AMD: R8xx Speculation

Discussion in 'Architecture and Products' started by Shtal, Jul 19, 2008.

?

How soon will Nvidia respond with GT300 to upcoming ATI-RV870 lineup GPUs

Poll closed Oct 14, 2009.
  1. Within 1 or 2 weeks

    1 vote(s)
    0.6%
  2. Within a month

    5 vote(s)
    3.2%
  3. Within couple months

    28 vote(s)
    18.1%
  4. Very late this year

    52 vote(s)
    33.5%
  5. Not until next year

    69 vote(s)
    44.5%
  1. flopper

    Newcomer

    Joined:
    Nov 10, 2006
    Messages:
    150
    Likes Received:
    6
    I use watercooling and will have to get a block for the upcoming 5870.
    I havent found the card to be too noisy, my 4870, can run gaming fine at 40% fan speed.
     
  2. rpg.314

    Veteran

    Joined:
    Jul 21, 2008
    Messages:
    4,298
    Likes Received:
    0
    Location:
    /
    SFR=split frame rendering. My chinese is obviously worst amongst the inhabitants of b3d, so I'll have to ask people with better mandarin than mine to please explain what went on there?

    If indeed 5870x2 does sfr, and is a form of a shared memory xfire, instead of the usual distributed-memory-like xfire, this is really exciting.
     
  3. seahawk

    Regular

    Joined:
    May 18, 2004
    Messages:
    511
    Likes Received:
    141
    He means that RV790 has more Flops then GT200, but does it outperfomr the GT200 by the same margin? And that is GT200, which tells only little about GT300.
     
  4. LunchBox

    Regular

    Joined:
    Mar 13, 2002
    Messages:
    901
    Likes Received:
    8
    Location:
    California
    I think what Ailuros meant was that 4870 X2 has more FLOPS than the 295 GTX if you look at the paper specs yet it delivers more frames in the gaming scene. Although this might also depend if the game was tailored to run more efficiently on NVidia hardware. If ATI plays their card right, and most DX11 games will base their engine on their Hardware, it might tip the balance to their favor.
     
  5. Ailuros

    Ailuros Epsilon plus three
    Legend Subscriber

    Joined:
    Feb 7, 2002
    Messages:
    9,442
    Likes Received:
    181
    Location:
    Chania
    How hard is it to answer the simple question how many theoretical maximum FLOPs a GTX295 has compared to a 4870X2 and in extension why the latter isn't faster than the first across the board.

    Your A= X FLOPs, B= Y FLOPs thus C must have Z FLOPs equasion is utter nonsense whether you like it or not.
     
  6. Xentropy

    Newcomer

    Joined:
    Apr 14, 2006
    Messages:
    28
    Likes Received:
    3

    I have heard that SFR is Scissor Frame Rendering actually for the upcoming 58xx cards. Not entirely sure what that is, but supposedly a new mode of operation.

    EDIT: Ahh, I guess Scissor Frame Rendering is what AMD calls it, and Split Frame is what Nvidia calls it.
     
    #2446 Xentropy, Sep 10, 2009
    Last edited by a moderator: Sep 10, 2009
  7. rpg.314

    Veteran

    Joined:
    Jul 21, 2008
    Messages:
    4,298
    Likes Received:
    0
    Location:
    /
    Good point. Then it must be because that 4870x2 loses out on other parts of the system, ie rops, tmu's, rasterizer speed, overall efficiency, driver quality etc.

    Agreed that this line of thinking may not be correct, but, surely, you agree that if gt300 loses to 5850 in peak flops, it is *highly* unlikely that it will be able to beat 5870x2, don't you? Just like it happened with 48x0 and gtx280.
     
  8. DegustatoR

    Veteran

    Joined:
    Mar 12, 2002
    Messages:
    1,489
    Likes Received:
    233
    Location:
    msk.ru/spb.ru
    Ehm, yeah, no.
    Assuming that little changes between DX10 and DX11 generation in comparable ALU utilisation (and that may already be too much of an assumption) NV will need ~3.35TFlops to be on par with 4.5TFlops AMD's card.

    New as in "3dfx VSA-100-level new"?
     
  9. rpg.314

    Veteran

    Joined:
    Jul 21, 2008
    Messages:
    4,298
    Likes Received:
    0
    Location:
    /
    Quick googling says scissor frame rendering=split frame rendering.
     
  10. Alexko

    Veteran Subscriber

    Joined:
    Aug 31, 2009
    Messages:
    4,515
    Likes Received:
    934
    Then look at it this way:

    4850 to 5870:

    1.0 to 2.72 TFLOPS, i.e. +172%
    63.6 to 153.6 GT/s, i.e. +141%

    Keeping in mind that the 4850 was indeed somewhat starving, it doesn't look very good. I hope I'm wrong, though.

    Edit: I'm basing this on apoppin's specs. If you believe Fudo, then the 5870 is a bit more balanced.
     
