Announcement

Collapse
No announcement yet.

NVIDIA Rolls Out The Volta-Based Quadro GV100

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • OneBitUser
    replied
    Originally posted by theriddick View Post
    Multi core GPU's would be expensive, and completely cut out the small form factor market. Also the heat involved with cooling such a card could put people off also. BUT if they reduced the power draw of the vega gpu significantly, and slapped two on a die, then they could still come under 300W and outperform a 1080ti, maybe.
    Multi-core GPUs could be the solution to AMDs problems with a model similar to the Ryzen CPUs.

    Compared to Vega 64 I believe it would be much cheaper to manufacture 2-4 Polaris chips and shared 8 GB stack of HBM2 and then connecting the GPUs with Infinity fabric.

    Due to the nature of the interposer and the HBM2 memory the package size (i.e. GPU+RAM chips) might even be smaller than on a regular GPU+GDDR5X card, meaning that you could have ITX-sized cards with enough space for a strong VRM.

    Leave a comment:


  • Vistaus
    replied
    Originally posted by theriddick View Post
    Multi core GPU's would be expensive, and completely cut out the small form factor market. Also the heat involved with cooling such a card could put people off also. BUT if they reduced the power draw of the vega gpu significantly, and slapped two on a die, then they could still come under 300W and outperform a 1080ti, maybe.
    Water cooling should do the trick ;-)

    Leave a comment:


  • theriddick
    replied
    Multi core GPU's would be expensive, and completely cut out the small form factor market. Also the heat involved with cooling such a card could put people off also. BUT if they reduced the power draw of the vega gpu significantly, and slapped two on a die, then they could still come under 300W and outperform a 1080ti, maybe.

    Leave a comment:


  • pcxmac
    replied
    AMD going with TSMC might actually help them catch up substantially. Not that I know what kind of IP TSMC can provide AMD's process, but it's probably a bit better than what GloFo has at the moment. AMD is like a crude truck with more torque but less ponies, AMD needs to use it's fabric and double their processing power every generation until NVIDIA is force to break out of cycle like Intel. That and with some bad press like Nvidia has been having should do well enough to draw even in 2-3 years. But AMD seriously needs to double their potential, Nvidia will drive up the prices, but if they can get cards in at 3-400 watts with 2+ dyes on board, with options for both HBM and GDDR, they will have a very powerful platform. Nvidia has only their software and and some silicon process keeping them ahead at the moment. Vulkan and their O/S efforts will hopefully pay off soon.
    Last edited by pcxmac; 03-28-2018, 04:28 AM.

    Leave a comment:


  • theriddick
    replied
    If AMD released a 7nm Vega GPU then I think they'd do ok. 7nm should allow them to drop power consumption allot and also enable the ability to reach 1080ti performance levels, that would be enough for quite a few people to come back to AMD.

    The issue I see is the price gouging and high cost of HBM2, AMD use to sell cheaper cards then competitor (by a little bit) but now AMD cards are significantly more pricey compared to NVIDIA, a BAD precedence when you only have %8 market share (gamers). The miners are not going to be loyal long term, both companies are going to find this out the hard way!

    Leave a comment:


  • coder
    replied
    Originally posted by ethana2 View Post
    nVidia made some *really* strong moves towards big iron and cloud neural network computing today, I hope AMD can keep up :/
    Meh, AMD being behind in HPC and Deep Learning is nothing new. Nvidia is accelerating like a rocket. Catching up to them seems nearly impossible.

    What AMD needs to do is to find a niche (or a few). They sort of did that with consoles, but the problem is it's a cost-sensitive market with fairly low-growth.

    I like their commitment to open source and standards, but standards are nearly always slower & more labor-intensive than proprietary. The open source strategy is also about playing a long game, and I feel like it hasn't yet started paying big dividends.

    Leave a comment:


  • nuetzel
    replied
    Sorry, German:

    Nvidias 3100-Euro-Grafikkarte Titan V verrechnet sich angeblich
    https://www.heise.de/newsticker/meld...h-4001320.html

    Leave a comment:


  • ethana2
    replied
    nVidia made some *really* strong moves towards big iron and cloud neural network computing today, I hope AMD can keep up :/

    Leave a comment:


  • coder
    replied
    Originally posted by pinguinpc View Post
    They've managed to keep Titan V in stock for at least 1.5 months, now.

    https://www.nvidia.com/en-us/titan/titan-v/

    While $3k would seem to be plenty of incentive to do so, consider the same GPUs are going into $10k Tesla products that are also in high demand. The 12 nm chips are so enormous that I don't imagine they can profitably be sold for much less.

    Oddly, Titan XP has gone in & out of stock in the US, but remained available in India. I wonder if that has anything to do with the fact that they still don't (can't?) sell Titan V there.

    http://www.nvidia.in/graphics-cards/...scal/titan-xp/

    Leave a comment:


  • coder
    replied
    These aren't as "new" as they claim. They were obviously in the DGX Station (or maybe a 16 GB variant).

    They're basically Titan V with fully-enabled (double-memory) GV100 and NVLink activated.

    Leave a comment:

Working...
X