Nvidia reportedly developing a dual-GM200 graphics card

Scorpus

Posts: 2,156   +238
Staff member

Nvidia is reportedly working on a new dual-GPU graphics card that would take two flagship GM200 chips, currently used for the GeForce GTX 980 Ti and GTX Titan X, and put then on the one board.

The card will allegedly feature "Titan" branding, and could be a doubling of the fully-unlocked Titan X. This would mean both GPUs would feature 3,072 CUDA cores, 192 TMUs and 96 ROPs, with 12 GB of GDDR5 available to each GPU through a 384-bit bus. In total we'd be looking at 6,144 CUDA cores and 24 GB of memory, although due to asset duplication only 12 GB would be available to games.

It should be noted that the specifications for this new gual-GM200 card haven't been confirmed, so it's possible Nvidia would opt for a dual GTX 980 Ti setup instead.

Like with all dual-GPU cards from Nvidia, it will perform essentially identically to having two separate graphics cards connected in SLI, as the dual-GPU card uses SLI to connect both GPUs. Cooling solutions tend to be a bit better on dual-GPU cards compared to traditional SLI setups, but you can be sure this card will still consume a lot of power.

Nvidia's dual-GM200 graphics card will reportedly be launching very shortly, although the price hasn't been revealed just yet. Hopefully Nvidia decides to price the card more sensibly than their last dual-GPU card, the $3,000 GTX Titan Z, which was considerably more expensive than simply buying two single-GPU cards and hooking them up in SLI.

Permalink to story.

 
I heard the last Titan Z was a flop, at $3000 the market of buyers was too small for that price bracket. And now looks like they are making another one.
 
I would like to 3-way SLI these...or maybe just go all our and 4-way SLI???
 
Of course Nvidia is going to compete with AMD on all fronts and they already announced a dual GPU Fiji.

AMD does have the advantage. With HBM, a dual GPU card is pretty much going to be a regular sized card. In addition to that, energy efficiency should be good because the memory placement and die placement are already decided. Not a huge need to optimize.
 
With DX12 potentially there will be access to 24GB of dedicated video memory, damn... Also the comment on cooling solutions being better on dual GPU cards, yes they may have to overcome a larger heat load but from what I've seen in past dual GPU cards they run hot and loud, perhaps this will be different. I'd buy two whole computers for the price of one of these, thank you very much.
 
I heard the last Titan Z was a flop, at $3000 the market of buyers was too small for that price bracket. And now looks like they are making another one.
Overpriced and under performing, although the Titan Z isn't the latest dual GPU card from Nvidia. That honour goes to the Tesla K80 ( dual GK 210's w/ 24GB of GDDR5) which has been in the market for a while, and just been announced for Microsoft's Azure cloud platform.

As far as a GeForce branded GM 200 product, it seems a case of halo wrestling with AMD as per usual. SLI scaling being what it is, bang-for-buck is a non-starter from the outset. A dual GM 204 would make more sense as a consumer product, but wouldn't be an option against a dual Fury card I suspect.
 
Wouldn't DX12 give all 24GB of VRAM available to games? I could've sworn it had something that would address this.
 
I have a 4K bought the Devils canyon and a 970 with 8G ram after which I found out I need 2 980 TI and 16 G ram and Really need a Intel 6 core option. To run on Max Res. So I will be getting this if not over 1,000 $$. By the way my 970 Kicks &^%.
 
I'm still waiting for pascal and hoping that it won't be a wallet killer.

Seeing as there entire range of cards will be moving to pascal there should be a card for everyones budget. I am curious as to the name of the new range of cards next year. Surely there aren't going to call them things like 1070GTX, 1080GTX etc they need a new naming convention I think. I had a 670 and was holding out for pascal but recently I got a free 980ti so im happy for the time being now.
 
Back