Radeon HD 2900 (R600) reviews thread

Discussion in '3D Hardware, Software & Output Devices' started by Geeforcer, May 10, 2007.

  1. Razor1

    Veteran

    Joined:
    Jul 24, 2004
    Messages:
    4,232
    Likes Received:
    749
    Location:
    NY, NY
  2. BRiT

    BRiT Verified (╯°□°)╯
    Moderator Legend Alpha

    Joined:
    Feb 7, 2002
    Messages:
    15,573
    Likes Received:
    14,169
    Location:
    Cleveland
    Even threads talking about 3dfx Rampage? :wink:
     
  3. Tim Murray

    Tim Murray the Windom Earle of mobile SOCs
    Veteran

    Joined:
    May 25, 2003
    Messages:
    3,278
    Likes Received:
    66
    Location:
    Mountain View, CA
    Yes, because the question of "how much of NV30 was based on Rampage" is inevitably raised.
     
  4. Geeforcer

    Geeforcer Harmlessly Evil
    Veteran

    Joined:
    Feb 6, 2002
    Messages:
    2,297
    Likes Received:
    465
    No, the question whenever a new card comes out is: "Is this finally the card that could have come close to Rampage, had it been released? For example, it is possible that 4 OCed 8800 Ultras could matched the power of the Rampage.
     
  5. Trawler

    Regular

    Joined:
    Mar 24, 2002
    Messages:
    250
    Likes Received:
    1
    Cmon, everyone knows it was nv40 that was based on Rampage! :wink:
     
  6. silent_guy

    Veteran Subscriber

    Joined:
    Mar 7, 2006
    Messages:
    3,754
    Likes Received:
    1,380
    It should also be noted that your statement is complete fiction.

    Fabless companies pay per wafer, not per chip.
     
  7. Star_Hunter

    Newcomer

    Joined:
    Jan 21, 2003
    Messages:
    77
    Likes Received:
    0
    Location:
    Michigan
    Well maybe thats not the case anymore but I know I read something that said that once I guess it was wrong.

    While I don't think this was where I got the idea from its the only one I could find in a few minutes

    "In addition, half of the inventory bloat was due to ATI underestimating the yields on its wafer by a factor of roughly 40% (i.e., the wafers yielded 40% more die than expected), further compounding its inventory glut (under ATI’s die buy model, ATI purchases on a per-die as opposed to per-wafer basis, implying that it has secured the price per die based on theoretical yields from the fab, and has committed to purchase the entire wafer no matter what the yield)"

    http://forum.beyond3d.com/showthread.php?p=553087#post553087

    I guess I miss understood that comment.
     
  8. AlexV

    AlexV Heteroscedasticitate
    Moderator Veteran

    Joined:
    Mar 15, 2005
    Messages:
    2,528
    Likes Received:
    107
    You did. TSMC and other foundries would be out of business otherwise.
     
  9. Geo

    Geo Mostly Harmless
    Legend

    Joined:
    Apr 22, 2002
    Messages:
    9,116
    Likes Received:
    213
    Location:
    Uffda-land
    I'm going to guess that TSMC employs an actuary or two that Lloyd's would be more than happy to have on staff, and will take either end of the die vs wafer proposition you'd like them to, and have the end results to TSMC's profits not differ materially.
     
  10. jb

    jb
    Veteran

    Joined:
    Feb 6, 2002
    Messages:
    1,636
    Likes Received:
    7

    Why wait till monday, tell us now? :)

    Ok so I guess only I thought that was funny huh?

