ATI Talks R5xx

Discussion in 'Beyond3D News' started by Dave Baumann, Nov 1, 2004.

  1. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,079
    Likes Received:
    648
    Location:
    O Canada!
    Following the release of the Mercury Market share figures, ATI held an Analyst conference last week in which numerous point of ATI's business were discussed. Reports indicate that following the meeting Goldman Sachs have come back with numerous impressions of ATI's R5xx series, some of which are inline with elements that we've discussed before.

    The reports state that ATI have confirmed they are due to launch their new architecture, the basis of the R5xx series, in the first half of 2005. Goldman also believes the architecture to be based on Shader 3.0 and be very focused on the memory interface and bus. The expectation is that GDDR4 memory technology will ramp in production in late 2005 and the memory interface will be compatible with it and be designed to scale to speeds of 1.2GHz (2.4GHz effective) over its lifetime.

    Goldman also state that checks indicate that ATI have already taped out the products on TSMC's 90nm node. This goes against ATI's previous public statements that they were loath to transition to a new architecture at the same time as moving to a new process, especially as there appear to be no other products yet produced at 90nm from ATI - reports suggest that the high end product refreshes for R480 and R430 will utilise 130nm low-k and 110nm respectively. However, this is inline with ATI's CEO, Dave Orton's comments that they weren't sure how producible their Shader 3.0 architecture would be on 130nm and that they were looking to 90nm for it.

    Interestingly Goldman also suggests that the first part may have a "relatively smaller die size" - presumably this is in comparison to R420. Moving from 130nm to 90nm can reduce die size by about 50% for a similar architecture, however ATI have previously suggested that the move from FP24 to FP32, as dictated for by Shader Model 3.0 for full precision in the Fragment Pipeline, would result in a 25% increase in transistors alone for the fragment shader ALU’s. Seeing as the fragment shaders are part of the largest element of the die, and the extra features for SM3.0, such as vertex texturing, will require many more transistors, it will be interesting to see what the performance composition will be like when it is ultimately announced. Should 90nm yields be favourable, though, a smaller die size could result in greater availability - an issue that has evidently plagued all high end parts based on 130nm in the current cycle.
     
  2. digitalwanderer

    digitalwanderer Dangerously Mirthful
    Legend

    Joined:
    Feb 19, 2002
    Messages:
    17,267
    Likes Received:
    1,782
    Location:
    Winfield, IN USA
    Taped out already and due first half of next year? Bang-on mate! :D
     
  3. Druga Runda

    Druga Runda Sleepy Substitute
    Regular

    Joined:
    Sep 30, 2003
    Messages:
    647
    Likes Received:
    43
    so hopefully it will be available next September :lol:
     
  4. Pete

    Pete Moderate Nuisance
    Moderator Veteran

    Joined:
    Feb 7, 2002
    Messages:
    4,944
    Likes Received:
    347
    Et tu, Dave?

    Now you're suggesting launch first half and availability end of 2005, too? Or can they use plain old GDDR3 or even DDR1 with it until GDDR4 ramps up?

    Wait a sec--they're not using the exact same GPU in both Xbox 2 and PC, are they? Are they contractually obligated to wait until after Xbox 2 ships to launch PC-side? :?
     
  5. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,079
    Likes Received:
    648
    Location:
    O Canada!
    Given that R300 support DDR and DDR3, then R420 DDR, DDR2 and DDR3 I think it would be safe to assume that the high end R520's would start off with the highest DDR3 speeds that are sensible and once they start getting breathless, move across to DDR4 which will inevitably be faster.
     
  6. Richard

    Richard Mord's imaginary friend
    Veteran

    Joined:
    Jan 22, 2004
    Messages:
    3,508
    Likes Received:
    40
    Location:
    PT, EU
    Going directly to 90nm is pretty ballzy of them. I wonder what is their expected thermal density on those things. I've always looked at ATi for "small, fast, quiet and cool" so I'm very curious to see if they break any of these four qualities.
     
  7. JoshMST

    Regular

    Joined:
    Sep 2, 2002
    Messages:
    465
    Likes Received:
    18
    Very interesting! Hopefully TSMC's 90 nm process will be up to the task, which I think it will. It is funny to see both ATI and NVIDIA talking about extending the design cycle, but we really haven't seen either of them slow down now that things have gotten competitive. Seems every since the release of the R300, both of them have been going gangbusters (though admitedly, most of the R3x0 products haven't been significantly different from each other). While the R420 wasn't a full generational step, it did really expand the architecture to a pretty hefty degree (in my opinion). On the NVIDIA side, NV30/NV35/NV40 all appeared within the past 2 years, and I consider the NV30 and NV35 to be about the same type of step for NVIDIA as the R360 to R420 was.

    Of course, I could just be talking out my a$$. It happens you know.

    Still, good info here and I can't wait to see what ATI unveils in late Spring.
     
  8. John Reynolds

    John Reynolds Ecce homo
    Veteran

    Joined:
    Feb 7, 2002
    Messages:
    4,490
    Likes Received:
    262
    Location:
    Westeros
    nVidia launched their GF3s months before the Xbox reached the market. No, they weren't identical (the console GPUs had a second vertex unit) and it's possible ATI's contract for the next console is different, but I doubt it.
     
