Page 204 of 247 FirstFirst ... 104154194202203204205206214 ... LastLast
Results 2,031 to 2,040 of 2461

Thread: 7XXX speculation thread.

  1. #2031
    Quote Originally Posted by 265586888 View Post
    Full-speed DPFP operations isn't providing anything good for the consumer/gaming market at this point. Professional markets need that, but it isn't at the top of the list. Continuous efficiency improvements will yield better results, then full-speed DPFP operations will be the icing on the cake.
    Yes, gamers, nothing, vast majority of home use, nothing but there are people who work solo and would consider porting code to OpenCL if it offered a significant price/performance increase, and the more adoption of OpenCL no matter where, will raise its profile, there ARE groups of people who work on projects at home and who can't afford professional cards.

    AMD's reputation/profile in terms of GPGPU/professional type situations is, non existent isn't exactly fair, and isn't exactly unfair.

    Nvidia have a VERY established base in the professional market and can make more profit on one sale there than on a crapload of desktop card sales. AMD needs to make inroads, both by being considered serious contenders and having lots of people talking up how good the hardware is. One way to do that is to remove what is simply an artificial driver based limit on desktop hardware.

    It's not a massive thing, its not going to effect everyone, but it's one thing they can do, one niche which could translate into more in the future.

  2. #2032
    I thought about that a while ago but I'm afraid it will be as successful as "removing the 4P tax".

    I agree it makes sense, but sadly these markets just don't care about how cheap they can get the performance for, they only care about the performance. Dislodging incumbents in these markets is nigh-on impossible.

  3. #2033
    DH claims the 9th of January, and "Crossfire efficiency is also improved", which is pretty interesting considering how great the scaling is already.

    http://translate.google.com/translat...kesinlesti.htm

  4. #2034
    Quote Originally Posted by James View Post
    I thought about that a while ago but I'm afraid it will be as successful as "removing the 4P tax".

    I agree it makes sense, but sadly these markets just don't care about how cheap they can get the performance for, they only care about the performance. Dislodging incumbents in these markets is nigh-on impossible.
    Yup, but as I said, there are a lot of people who would try AMD purely because of the performance/cost attraction for either smaller companies or amateurs doing solo or small projects, where open source type software might also appeal.

    If you can generate a little buzz with several forums talking about how Tahiti DP performance is so good it was well worth switching from Cuda..... that is better than not having anyone saying that.

    Like I said AMD faces an uphill battle in the professional market, something they really desperately need to start making money in, because of their reputation.

    Its small and it won't convinced some massive company switching overnight, but you take every gain you can get everywhere, and there really isn't a downside to not artificially crippling hardware. Most bigger companies and proper projects will pay for the support of the professional cards, but basically the more amateur's starting and porting projects to OpenCL and AMD hardware, the better.

  5. #2035
    >intel 4004
    Join Date
    Jun 2009
    Posts
    11,281
    Full speed DPFP capabilities requires extra efforts to the hardware design.
    And it will end up to the question of "To the company, is full speed DPFP worth the cost or not".

  6. #2036

  7. #2037
    I was saying on another forum that, CES is 10th Jan, so launching the day before so that the biggest release and newest release is the 7970 for all of CES with everyone able to show their versions of cards(or at least, their different stickers ) all through CES. If AMD is launching in Jan which was essentially certain, if it was any time other than CES you'd be surprised.

    As for DP functionality, I could be wrong but I was under the impression the "limits" on DP in top end cards are software/bios limits, not hardware. IE a quadro whatever is identical to a 580gtx, just the drivers tell the 580gtx to run at a lower DP rate.

    Hmm, I might be wrong about AMD anyway, according to the AMD site the 5870 has 2.72tflops SP/544Gflops DP and the V9800 Firepro has 2.64/528 so only clock speed differences and no artificial limit unless I'm missing something.

    I'm fairly sure Nvidia do it, I thought AMD did it. oh well.

    Also just to clarify, I'm talking about full speed DF as in what any version of the card is able to do, as in no artificial driver based limit for various cards aimed at different markets. A 1:1 SP/DP ratio is something entirely different, almost certainly worthwhile for those that only want a card for its DP rate, but for AMD a very small market to tape out an entire new line of cards for, for Nvidia, bit surprised they haven't done it already.

  8. #2038
    Senior Member
    Join Date
    Jan 2010
    Posts
    3,479
    Quote Originally Posted by Ailuros View Post
    You can add almost another one to that or just count a wee bit less than one for dp.
    I expect that if the HD7970 can do 4TFLOPS SP, then the FirePro of the same die will get 2TFLOPS DP (1/2 rate), but the HD7970 will only do 1TFLOPS DP (1/4 rate).

    HD7950 (and future lower-end GCN products) might keep the 1/4 rate for DP, but AMD could differentiate and drop it down to 1/8th rate, especially if that helps TDP.

  9. #2039
    Or as AMD said GCN offers options between 1:2 and 1:16, they simply opted for 1:4 in hardware (and not enforcing it by an artificial limit) for the first high end GCN incarnation (entry level will probably be even lower) and only the high end version of a future GPU will offer a 1:2 rate.

  10. #2040
    Senior Member
    Join Date
    Jan 2010
    Posts
    3,479
    Quote Originally Posted by Gipsel View Post
    Or as AMD said GCN offers options between 1:2 and 1:16, they simply opted for 1:4 in hardware (and not enforcing it by an artificial limit) for the first high end GCN incarnation (entry level will probably be even lower) and only the high end version of a future GPU will offer a 1:2 rate.
    Depends on how aggressive AMD want to be. Intel's HPC chip is going to get 1TFLOP, so being able to beat it by a factor of two before it is even released would be a major achievement to crow about.

    Regardless, sometimes you have to stop and think "2048 32-bit FPUs on a single bit of silicon!" My first FPU was a 25MHz 68882...

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
WordPress Appliance - Powered by TurnKey Linux