  11. Ailuros

    Ailuros Epsilon plus three
    Legend Subscriber

    Joined:
    Feb 7, 2002
    Messages:
    9,442
    Likes Received:
    181
    Location:
    Chania
    Well if you go then into that kind of speculative math, wouldn't you say that fillrates and what not should be part of the equasion?

    I wouldn't expect in my wildest dreams a D12U to be able to beat a Hemlock. But then again not only because it might end up in excess of 5 TFLOPs but rather that the equasion this time has 2*80 TMUs @ =/>825MHz, 2*32 ROPs @ =/>825MHz and a long line of et ceteras.

    If NV releases even under 40nm a dual chip D12U SKU as some rumors want them then yes I can easily believe that it could beat a Hemlock. But then again I could imagine such a beast to have way over 200 Gigatexels in raw fillrate.
     
  12. w0mbat

    Newcomer

    Joined:
    Nov 18, 2006
    Messages:
    234
    Likes Received:
    5
    I think 1.3GHz GDDR5 is over 160GB/s.
     
  13. Ailuros

    Ailuros Epsilon plus three
    Legend Subscriber

    Joined:
    Feb 7, 2002
    Messages:
    9,442
    Likes Received:
    181
    Location:
    Chania
    Back to the really interesting stuff; since that's an ever repeating rumor since day one the first details about Evergreen appeared (in one way or another) it sounds highly likely that there's something to it after all.

    Burning question would be how do they scale geometry with at least the same efficiency as AFR.

    If my layman's math ain't wrong it should be 166.40 GB/s
     
  14. Kaotik

    Kaotik Drunk Member
    Legend

    Joined:
    Apr 16, 2003
    Messages:
    9,070
    Likes Received:
    2,941
    Location:
    Finland
    The problem is that no-one knows the level / section bench was done on, on some levels GTX295 paired with ì7 965 Extreme stays still under 30 FPS, on some, obviously as your post shows, it's well over 30FPS with mere Core2 Duo

    edit:
    Browsing the Chiphell thread downwards, it seems in their tests GTX295 gets 25FPS AVG which I assume would be same level/section bench, but it seems just far too slow to be true
     
    #2454 Kaotik, Sep 10, 2009
    Last edited by a moderator: Sep 10, 2009
  15. Kaotik

    Kaotik Drunk Member
    Legend

    Joined:
    Apr 16, 2003
    Messages:
    9,070
    Likes Received:
    2,941
    Location:
    Finland
    ATI's old Scrissor-mode was Single/split frame rendering indeed, but not spit in half from the middle, but in several tiles.
     
  16. no-X

    Veteran

    Joined:
    May 28, 2005
    Messages:
    2,336
    Likes Received:
    297
    You are probably confusing super-tiling and scissors. Scissors mode splits the frame horizontaly or verticaly in two parts, which can be different in size. Super-tiling splits the frame in chess-board style.
     
  17. keritto

    Newcomer

    Joined:
    Apr 3, 2009
    Messages:
    143
    Likes Received:
    0
    330mm2 and 399 bucks. It's no surprise that ATi need to put wasted chips into their HD5850 offerings :oops: but on the other hand we should be happy that now we'd had 1440 SPs (-10%) and not 1280SPs into HD5850 (Pro) as the rumors said. And ATi leave some spare room to offer GT parts on 1280 SPs. Pretty good yield they have when they need only to disable 10% of chip for Pro parts and not even 25% or more like in times we last saw this disabling quads policy in R500 series.

    I'm a little dissapointed they went back to huge chip design and eat they tounge out after they swear on no huge chip designs policy in future and competing for performance crown. 330mm2 and little below 200W TDP is now became small, sick And as they really needed all that 80TMUs, what we now have to expect from nV?! ...120TMUs at least as their outer chip parts works on moderate speeds since G70, especially beefy top end chips.
     
  18. keritto

    Newcomer

    Joined:
    Apr 3, 2009
    Messages:
    143
    Likes Received:
    0
    Since when they went back on designs for putting memory chips on both sides just for measly 128b. Another disappointment?

    Those Cypress huge card for me should be announcement of 384b bus. Now at least nobody could say that they have too little die size to put such controller on it. :lol: 330mm2 damn
     
  19. leoneazzurro

    Regular

    Joined:
    Nov 3, 2005
    Messages:
    518
    Likes Received:
    25
    Location:
    Rome, Italy
    If you look at the GT300 rumors, that die is small compared to the rumored 500+ mm^2 of the next Nvidia flagship and Cypress die is still much smaller than the current GT200b. Consider also that yields go down more than linearly when die size increases. I think ATI is not saying "we'll made every generation smaller and smaller" but that they aim to performance/die size efficiency and they want to undercut the competition on price.
     
  20. Kaotik

    Kaotik Drunk Member
    Legend

    Joined:
    Apr 16, 2003
    Messages:
    9,070
    Likes Received:
    2,941
    Location:
    Finland
    The Cypress models are 1GB and 2GB, it's 256bit bus (or in theory 512bit, but that's not going to happen)
     
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...