠ
    sorry for being a newb, but what can the fillrate graphs tell you :?: :???:

    If you have the games in alphabetical order for the pages, could you list the games in alphabetical order on page 2? :p It's a really minor thing at the moment, but there are only certain games I'm interested that I want to see. That's just until you get the named index working.
     
  12. Neeyik

    Neeyik Homo ergaster
    Veteran

    Joined:
    Feb 6, 2002
    Messages:
    1,231
    Likes Received:
    45
    Location:
    Cumbria, UK
    They tell you, for one, whether the graphics card can keep up with the pixel demand placed up on it; they quickly point out where a test is CPU or GPU limited, and in the latter case possible causes for the performance bottleneck such as vertex processing or memory bandwidth.
     
  13. CarstenS

    Veteran Subscriber

    Joined:
    May 31, 2002
    Messages:
    4,797
    Likes Received:
    2,056
    Location:
    Germany
    Nice work - at last somebody who didn't shy away from the work to show performance of AA and AF separately!

    But one point's left for criticism: Which driver setting did you use on - especially on R5xx and G7x? After all G80 uses a very high filtering quality and competing last-gen products should be set to do their best too.
     
  14. fellix

    fellix Hey, You!
    Veteran

    Joined:
    Dec 4, 2004
    Messages:
    3,486
    Likes Received:
    397
    Location:
    Varna, Bulgaria
    Would be nice to include some S.T.A.L.K.E.R. numbers with Global Illumination option enabled. ;)
    Call of Juarez is also a good selection to bench with, and that applies to TD:U, too.
     
    #34 fellix, Mar 27, 2007
    Last edited by a moderator: Mar 27, 2007
  15. AlBran

    AlBran Ferro-Fibrous
    Moderator Legend

    Joined:
    Feb 29, 2004
    Messages:
    20,650
    Likes Received:
    5,757
    Location:
    ಠ_ಠ
    Thanks. So CPU limited is when the slope of the line is high and GPU limited is when the line is more flat? How do you determine vertex processing or memory bandwidth limitations? Different benchmarks? :oops:
     
  16. Skinner

    Regular

    Joined:
    Sep 13, 2003
    Messages:
    871
    Likes Received:
    9
    Location:
    Zwijndrecht/Rotterdam, Netherlands and Phobos
    Indeed.

    Btw, what does Global Illumination (heard the word before ;)) do and how can I enable it?
     
  17. Neeyik

    Neeyik Homo ergaster
    Veteran

    Joined:
    Feb 6, 2002
    Messages:
    1,231
    Likes Received:
    45
    Location:
    Cumbria, UK
    A total linear graph indicates that the graphics card is not pixel bound at any of the test resolutions; however, if the line tends to flatten off beyond certain resolutions then the card is becoming more limited by its ability to output (and/or process) pixels. By altering the core and memory speeds accordingly, one can then see if this limitation is due to insufficient memory bandwidth. As for being vertex limited, you really need to know what the test is like to pick this up.
     
  18. fellix

    fellix Hey, You!
    Veteran

    Joined:
    Dec 4, 2004
    Messages:
    3,486
    Likes Received:
    397
    Location:
    Varna, Bulgaria
    Check for description in Wikipedia.
    As for the game option, bring down the console and type r2_gi 1 to enable it (0 to disable) -- it definitely taxes the pixel shader pipes at maximum. Curious to see how G80 compares here, against R580 and G71.
    This particular title is very demanding upon the videomemory subsystem (thanks both of the large texture load and the deferred rendering).
     
  19. demalion

    Veteran

    Joined:
    Feb 7, 2002
    Messages:
    2,024
    Likes Received:
    1
    Location:
    CT
    Thoughts

    This largely echoes some other comments, but:

    • I think differences in AF implementation effects on image quality should be highlighted in this type of article when talking about AF performance, especially as there is a handy article to cross-reference for discussion (btw, that article's Page Navigation doesn't have page names for me, just page numbers...should be corrected?).
      Also, it is a disappointment to have a lack of clarity of whether the X1950 was using an AF mode more comparable to the 7xxx series or the 8xxx series...actually, I think both would have made its inclusion far more meaningful. I also think the implication that "out of the box" settings having any special meaning doesn't fit the level of knowledge this site has traditionally tried to encourage in its readership (and every hardware review site should encourage, IMO).
      Did I miss something in the article that addressed some of this?
    • I, too, miss the fillrate graphs. OTOH, I'm sort of excited by what could be accomplished when combining the presentation know-how of the "Rage3D fps graphs" with a fresh take on implementing the "Beyond3D fillrate graphs"...please give the same sort of creative thought to presentation/polish for including them. :)
    • Details about dynamic branching performance seems quite a major omission. What was the thought there? This seems to make the article more about the performance on the games selected rather than performance of the architecture, disappointing to me because it is both a significant factor in comparison between IHV approaches and (AFAIK) a significant point of progress over previous nVidia architectures.
      If I've missed discussion of this in an article, please correct me?

    Also, in general, I think more cross-referencing to articles dealing with certain things more in depth should be done, like the AF analysis I mentioned above (as well as AA, branching if I missed it in an article, etc.). I'd think this would both enhance user edification, as well as bump up ad revenues in a completely synergistic way (am I mistaken about the latter?).

    Oh, BTW: I hope a return to linking to comment threads directly from the front page snippet is in the cards?
     
    #39 demalion, Mar 28, 2007
    Last edited by a moderator: Mar 28, 2007
  20. trinibwoy

    trinibwoy Meh
    Legend

    Joined:
    Mar 17, 2004
    Messages:
    10,428
    Likes Received:
    426
    Location:
    New York
    Woo, thanks for the updated navigation links guys!
     
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...