nVidia verified that the GTX 780 will be based on the GK110...

[Dimitris Tolios]

According to the SweClockers’s source, NVIDIA will launch it a new flagship GeForce model with Kepler GK110 GPU — the GTX 780. Also the site is claiming that the release might take place even sooner than VR-Zone reported this week. The new GTX 780 would arrive later this year, or more likely, during the first months of 2013 – so this could be March too.

 

Source: Videocardz.com

 

Some more reading on the GK110 GPU core: http://techreport.com/articles.x/22989

 

- Good news for the GPU rendering community - GK110 will be found outside of Teslas!

"The GF110-based Tesla M2090 card is rated for a peak of 666 DP gigaflops, and Nvidia claims the GK110-based Tesla K20 will exceed one teraflops". - so it is safe to guesstimate that this will be true for the GTX equivalents: the 780 will be roughly 50% faster than the 580.

 

Bad news: it is a huge chip, more than double the size of the GK104 used in 680s, and that still gets mediocre yields with the current maturity of the manufacturing process in nvidia's supplying foundries.

 

I wonder what AMD has cooking in the 8000 line to force nVidia announce a "gaming card" that won't be making them nearly as much money as the GK104 per card sold (unless the pricing for the GTX 780 is preposterous).

[Valtiel]

Compelling specs. Have they said anything about expecting a process shrink to make that many transistors viable in the consumer space?

 

So they're somewhat recycling the Kepler architecture for a SKU increment that suggests a complete generational refresh? I know this isn't the first time either NVidia or AMD has done it and this is more of an "unlocking" of the full Kepler than a rebrand but it's still less than comfortable. I suppose it does conserve R&D to make the 700 series derivative of this.

 

Neat.

[Dimitris Tolios]

The 110 is a huge chip. I don't know if they will really push the 780 hard, or just use it as a step-up from the GK104 before the GK114 hits production, recycling some of the GK110 chips that did not pass the validation for K20 Teslas so far...

 

That would allow them to claim the "fastest" card, and maybe develop the PCBs etc for the equivalent Quadro line.

 

The Kepler was never locked...it is just a "throttled" public release to ensure profits as they keep scaling performance 10-15% per generation.

The GK104 is a massive upgrade over the GF110 - much more than 15% - in the market that really makes them money - Games.

The GK110 is clearly not a gaming oriented design.

[Jeff Mottle]

I wonder if this is related to the rumors I've been hearing about NVIDIA. I've heard on a few occasions that in the near future there will be a lot less of a separation if any is made at all between their Quadro and GeForce lines. It seemed they might drop the delineation. All rumors for now, but the above certainly suggests a move in that direction.

[Dimitris Tolios]

I think the last 5-6 years at least, almost all of the Quadro and Teslas were based on the same hardware as GeForces as far as the processor would go. Yes, the pro cards would have in some cases ECC ram and/or more VRam all-together for the top models, but the actual performance differences were mostly driver based. It has nothing to do with TFlops/GTexels etc.

 

The GK110 was a chip that was probably not intended to be released on a gaming card. It is clearly a pottent number crunching GPU, aimed to fill the niche that the GK104 could not, despite it providing amazing improvement in the gaming field over the Fermi cards for consumers: mainly much better thermals/power draw (like 40% better while being considerably faster), and amazing profit margin improvements for nVidia since the new lithography was much more efficient per wafer for them.

 

There has been so much cry for a GK110 from the nVidia fanboys I guess, that they are doing it for the heck of it.

 

The fan-boys were never satisfied and always speculating:

* oh, GK is memory starved. Yes the 660ti is faster than the 580, but how fast would it be if it had more...

* oh, the 680 would do some much more with 384bit mem bus.

* oh, we don't have OpenCL performance like AMD, and we don't compute better than the 580, so we have to whine about it despite not actually knowing what OpenCL or even CUDA is and if we are doing something with it.

* oh, the GK104 is named with a 4 so it is not the full Kepler card, we were tricked...

* oh, that, or the other...

 

Silly forum-warrior stuff for "kids" that have 4-digits to spend every year for a new gaming PC, and too much time to brag about it on the forums.

 

GK110 is a relatively slow clocked GPU aimed probably for super-computers, and that's were all the GK110 production (at least the cherry picked chips) has been going last few months.

 

We will be lucky if we can get some of that in an affordable package.

As a gaming card, I doubt it will be much faster than a 690, due to core clocks, so I would not expect to be more expensive than a 690.

I don't know how many of the features will be locked out due to drivers, so I cannot speculate on the performance of the 780 as either a GPU renderer, or viewport accelerating alternative.