    /me hangs up his short lived career as a comedian :D
     
  11. neliz

    neliz GIGABYTE Man
    Veteran

    Joined:
    Mar 30, 2005
    Messages:
    4,904
    Likes Received:
    23
    Location:
    In the know
    I have no problem telling you now though.. but it'll be based on BS. I first need to bring RB6:LV and my 2900 together.. my brother has my copy of rb6.. :(
     
  12. silent_guy

    Veteran Subscriber

    Joined:
    Mar 7, 2006
    Messages:
    3,754
    Likes Received:
    1,380
    It's hard to misunderstand the comment in any other way than 'ATI pays per working die'. Maybe it wasn't as much fiction as I though it was.:wink:

    So let me qualify: some companies, like LSI Logic and IBM, will happily produce your chips at a fixed cost per working die. But in that case, those companies have full control over the backend process: they are given a netlist and they are responsible for scan and BIST insertion, analog cells, memories, P&R, the works. In such a case, they have full control and understanding of the whole chip and that makes it much easier to avoid design related yield problems.

    Customers with a COT (Customer Owned-Tooling) flow, do everything themselves and dump a GDS2 file at the fab. It's basically a black box. This makes it harder and riskier to estimate correct yields (as apparently evident in the case of R520?). It also makes it a bit harder to set price models when you're dealing with redundant logic and lower-functionality parts.

    So I always worked in the fixed wafer cost model and thought that this was the mode of operations of TSMC/UMC/...

    As Geo wrote, a decent actuarian should find a way to make the numbers work for both parties and apparently that's what happened for R520.
    But since in that case the fab is taking on all the risk, the customer is going to pay a more stable but higher price for it.
     
  13. Jawed

    Legend

    Joined:
    Oct 2, 2004
    Messages:
    10,875
    Likes Received:
    767
    Location:
    London
    R520 may have been ATI's first PC GPU with fine-grained ALU redundancy.

    Admittedly, in R520, this would have been 1 extra ALU for each 4, so not amazingly fine-grained (as it would seem to be in R600, 1 for each 16).

    So, perhaps the redundancy worked better than expected? That seems pretty unlikely though, doesn't it?

    Presumably fabs and customers, jointly, are used to fine-grained redundancy when memory is a major part of a die. Again, relatively speaking, R520 was prolly a step-change in register file size over previous GPUs.

    So, perhaps the combination of a large increase in memory (with attendant redundancy) and fine-grained ALU redundancy caused the effect.

    Jawed
     
  14. Sxotty

    Legend Veteran

    Joined:
    Dec 11, 2002
    Messages:
    5,091
    Likes Received:
    453
    Location:
    PA USA
    That is simply not true. I received two vouchers one without those purchases and one with a 9600 of some type.
     
  15. jb

    jb
    Veteran

    Joined:
    Feb 6, 2002
    Messages:
    1,636
    Likes Received:
    7

    Well I was just making a play on words to try to be funny. :p

    And I would rather have your BS than others, so we will be waiting to see what you have for us on Monday! :)
     
  16. neliz

    neliz GIGABYTE Man
    Veteran

    Joined:
    Mar 30, 2005
    Messages:
    4,904
    Likes Received:
    23
    Location:
    In the know
    keep trying! ;)

    Hope to get it finished by then.
     
  17. Twinkie

    Regular

    Joined:
    Oct 22, 2006
    Messages:
    386
    Likes Received:
    5
  18. neliz

    neliz GIGABYTE Man
    Veteran

    Joined:
    Mar 30, 2005
    Messages:
    4,904
    Likes Received:
    23
    Location:
    In the know
  19. Twinkie

    Regular

    Joined:
    Oct 22, 2006
    Messages:
    386
    Likes Received:
    5
    Could it be because of the bandwidth advantage that the R600 has? Id say its a possibility since the res is at 1920x1200 and 2560x1600. Interesting results especially the DX10 results.
     
  20. neliz

    neliz GIGABYTE Man
    Veteran

    Joined:
    Mar 30, 2005
    Messages:
    4,904
    Likes Received:
    23
    Location:
    In the know
    The thing is that Far Cry so far has shown the XT to sometimes be outperformed by a 1950Pro, this new benchmark completely flips the tables.

    The DX10 tests were interesting too, but I'm even more amazed about the performance of the Vista drivers in some situations, cases where it's 15% faster should be a testimony to what can be achieved with vista driver optimization.
     
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...