Page 6 of 10 FirstFirst ... 45678 ... LastLast
Results 51 to 60 of 91

Thread: Nvidia?s five new Keplers raise a red flag

  1. #51
    Quote Originally Posted by Medallish View Post
    and it seems AMD is making a move on nVidia's cash cow with GCN.
    I don't think they have anything comparable to Tesla, I mean they don't have HPC dedicated cards. Also it seems they have only two pro cards using the last gen of their chips (V3900 and V4900) and they seem to be low profile compared to Cayman variants (note I might be completely wrong, as I only used Wikipedia ).
    Speaking for myself.

  2. #52
    8-bit overflow
    Join Date
    Jun 2010
    Posts
    378
    They do have FireStream boards. The most recent one though is based on the HD 5870.

  3. #53
    Hi everybody,
    Cayman was not a planned chip. The advantages of VLIW-4 over VLIW-5 might not have been big enough to justify the extra amount of driver work. With GCN, the HPC Driver department could start into a new design cycle. The product cycles seem to be a bit longer in the professional business, where PCs have a minimum lifetime of est. 3 years (taxes), and supercomputers have long project planning times. Not to mention the software validation effort. So with 28nm and GCN, we see a significant amount of improvements, and AND`s driver department should be aware of the chances. If they are not or their support for business customers falls asleep, it would be their own fault.

    Greetz,

    crs

  4. #54
    Senior Member
    Join Date
    Mar 2011
    Posts
    781
    Quote Originally Posted by laurent View Post
    I don't think they have anything comparable to Tesla, I mean they don't have HPC dedicated cards. Also it seems they have only two pro cards using the last gen of their chips (V3900 and V4900) and they seem to be low profile compared to Cayman variants (note I might be completely wrong, as I only used Wikipedia ).
    http://en.wikipedia.org/wiki/AMD_FireStream if anyone should be worried, imo it should be nVidia considering AMD has some great server tech and experience.

    But yeah their FirePro cards are still Cayman based, I think it'll be some months before we see any GCN based Firepro cards, it usually takes longer for new tech to reach the professional market, I'm guessing software needs to be pretty refined?
    Desktop R7 1800X@4.1GHz, MSI X370 Xpower gaming titanium(Bios 1.74), 2x8GB Corsair Vengance LED(Red) DDR4 3200MHz@2933MHz , 2xSamsung 830 256GB & Samsung 850 EVO 500GB SSD, Sapphire Fury X 1050MHz / 500MHz, Corsair AX1200i.
    Laptop: Lenovo Y700 AMD FX-8800p, R9 M385X, 2x8GB DDR3, 256GB SSD.
    Quote Originally Posted by vain View Post
    From my personal browsing experience, everyone is well endowed.

  5. #55
    Ha yes indeed there's FireStream. Did it get any design win in some supercomputer as nVidia did?

    @Medallish: indeed software has to be pretty refined, in particular because most GPU have *many* bugs. BTW does Tahiti have support for ECC? My understanding is that it is mandatory at least for HPC.

    And no matter how good AMD HW is, another big issue is that the HPC sites using GPGPU are now used to Cuda and I'm not sure they would want to spend time porting to OpenCL. nVidia still have the lead here and I'm willing to bet it will take many years before AMD makes significant inroads.
    Speaking for myself.

  6. #56
    >intel 4004
    Join Date
    Jun 2009
    Posts
    11,281
    Quote Originally Posted by Shadow Concept View Post
    Basically my point is that people tend to compare stock card to stock card - overclocking headroom and performance of factory overclocked cards is generally thrown to the wayside. What's the point of having all these overclocked cards if ignorant enthusiasts read a few reviews, see 680 wins the benchmarks and spend $50 extra to get one, when the factory overclocked 7970 is the same price/cheaper and still has more overclock potential?
    A new bin and a new name, reference designs are supposed to be safe, or there will be virtually a trillion words spread all over a million stickers on the card to remind the user that "use at your own risk, sucka".

    It would be pointless for an official 1 GHz card with less overclocking headroom and more marketing money go to waste. Wouldn't a badge for AIBs to use already adequate for marketing purpose? When people disregard overclocked cards, they will do the same even with a new name and/or something they don't see (e.g. a new bin). So I am doubtful about the mindshare change.

  7. #57
    >intel 4004
    Join Date
    Jun 2009
    Posts
    11,281
    Quote Originally Posted by laurent View Post
    BTW does Tahiti have support for ECC? My understanding is that it is mandatory at least for HPC.
    Yes, end of answer.

  8. #58
    >intel 4004
    Join Date
    Jun 2009
    Posts
    11,281
    GCN is supposed to be less driver-dependent, because it doesn't require the driver to pack instructions any more. So efficiency goes up there.

  9. #59
    Senior Member
    Join Date
    Mar 2011
    Posts
    781
    Quote Originally Posted by laurent View Post
    Ha yes indeed there's FireStream. Did it get any design win in some supercomputer as nVidia did?

    @Medallish: indeed software has to be pretty refined, in particular because most GPU have *many* bugs. BTW does Tahiti have support for ECC? My understanding is that it is mandatory at least for HPC.

    And no matter how good AMD HW is, another big issue is that the HPC sites using GPGPU are now used to Cuda and I'm not sure they would want to spend time porting to OpenCL. nVidia still have the lead here and I'm willing to bet it will take many years before AMD makes significant inroads.
    Looks like it did.

    Well I wonder if Cuda really is that widespread amongst HPC's, wouldn't it make more sense for them to use OpenCL, both for stability and compatability?

    *Edit* Found some info on that:
    http://www.hpcwire.com/hpcwire/2012-...d_on_cuda.html

    And let's not forget, AMD is in the HPC business and have been longer than nVidia, there's a reason why atleast 4 of the top 10 supercomputers uses Opteron processors.
    Desktop R7 1800X@4.1GHz, MSI X370 Xpower gaming titanium(Bios 1.74), 2x8GB Corsair Vengance LED(Red) DDR4 3200MHz@2933MHz , 2xSamsung 830 256GB & Samsung 850 EVO 500GB SSD, Sapphire Fury X 1050MHz / 500MHz, Corsair AX1200i.
    Laptop: Lenovo Y700 AMD FX-8800p, R9 M385X, 2x8GB DDR3, 256GB SSD.
    Quote Originally Posted by vain View Post
    From my personal browsing experience, everyone is well endowed.

  10. #60
    Quote Originally Posted by Medallish View Post
    The only real part of AMD's Radeons that have to stand up to Intel is the ones on AMD's APU's, of course the dedicated card market AMD very much only has to compete with nVidia, and it seems AMD is making a move on nVidia's cash cow with GCN.
    Nvidia's cash cows will be dead soon as Haswell with AVX2 arrives and IGPs start to give for free what Nvidia charges an arm and leg to give. I pretty much doubt that they will be able to fund development for something beyond Broadwell/Excavator unless Tegra takes off.

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