This Is The 'Fastest GPU Ever' Made

NVIDIA has just announced an insane new GPU. Designed for (wealthy) graphic professionals, the Kepler-based Quadro K6000 is, apparently, "the fastest and most capable GPU ever built".

With twice the capacity of the Quadro 6000 that it replaces, the specs on this thing are just crazy: it has 12GB of super-fast DDR5 graphics memory, 2880 streaming multiprocessor cores, ultra-low latency video I/O, and the ability to drive four simultaneous displays at up to 4K resolution. So, you probably need this on your home PC, right?

Just as an example of the power of this thing, Nissan can apparently load almost-complete vehicle models using it. And that's incredibly impressive. But impressive will come at a cost. While there are no firm pricing details available at present, the K5000 costs $US2250 — so brace yourself. It will be available later in the year. [NVIDIA]


    The big question that's on everyone's minds "can they be SLI'd?" XD

    I thought because the Quatro's arn't part of the Geforce line that their drivers don;t support games very well, they are for workstations not gamers.

      Pretty much my understanding too. Sure the core tech is the pretty much the same, but from then on they deviate.

      Yeah, for design/CAD, not gaming.

        I wish, CAD user here. CAD software is bottlenecked at the cpu since the 3d modelling kernel can only utilise one thread. I hate staring at my task manager during a long load seeing only one of eight cores being utilised. This would be more for running simulations and rendering.

        Last edited 25/07/13 1:14 pm

          Good for ray tracing. And yeah definitely rendering. Good for people in the video effects industry

          That's related to software you're using though, not hardware... Not that I use or know of any software, but there should be something out there that can utilise all available hardware to do renderings. Heck, if Universal Media Server (PS3 Media Server) had the option to use multi-core CPU and GPU, CAD software should as well!

            It's to do with how CAD works as opposed to a game, Since most parts of a model are built off each other or are positioned relative via constraints it's very math heavy, very much alike a long running equation that requires the answer to the previous equation.
            So if you have once process start halfway through the model it wont get far as it needs to know it's XYZ position first before it can continue to position the geometry it's creating. The long loads are a tradeoff for otherwise creating 1000s of individual models all using absolute positioning to create the final assembled model -each of which cannot be mass manipulated if you decide to change something- individual models and absolute positioning is a very VERY long process to do by hand.

            After the model is made, animated, textured, and exported for something like a game engine or such, it doesn't need to know about rebuilding geometry as much as it just gets wrapped up and becomes a hollow bunch of polygons which takes hardly any time to load and it's where multithreading can take over using functions like tessellation to add more detail, and stuff like textures just wrap around and can be loaded while the models basic polygon structure gets created.

Join the discussion!

Trending Stories Right Now