NVIDIA GeForce 6600 GT Review

Discussion in 'Beyond3D Articles' started by Dave Baumann, Sep 7, 2004.

  1. PatrickL

    Veteran

    Joined:
    Mar 3, 2003
    Messages:
    1,315
    Likes Received:
    13
  2. Hyp-X

    Hyp-X Irregular
    Veteran

    Joined:
    Feb 6, 2002
    Messages:
    1,170
    Likes Received:
    5
    I found the GeForce 6600GT vs 5800Ultra comparsion interesting

    Both are clocked @ 500MHz
    Both has RAM @ 500MHz
    Both can output 4 pixels only (a point criticized a lot in 5800)
    Both can output 8 "Zixels"
    Both have "8 TMU" (yes it's not a correct term)
    Both has 128bit bus (another point criticized in 5800)

    Now it's clear that neither it's 4px output nor it's 128bit bus was the problem with 5800 - only it's pixel pipeline performance.

    Also I think if you'd remove the PS3.0 functionality and FP filtering/blending you'd get a similar transistor count.
    That's the card that nVidia should have made instead of 5800Ultra...
     
  3. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    And why would that be interesting?
     
  4. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0

    Another interesting thing I noticed was how the Fillrate achievement was much greater on the 6600GT yet the same bandwith. Think this might have something to do with the high latency on DDR2?
     
  5. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,079
    Likes Received:
    648
    Location:
    O Canada!
    Well, obviously I don't know what their cost structure is like so I can' really say much, but I did wonder about the die size for a mainstream part when I ripped the HSF off. It will be interesting to see the die size of RV410 - since the FX series, and beyond, it looks like NVIDIA have been carrying a transistor premium over ATI and we'll see if that continues. I've wondered about the costs for NV40 as well, so it will be interesting to see what their margins are like when they have fully transitioned to NV4x.

    Not entirely sure - it might be vertex performance actually. I'll have to have a think about that.

    Interesting you noticed that - I was talking about that with NVIDIA on friday when I met with them, as I'd seen it the nigh before when I was doing some testing - I forgot to mention it in the review though!

    There's the capability for more in the chip table (or the detail chip/board pages from the tables) than is actually displayed - it's only displayed if there is data. Not really done die sizes before since I've not have very accurate data.

    I wrote that after taking the numbers - after I wrote it I realised I should have included the 5800 Ultra in the number, but didn't have time to in the end (Hmmmm, perhaps I'll do another article comparing them...?)

    WRT the criticism’s, in isolation I don’t think they would have been criticised - my beef with the 4x2 was the fact that when I pointedly asked the configuration they told me 8x1 (likely as a reaction to R300), probably thinking that I as "Press" (in general) wouldn't be able to distinguish where this type of arrangement is in the grand scheme of things in game utilisation at that point. I think the 128-bit itself wasn't actually the issue, more the fact that in comparison to 9700 it lost out on a few GB of bandwidth.

    In general though, I think there are a number of reasons why 6600 is landing out much better - first, because of its 2 quad internals, I think it can reject more pixels at the top of the raster pipe, then the pipelines are a little more flexible and I also believe a few elements such as the colour compression on NV30 were "not quite as optimal as they should have been", I'd also heard that fixed function fog wasn’t correct, so a lot of the time that ended up going through the fragment pipe which slowed things down (IIRC there were a number of initial benchmarks where people noted fog was missing?).

    Mmmm, perhaps I should do a comparison...
     
  6. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    And why would that be interesting?[/quote]
    Perhaps because SA alone isn't indicative of cost? Compare large SA with low layers & small SA with high layer #. Chip SA x layer # required for useful comparison.
     
  7. Bjorn

    Veteran

    Joined:
    Feb 6, 2002
    Messages:
    1,775
    Likes Received:
    1
    Location:
    Luleå, Sweden
    I seem to remember that a lot of people critized the NV30 for it's 4*2 architecture. Not only for the deception (4*2 = 8*1) but thinking that it was the one of the problems with it's bad performance. And seeing what you can do on a 128 bit bus and 4*2 architecture then you again have to wonder what they did when they designed the NV3x. (Not completely fair of course since better archtectures, compression techniqes, rejection.. doesn't invent themselves)

    As a sidenote, it'll be rather interesting to see how the low end NV4x, RV4xx cards will perform when they are available. Perhaps even the 64 bit bus versions won't perform as bad as we might think.
     
  8. MuFu

    MuFu Chief Spastic Baboon
    Veteran

    Joined:
    Jun 12, 2002
    Messages:
    2,258
    Likes Received:
    51
    Location:
    Location, Location with Kirstie Allsopp
    That's right, and there was talk of a "hardware fog bug" in NV30 prior to the NV31/NV34 launch. No idea whether it was fixed for the mainstream parts, but I can't remember any reviewers pointing out similar problems. Obviously there's been a workaround for NV30.

    Maybe you should. It seems like a pretty "B3D article-worthy" subject. :D
     
  9. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,079
    Likes Received:
    648
    Location:
    O Canada!
     