  9. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,079
    Likes Received:
    648
    Location:
    O Canada!
    If they have already taped out then they appear to have built themselves some contingency - its probably possible to hit 3 spins and still launch in H1 '05.

    I would suggest that the publicised issues that Intel have had with 90nm aren't necessarily a precident as they are using very different processes, different architectures and different design methodologies.
     
  10. JoshMST

    Regular

    Joined:
    Sep 2, 2002
    Messages:
    465
    Likes Received:
    18
    I really wonder what some of these initial wafer costs are going to be like for this type of design. On older processes these companies were paying upwards of $20,000 per wafer for test runs on new products, but that number gets much bigger the faster they want the process to go. For the basic 20K they can get a wafer back in about 8 weeks or so (not including packaging time). If they pay upwards of 150K per wafer, they can get the first runs back in a matter of 2 weeks or so. For 90 nm, I am betting that cost has gone up dramatically.

    The next problem is that the masks for the R5x0 will probably be the most expensive in the history of fabrication. I would not be surprised if the R5x0 from ATI is comprised of 350 million transistors (though Dave probably has some factual information about that one). Getting this out on 90 nm by late Spring is going to cost a pretty penny for ATI. If it has in fact already taped out though, they have given themselves ample room to make changes depending on the outcome of the first wafer runs.

    This of course makes me very curious about what the other side is doing as well. Will NV stick with IBM and go with their 90 nm process? What exactly do they have up their sleeve for late this winter and spring? Oh, the joys of speculation!
     
  11. Pete

    Pete Moderate Nuisance
    Moderator Veteran

    Joined:
    Feb 7, 2002
    Messages:
    4,944
    Likes Received:
    347
    Then perhaps the differences b/w GDDR4 and previous types isn't as great as I thought. Wouldn't doubling the bits transferred per clock require some low-level architectural tweaking? Maybe ATi's current "programmable" memory controller was a hint of things to come.

    Good point, John. MS is supposed to reveal Xb2 in January, anyway, so it's not like ATi would be letting too many cats out of the bag by launching in Q1.
     
  12. Mark

    Mark aka Ratchet
    Regular

    Joined:
    Apr 12, 2002
    Messages:
    604
    Likes Received:
    33
    Location:
    Newfoundland, Canada
    So it's a given that the R5xx will support SM3.0 and that it'll have, at least, some form of new memory controller. What about transistor count, does anyone want to wager a guess?

    (Am I the only one who likes the months leading up to a new release, with all the speculation and discussions and excitement, more than the new release itself?)
     
  13. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    Great, this just means more SoF2 clones....
     
  14. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    if they say r5xx being smaller than r420 does that apply to both r500 and r520?

    I would guess the r520 has around 230mil. transistors but not necessary more pipelines than r420
     
  15. phenix

    Regular

    Joined:
    Feb 22, 2003
    Messages:
    620
    Likes Received:
    1
    Location:
    Cambridge, MA
    What about TSMC's 90nm process? Any SOI, strained silicon, or low-K properties? do you think it would be better or worse than IBM's 90nm?
     
  16. AlphaWolf

    AlphaWolf Specious Misanthrope
    Legend

    Joined:
    May 28, 2003
    Messages:
    8,476
    Likes Received:
    325
    Location:
    Treading Water
    Is $20k or even $150k for a test wafer really a significant amount of money to ATI? Their revenues are approaching $2,000,000,000.
     
  17. gandalfthewhite

    Newcomer

    Joined:
    Feb 21, 2004
    Messages:
    43
    Likes Received:
    0
    agreed ratch its always more fun

    my wager is 280 million transistors by the way
     
  18. Khronus

    Newcomer

    Joined:
    Apr 15, 2004
    Messages:
    62
    Likes Received:
    2
    Lets get a R520 transistor betting pool going, maybe Dave could throw in a freebie item for the winner who comes closest to the transistor count without going over! :D
     
  19. 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
    it's low-K, and I assume they don't have any sort of SOI quite yet since we just had this.
     
  20. JoshMST

    Regular

    Joined:
    Sep 2, 2002
    Messages:
    465
    Likes Received:
    18
    AlphaWolf: Yes, it can be a large amount of money, especially when they don't just buy one wafer at a time. While the wafer costs are impressive, what is even more impressive are the photo masks used. Every time a revision is made, a new set of photo masks need to be produced, and these cost many millions of dollars.

    So while their revenues approach $2 billion, how many tapeouts and initial test wafers do they do in a year? If they could make 2 billion and change, and only have to do one tapeout a year, then those costs are minimal. When they have a whole lineup of products, and more than 20 of these types of tapeouts a year, it leads to a staggering amount of money. While not all of them will be the same price as a R5x0 type tapeout and initial production, they all add up nonetheless.

    Also, out of that $2 billion, how much of that is cost of goods sold? What are ATI's gross margins on those products sold? Once you start figuring out all of those different aspects, you will find that an initial batch of say 20 wafers that are needed very quickly, plus the costs of the photo masks could quickly lead to about $40 million for an initial tapeout and production for that product. This doesn't include any type of respin effort, or a major revision. When a company like ATI gets excited about a 40 to 60 million dollar profit for a quarter, you know that they had to watch every penny to get that amount. When the cost of a bad tapeout can quickly eat up that profit, you know the rest of the business is pretty nip and tuck.
     
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...