Nvidia G71 - rumours, questions and whatnot

Discussion in 'Pre-release GPU Speculation' started by ToxicTaZ, Dec 4, 2005.

Thread Status:
Not open for further replies.
  1. ToxicTaZ

    Newcomer

    Joined:
    Mar 6, 2003
    Messages:
    52
    Likes Received:
    2
    Location:
    Edmonton
    90nm G71 vs 90nm RSX Should be the same just -200MHz


    Well it looks to me the 90nm RSX is a cut down 90nm G71? it mite look like this...

    110nm G70 (7800 Ultra) 24 Pipe @550MHz/900MHz 1.1ns (256-bit) K4J52324QC-BJ11

    90nm G71 (7900 Ultra) 32 Pipe @750MHz/1000MHz 1.0ns (256-bit) K4J52324QC-BJ10

    90nm RSX 32 Pipe @550MHz/700MHz 1.429ns (128-bit) K4J52324QC-BC14


    Most people in this thread say it looks more like this?

    110nm G70 (7800GTX 512) 24 Pipe @550MHz/900MHz 1.1ns (256-bit) K4J52324QC-BJ11

    90nm G71 (7900 ?) 24/32 Pipe @650MHz/900MHz 1.1ns (256-bit) K4J52324QC-BJ11

    90nm RSX 24 Pipe @550MHz/700MHz 1.429ns (128-bit) K4J52324QC-BC14

    well take your bet?
     
  2. walterman

    Newcomer

    Joined:
    Dec 27, 2005
    Messages:
    8
    Likes Received:
    0
    Thank you very much for the replies with this nice info. Now i understand the picture better.
     
  3. Ailuros

    Ailuros Epsilon plus three
    Legend Subscriber

    Joined:
    Feb 7, 2002
    Messages:
    9,429
    Likes Received:
    181
    Location:
    Chania
    If you'd concentrate exclusively on MADDs I'd rather say:

    R580:
    48 ALUs * 4 MADDs = 192 MADDs

    G7x@90nm:
    32 ALUs * 8 MADDs = 256 MADDs

    Now you need to mulitply with the according clockspeed to reach purely theoretical GFLOPs, but since I expect R580 to be clocked higher the difference might be smaller in the end than you've pictured it in terms of pure arithmetical throughput.

    ***edit: to make that a bit more clear....

    192 MADDs/clk * 0.65GHz = 124.8 GMADDs/clk = 249.6 GFLOPs

    256 MADDs/clk * 0.55GHz = 140.8 GMADDs/clk = 281.6 GFLOPs

    and yes minus texture OPs always in the latter case.
     
    #383 Ailuros, Dec 28, 2005
    Last edited by a moderator: Dec 28, 2005
  4. KimB

    Legend

    Joined:
    May 28, 2002
    Messages:
    12,928
    Likes Received:
    230
    Location:
    Seattle, WA
    Since the R580 is on the same process as the R520 but the G71 is on a new process, I don't really see how you'd come to that conclusion.
     
  5. neliz

    neliz GIGABYTE Man
    Veteran

    Joined:
    Mar 30, 2005
    Messages:
    4,904
    Likes Received:
    23
    Location:
    In the know
    Moving to a smaller process doesn't guarantee massive speed increases from the get go, right? I think we'll see a lot of the zoning throttle back in nv's design.. ATi just had half a year longer to work on their 90nm design.. that's all..
     
  6. Ailuros

    Ailuros Epsilon plus three
    Legend Subscriber

    Joined:
    Feb 7, 2002
    Messages:
    9,429
    Likes Received:
    181
    Location:
    Chania
    Unless you're expecting an 8 quad GPU with =/>380M to be clocked a lot higher than 600MHz, it would be a totally different story indeed.

    IMO it's either significantly high clockspeeds with 6 quads or something like 550-600MHz and 8 quads. Both at the same time sound way too optimistic for my taste, new process back and forth.

    ***edit: clockspeed or ALU throughputs don't worry me that much to be honest.

    http://www.xbitlabs.com/articles/video/display/games-2005_12.html

    I can see here a 24% loss from noAA to 4xAA on the R520 in 1600 and 45% for the 512 GTX. There's the real spot where NVIDIA must or should have concentrated on IMHLO.
     
    #386 Ailuros, Dec 28, 2005
    Last edited by a moderator: Dec 28, 2005
  7. KimB

    Legend

    Joined:
    May 28, 2002
    Messages:
    12,928
    Likes Received:
    230
    Location:
    Seattle, WA
    Yeah, that's about what I'm expecting too. But I'm also not expecting the R580 to clock any higher than the R520 (or, at least, not significantly so).
     
  8. Ailuros

    Ailuros Epsilon plus three
    Legend Subscriber

    Joined:
    Feb 7, 2002
    Messages:
    9,429
    Likes Received:
    181
    Location:
    Chania
    But 650MHz hypothetical gives ~250GFLOPs vs. 280 for G7x@550MHz. Now take the memory controller into the equation and fore mentioned percentage differences, and I doubt that even 600MHz will save the latter's day, unless NV has moved to other changes we're not aware of yet.
     
    #388 Ailuros, Dec 28, 2005
    Last edited by a moderator: Dec 28, 2005
  9. KimB

    Legend

    Joined:
    May 28, 2002
    Messages:
    12,928
    Likes Received:
    230
    Location:
    Seattle, WA
    I don't know. A G71 with 32 pipelines at 550MHz should be a good 33% faster than the current GeForce 7800 GTX 512, which places the product somewhere in the range of 60% or so faster than the Radeon X1800 XT (obviously rather dependent upon which benchmarks you look at, but I don't think putting the 512 at 20% faster than the XT is out of line).

