Slashdot Mirror

← Back to Stories (view on slashdot.org)

NVIDIA Announces New Quadro M6000 With 24GB Memory Buffer For Heavy Workloads (hothardware.com)

Posted by msmash on from the memory-hunger dept.

Reader MojoKid writes: Some might say there's no such thing as a one-size-fits-all solution when it comes to computing, and that's especially true for workstation graphics professionals who need varying levels of performance and memory space. For that reason, NVIDIA is now offering a version of its Quadro M6000 graphics card with 24GB of GDDR5 memory, twice as much memory as much as the original model. According to NVIDIA, customers rendering datasets larger than 12GB can experience up to 5X faster performance compared to the previous Quadro M6000. Like the 12GB version, the new 24GB Quadro M6000 is based on NVIDIA's Maxwell architecture. It has 3,072 CUDA cores, a 384-bit memory bus, four DisplayPort 1.2 connectors, a single DVI-I connectors, and a maximum power consumption rating of 250W. In addition to the doubling the memory buffer, NVIDIA added a few other features, including more GPU clock options, greater software temperature control to keep the GPU temp below the point where throttling occurs, and a new under-power boot message if the card is ever under powered.

22 of 43 comments (clear)

  1. Holy crap ... by gstoddart · · Score: 5, Insightful

    I remember being all excited to have a video card with 1MB of RAM which would do 1024x768.

    My desktop has 16GB of RAM.

    What kind of porn are you guys watching you need a 24GB video card?

    --
    Lost at C:>. Found at C.
    1. Re:Holy crap ... by simcop2387 · · Score: 5, Informative

      Volumetrically modeled porn. Also we're watching neural networks excite each other.

    2. Re:Holy crap ... by Iamthecheese · · Score: 1

      I think VR porn will need that. Self-aware VR porn will need much more. On a more practical note when this amount of video memory is common it will be possible to buffer games far enough into the distance and with good enough textures to reach retinal quality. Video will finally be literally indistinguishable from real life.

      --
      If video games influenced behavior the Pac Man generation would be eating pills and running away from their problems.
    3. Re:Holy crap ... by gstoddart · · Score: 2

      As much as I understand it's not literally all porn in the sense of boobies ... volumetrically modeled porn and neural network porn is no different from food porn and automobile porn.

      It's all porn of some kind for someone, even it allegedly not porn. ;-)

      --
      Lost at C:>. Found at C.
    4. Re:Holy crap ... by Hentes · · Score: 1

      Quadros are professional grade cards for content creators, not consumers.

    5. Re:Holy crap ... by blogagog · · Score: 1

      I'm less interested in the porn than I am the quality of the shadows behind the porn. For that, you need a really good graphics card.

    6. Re:Holy crap ... by JohnStock · · Score: 1

      Thats advanced. I remember having a ZX81 with 1KB of RAM with part of that being screen memory

  2. CUDA by orledrat · · Score: 1

    Quite the CUDA card, this one..

    Until there's HBM, with its ultra-wide memory busses, this thing seems a likely choice for mem-buffer intensive crunching.

    1. Re:CUDA by fuzzyfuzzyfungus · · Score: 2

      What surprises me(especially since AMD is clearly capable of it, since the PS4 does it, and it'd be a bit of a shock if Nvidia and Intel were incapable of following suit or already have something ready to go) is that CUDA(or OpenCL)-targeted systems still haven't made any moves(at least not commercially visible ones) toward unified memory architecture; and/or allowing the relatively expansive supply of motherboard memory expansion to use GDDR and connecting the GPU via a faster interface than PCIe if needed.

      As much as it would probably involve selling more Nvidia GPUs than they would like, AMD seems like the one that might actually be particularly interested in this, since it would at least get opterons into servers: They've been pushing HyperTransport as a multi-vendor high speed interconnect(first under 'Torrenza', now under the 'Heterogenous System Architecture Foundation', with some fancy FPGAs available that can interact with Opterons in multi-socket boards.

      If GPU compute cards are shipping with increasingly monstrous amounts of RAM, that seems like an opportunity for HSA to move the GPU closer to being a first-class citizen along with the CPU; and allowing RAM expansion.

    2. Re:CUDA by orledrat · · Score: 1

      What surprises me(especially since AMD is clearly capable of it, since the PS4 does it, and it'd be a bit of a shock if Nvidia and Intel were incapable of following suit or already have something ready to go) is that CUDA(or OpenCL)-targeted systems still haven't made any moves(at least not commercially visible ones) toward unified memory architecture;

      CUDA has had all kinds of unified memory support since version 6, so for at least two years now, I reckon. Have a look at cudaMallocManaged(). The stack memory is "unified" as well now, as is the system allocator (since the GM2xx series).