  10. Pete

    Pete Moderate Nuisance
    Moderator Veteran

    Joined:
    Feb 7, 2002
    Messages:
    4,945
    Likes Received:
    350
    Yeah, the other "anomaly" was the 5900XT outperforming the 6600GT in 3DM01SE's 8-light geometry test. IIRC, the 5900XT no longer had separate fixed function T&L units, so its winning is curious, unless that test was influenced by its greater bandwidth or A64 platform. Otherwise, the 6600GT appears to be even or superior to the 5900XT in vertex performance, and that would seem logical, given they have the same number of VS units, but the NV4x's have been reworked. Could vertex performance really be holding the 6600GT back in your shader-heavy SC test? Andy's numbers seem to hint at such a limitation, given how close his 5700U's scores are to your 5900XT's (despite their wildly differing fillrates and platforms). Or is it just that the A64 allows for a much higher framerate ceiling than the P4 in your tests, or simply driver "inefficiencies"?

    It's a nice way to scale. Imagine how much simpler PC gaming would be to Joe Consumer if performance was the same across a product line, with only resolution scaling with price?

    BTW, any word on whether nV will release a four-pipe, two ROP 6200?

    I think this would be very well received, though I really doubt nV could have engineered a crossbar and four extra pipes into the NV30 at that time, even if they had thought of it. Maybe this can wait for a MB with both AGP and PEG slots, to reduce outside influences as much as possible? That way you can work in a MB review, as well.

    Can't wait to see how the X700P's eight pixels per clock stacks up to the 6600GT.
     
  11. KimB

    Legend

    Joined:
    May 28, 2002
    Messages:
    12,903
    Likes Received:
    218
    Location:
    Seattle, WA
    Well, the NV30 was only capable of about four FP operations per clock, while the NV43 is capable of sixteen, and is much more able to approach its peak FP performance. That's not a small difference for any DX9 game.

    Other than that, I've been saying it for a while: the NV3x architecture was clearly broken.
     
  12. Ichneumon

    Regular

    Joined:
    Feb 3, 2002
    Messages:
    414
    Likes Received:
    1
    Heh, you just weren't saying it when that was all Nvidia had.
     
  13. KimB

    Legend

    Joined:
    May 28, 2002
    Messages:
    12,903
    Likes Received:
    218
    Location:
    Seattle, WA
    No, I started saying it quite some time before the NV40 was released. But yeah, early on I just didn't realize what the problems with the NV3x were. I think I finally realized that the NV3x irreparably broken, oh, about a year ago now or so.

    Besides, the fact that I never purchased an NV3x should be a hint to you as to how much I thought about the cards....
     
  14. kemosabe

    Veteran

    Joined:
    Jun 19, 2003
    Messages:
    1,001
    Likes Received:
    16
    Location:
    Montreal, Canada
    Digit-Life wrote
    Food for thought? (Or rather, insert Dave's "No" here). :wink: :lol:
     
  15. KimB

    Legend

    Joined:
    May 28, 2002
    Messages:
    12,903
    Likes Received:
    218
    Location:
    Seattle, WA
    By the way, I just have to slip in a little something:
    http://www.beyond3d.com/forum/viewtopic.php?t=13231&postdays=0&postorder=asc&start=93

    Granted, I was talking about low-end parts, and the 6600 isn't quite low-end, but hell, it feels good to be essentially correct :)
     
  16. PatrickL

    Veteran

    Joined:
    Mar 3, 2003
    Messages:
    1,315
    Likes Received:
    13
    Well you could have said that ATI had a goodidea about Nvidia future products too :) With predictions like that there is little risk to be wrong .
     
  17. KimB

    Legend

    Joined:
    May 28, 2002
    Messages:
    12,903
    Likes Received:
    218
    Location:
    Seattle, WA
    Well, perhaps I quoted the wrong thing. But regardless, what I was saying at the time was that even if nVidia and ATI are close in the high-end market, I claimed that the NV4x architecture really had tremendous promise in the low-end. I think we have seen the first good evidence of this. For $200, the 6600 GT really is an incredible product. I can't wait to see what nVidia pulls off for the really low-end NV4x....I'm expecting good things.
     
  18. PatrickL

    Veteran

    Joined:
    Mar 3, 2003
    Messages:
    1,315
    Likes Received:
    13
    First 6600 GT is 249 € here, sol they will prolly be a bit less soon. Depending ATI card AND the avaibility of both cards, competition should be fun :)
     
  19. aZZa

    Newcomer

    Joined:
    May 31, 2003
    Messages:
    59
    Likes Received:
    0
    Location:
    the land of OZ
    How much improved is the memory controller in the 6600 over the 5800??? It seems to compare very well to the r9800pro/xt series, whilst the other could barely keep up with the r9700 series.

    How much bandwidth is actually being wasted/utilised in the 256 bit mem controllers on the high-end chips??? How much more efficient are the 128bit ddr memory controllers over the 256bit ddr controllers????

    The 6000GT now makes me think that PowerVR (series 5/6) might be able to get away with a similarly bandwidth graphics core, with significantly increased performance. Without TBDR or similar tech, it seems unlikely that many of the other smaller/lesser companies could reach such similar performance on such bandwidth. The 6600 chip seems to use memory bandwidth more optimally than previous generations.
     
  20. KimB

    Legend

    Joined:
    May 28, 2002
    Messages:
    12,903
    Likes Received:
    218
    Location:
    Seattle, WA
    I doubt it has much to do with the memory controller.
     
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...