    This should place the G71 squarely in the range of the (average) performance improvement we should expect from the R580 over the R520. Yes, there will likely be a game or two that really allows the R580 to shine, and it may prove to be more future-proof, but I still expect the G71 to remain highly competitive for today's games.
     
  10. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,081
    Likes Received:
    651
    Location:
    O Canada!
    Chalnoth, using R520 as a guide for clockspeeds is not necessarily the best idea since it was an abnormal product run and they had a fairly specific goal in mind when they did eventually release it. R580 is likely to have a greater time in the market, hence the SKU's will be planned differently.
     
  11. Ailuros

    Ailuros Epsilon plus three
    Legend Subscriber

    Joined:
    Feb 7, 2002
    Messages:
    9,429
    Likes Received:
    181
    Location:
    Chania
    That's a mere speculation on the added fillrate and doesn't take into account available bandwidth.

    Of course am I looking at games like FEAR because exactly there the ALU to texture relation reaches unnatural heights. In the link I posted above with no AA in 1600 it's 65 fps for the 512 GTX and 51 fps for the X1800XT. Add AA/AF and the 512 GTX drops to 35 fps, while the X1800 only to 39 fps and pulls actually ahead.

    Where the smaller drop in the latter case comes from is obvious and that's where I'm trying to get to.

    If I'd be asked what R520 lacks mostly in I'd say ALU throughput compared to G70. Once though that particular gap closes, I can't simply ignore the relatively small performance drop ATI's memory controller gives in 4xMSAA case scenarios. I'm not saying that it'll end up like that, I'm merely puzzled right now because just adding more quads for NV doesn't sound good enough.
     
  12. Geo

    Geo Mostly Harmless
    Legend

    Joined:
    Apr 22, 2002
    Messages:
    9,116
    Likes Received:
    213
    Location:
    Uffda-land
    There seems to be a consensus growing that R520 is going bye-bye entirely in favor of R580, rather than becoming the "one down" model. Assuming they had made that decision at the beginning of the production run, we have to consider the possibility that clearing out the inventory was the priority rather than pushing the clocks. I suspect that the 700mhz models we're seeing now from some AIB, under pressure from GTX512, is more indicative of what they could have really done for a top-end model if they'd been serious about it being a longer-term part. The early scuttlebutt was a target range of 700-750mhz. So looking at all that, and recognizing the marketing difficulties of a follow-on product in the same family having *lower* clocks, I think at this point I'd be moderately disappointed if the tippy-top R580 model doesn't have core clocks starting with a "7". At least that XTX one.
     
  13. dizietsma

    Banned

    Joined:
    Mar 1, 2004
    Messages:
    1,172
    Likes Received:
    13
    When is G71 due to be released again and what is the likely naming convention ? 7900 ??
     
  14. neliz

    neliz GIGABYTE Man
    Veteran

    Joined:
    Mar 30, 2005
    Messages:
    4,904
    Likes Received:
    23
    Location:
    In the know
    Q1 or early Q2
     
  15. KimB

    Legend

    Joined:
    May 28, 2002
    Messages:
    12,928
    Likes Received:
    230
    Location:
    Seattle, WA
    I'm not really sure that I buy that, since the X1800 XL product was there to take care of the slower-clocked parts. If they really wanted to push out inventory, then, they should have increased the clockspeeds, considering how the product fares against the 7800 GTX (256MB).
     
  16. neliz

    neliz GIGABYTE Man
    Veteran

    Joined:
    Mar 30, 2005
    Messages:
    4,904
    Likes Received:
    23
    Location:
    In the know
    REally? the X1800XL was launched based on parts with the soft ground problem, the cores that were very unstable above 500mhz not specifically a lower grade, but different.
     
  17. Ailuros

    Ailuros Epsilon plus three
    Legend Subscriber

    Joined:
    Feb 7, 2002
    Messages:
    9,429
    Likes Received:
    181
    Location:
    Chania
    The X1800XL competitor was and is IMO the 7800GT.
     
  18. walterman

    Newcomer

    Joined:
    Dec 27, 2005
    Messages:
    8
    Likes Received:
    0
    But the X1800 has 16 ROPs at 625MHz vs 16 ROPs at 550 MHz -> 625/550 MHz +-= 55/49 fps.

    http://techreport.com/reviews/2005q2/geforce-7800gtx/index.x?pg=2
     
  19. Ailuros

    Ailuros Epsilon plus three
    Legend Subscriber

    Joined:
    Feb 7, 2002
    Messages:
    9,429
    Likes Received:
    181
    Location:
    Chania
  20. Arun

    Arun Unknown.
    Moderator Legend Veteran

    Joined:
    Aug 28, 2002
    Messages:
    5,023
    Likes Received:
    300
    Location:
    UK
    It will always amuse me to see how many people consider a scene (or *gasp* game) to be limited by the same factors at the same time. Here's a stupid question, for which I don't have precise figures for (but could get rough ones if needed): How many of a character's vertex budget is used on its head? And what's the average percentage of visible pixels from a character's head, compared to total average percentage of visible pixels from that character, in an average FPS scene?

    That should give you a good idea of the changing fillrate/VS balance in a *single* model here, which explains why upclocking either of the clock domains on a G70 would increase performance in a single scene, but in different percentages. If you want an example of balance load changes within a single scene, a great example would be particles, because they can seriously fuck up everything (alpha blended ones, especially so). Heck, they've often even got completely different ALU:TEX ratios than normal 3D models!

    So walterman, don't try reasoning like that. Just don't.


    Uttar
     
Loading...
Thread Status:
Not open for further replies.

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...