    3. Re:CUDA by SoftwareArtist · · Score: 3, Interesting

      NVIDIA is in fact working on something like that. It's called NVLink and will, according to that page, enable "data sharing at rates 5 to 12 times faster than the traditional PCIe Gen3 interconnect". There's no commercial hardware yet that supports it, but it should be coming soon.

      --
      "I'm too busy to research this and form an educated opinion, but I do have time to tell everyone my uninformed opinion."
    4. Re:CUDA by fuzzyfuzzyfungus · · Score: 1

      I realize that the GPU can access main memory(the cheap and lousy ones have been doing it to save money for ages; and Nvidia has made the process a bit nicer for their compute users when they need access to lots of RAM); but the memory is 'non-unified' in the sense that a GPU accessing system RAM is (relatively speaking, it's still pretty fast in absolute terms) a second class citizen compared to a CPU accessing system RAM(there is an unavoidable increase in latency if the RAM is connected to a different CPU than the one accessing it and you add a Hypertransport/QPI hop in there); while even if you have heroic amounts of money to wave around you just don't have the option of GDDR5-or-equivalent speeds for system RAM.

      There is no 'zero overhead' option when you are talking about multiple memory controllers, hence NUMA; but my naive expectation would have been that CUDA-oriented systems would have moved more aggressively in the direction of resembling multi-socket CPU arrangements; but with GPUs in most of the sockets and GDDR-speed RAM across the board, rather than these cards with their increasingly enormous onboard RAM supply, connected (comparatively) slowly to the CPU and system RAM. I'm sure that they have their reasons, probably better than mine; but it seems surprising from my naive position.

    5. Re:CUDA by K.+S.+Kyosuke · · Score: 1

      CUDA has had all kinds of unified memory support since version 6,

      Sounds very dubious to me given the fact that to my knowledge, AMD is the only company that (in its most recent APUs) actually put its GPU on top of the CPU's memory access infrastructure, including the virtual address space. The GPU then accesses any relevant data immediately without copying, and any relevant drivers or infrastructure can be put into user space since the GPU doesn't access physical memory but is subject to memory protection in exactly the same way that multiple threads belonging to the same process (and running on multiple cores simultaneously) are. Meaning that not only is there zero copying of data but also any commands to the GPU units can avoid any kind of context switches or syscalls, making GPU calls as fast as calling CPU subroutines. Do not accept nVidia's imitations!

      --
      Ezekiel 23:20
    6. Re:CUDA by K.+S.+Kyosuke · · Score: 1

      Except that there's nothing "unified" about nVidia hardware's memory. nVidia hardware has its own memory.

      --
      Ezekiel 23:20
  3. Inferences? by Cederic · · Score: 1

    As someone deferring a hardware upgrade until the shiny new Nvidia GPUs come out with a step change in performance/watt should I worry that they're still issuing new 'top end' versions of the current generation?

  4. Obligatory by Voyager529 · · Score: 1

    Look, I know it's been almost ten years now, but just checking - are we still waiting for a card that can run Crysis?

    1. Re:Obligatory by U2xhc2hkb3QgU3Vja3M · · Score: 1

      I thought Witcher 3 was the new Crysis?

    2. Re: Obligatory by backslashdot · · Score: 1

      Imagine a Beowulf cluster of these.

    3. Re: Obligatory by sconeu · · Score: 1

      It could certainly run warm enough to cook some hot grits.

      --
      General Relativity: Space-time tells matter where to go; Matter tells space-time what shape to be.
  5. Pity by MobileTatsu-NJG · · Score: 1

    Too bad the software that would take advantage of this has been shifting away from the benefits that the Quadro brand typically offers and towards more game-engine like approaches that make gaming cards more cost-effective.

    --

    "I like to lick butts!" by MobileTatsu-NJG (#32700246) (Score:5, Informative)

  6. MojoKid is HotHardware.com's editor in chief by dave420 · · Score: 1

    MojoKid is HotHardware.com's editor in chief: Is this acceptable? There is no acknowledgement of his involvement in the site in question. It smacks of some sort of nonsense.

  7. Re:What people are using it for? by Rakarra · · Score: 1

    Autodesk's Maya, for instance. You can load in larger sets and more complex objects, and it'll be faster to manipulate (or the same speed at more complexity) because program won't have to swap data between video ram and system ram as much.

    The Foundry's Mari for high-detail texture mapping and shader creation, and Katana for scene lighting.

    Visualization, particularly computer graphics and CG effects are a big market here.