Nvidia GTX680 with 4gb vram coming soon :)

[Slinger]

http://www.evga.com/articles/00669/#GTX680FTW

 

I can't wait!!!!

[AubreyM]

It is very cool. I think that the price point will be higher than what I paid for my i7 PC. 4gb of ram would help with those really big scenes though

[DavidR]

I thought Kepler-series cards were going to be able to share system memory. I'd like to see some VrayRT-GPU tests to see if it works, or if we have to wait for GK110 Keplers. Anybody know?

[RyderSK]

I thought Kepler-series cards were going to be able to share system memory. I'd like to see some VrayRT-GPU tests to see if it works, or if we have to wait for GK110 Keplers. Anybody know?

 

 

Where did you hear this ? Very interesting.

 

I am excited for this card too, but tests outside of gaming show very poor results in OpenCL, yet to see how it performs in CUDA (Octane render), but it's pretty obvious nVidia dumbs down the general computing power in GTX series in exchange for only game performance.

 

Still, 4GB is 4GB ! We're getting somewhere.

[AJLynn]

I want to love this card but there's a bit more information needed before I'd make it a recommendation. When I first saw the spec, what jumped out at me immediately was that the number of "CUDA cores" (which is marketing speak for shader processors) tripled, relative to the 580, but the number of transistors only went from 3 billion to 3.5. What that means is that the number of transistors per shader unit must have dropped dramatically. This is taking the "wide but shallow" computing to a new level for nVidia, which could affect OpenCL and CUDA performance. Vray RT-GPU uses OpenCL and iray uses CUDA. (Similar renderers include Arion and Octane.)

 

Preliminary results are mixed. In some GPU computing tasks the 680 is extremely fast, while in others it's much worse than the Geforce 580 and competing Radeons. I see no high quality information yet on GPU rendering speed on the 680. Since the only thing that would make me recommend something this powerful with 4GB to a 3D artist (as opposed to a gamer) is GPU rendering speed, I'm holding off on recommending this until I see more.

[DavidR]

The next core revision (summer release) is going to be the big one for high-end gaming and Quadros.

[Slinger]

I want to love this card but there's a bit more information needed before I'd make it a recommendation. When I first saw the spec, what jumped out at me immediately was that the number of "CUDA cores" (which is marketing speak for shader processors) tripled, relative to the 580, but the number of transistors only went from 3 billion to 3.5. What that means is that the number of transistors per shader unit must have dropped dramatically. This is taking the "wide but shallow" computing to a new level for nVidia, which could affect OpenCL and CUDA performance. Vray RT-GPU uses OpenCL and iray uses CUDA. (Similar renderers include Arion and Octane.)

 

Preliminary results are mixed. In some GPU computing tasks the 680 is extremely fast, while in others it's much worse than the Geforce 580 and competing Radeons. I see no high quality information yet on GPU rendering speed on the 680. Since the only thing that would make me recommend something this powerful with 4GB to a 3D artist (as opposed to a gamer) is GPU rendering speed, I'm holding off on recommending this until I see more.

I totally agree with you Andrew. Testing needs to be done on this new "cuda" layout. I have a feeling it is not the same as the GTX580. I am more impressed that nvidia is bringing out cards with 3gb of vram.I will more than likely be picking up the eVGA FTW 4gb version when it is released ina few months, that is unless a new GTX comes out, which is already rumored.

[branskyj]

Octane user here. Also quite disappointed with the GTX680 in Octane- it actually equals GTX 460 in terms of performance (for now at least, since the software is not optimised for Kepler cards yet). Sad, considering I am eagerly waiting to upgrade my 2 GTX 460s. Here's hoping that the GTX 685s (later this year) will actually be what Kepler was expected to be.

The waiting continues.

[markperlick]

Octane user here. Also quite disappointed with the GTX680 in Octane- it actually equals GTX 460 in terms of performance (for now at least, since the software is not optimised for Kepler cards yet). Sad, considering I am eagerly waiting to upgrade my 2 GTX 460s. Here's hoping that the GTX 685s (later this year) will actually be what Kepler was expected to be.

The waiting continues.

 

I'm pretty sure the Octance guys will be able to optimize their code to GK104. It's some work since the architecture has somewhat changed. But personally I'd wait for GK110 later this year, it's a similar setup but more powerful if the rumors are correct.