Nvidia BigK GK110 Kepler Speculation Thread

about 50% more performance on gtx 680 means about 750 dollars MRSP (gtx 680 is 499 dollars)

 

Usually newer generations of hardware are more cost-effective than older generations.

 

799$ at launch?

 

That performance is making my mouth water! The price is crazy though, I wouldn't even pay half that for a GPU.

 

I do t upgrade that often- heck my 285s were doing a stellar job but the lack of DX11 was getting to me. If it is released at a 799 price point ugh too rich for my blood but maybe I can just grit my teeth order 2 and not upgrade for three four years. I just hope DX12 is not right around the corner.

 

I have no idea. But let me mention something else: there is really no good reason why anybody should linearly associate processing power and price. Or, at least, think that a linear association is fair.

To me, it comes down to this:
1. Will this part let me do things that a different, less expensive part won't be able to do?
2. How long will it be before I should expect to replace this part?

 

One thing..or a few things I find interesting.

Despite x1x GPUs traditionally being a new product line compared to x0x GPUs, GK110 sounds like it will be marketed in the same product line as GKx0x GPUs.

So that seems to imply that there isn't likely going to be a new product line released in the next 6-12 months. GK110 is basically it for anything new for a while and there aren't any GK11x derivatives.

And from a margins and thus profitability point of view, it allows them to release it at a much higher price than they would have been able to otherwise.

If it was a GTX 780, they would likely be limited to a 599-699 USD price. Enthusiasts are used to new product lines being faster than the previous product lines without pricing scaling in line with the performance increase. In other words +50% performance obviously doesn't lead to +50% price going from one product generation to the next. As we saw with the last few product generations +30-50% performance leading to any increase in price leads to a mass forum warrior uproar.

By keeping GK110 in the 6xx product stack it allows them to scale price closer to how performance is likely to scale (not 1:1 but closer). In other words, they can charge a lot more for it being a 6xx product than they would if it was a 7xx product. This is absolutely brilliant from a marketing point of view, IMO.

Regards,
SB

 

Reminds me a lot of the 7800GTX 512MB. Which also gives me this uncomfortable (and possibly totally incorrect) feeling that the Kepler technology is pretty old and won't be long until it's looking pretty outdated.

 

The "Titan" name allows Nvidia to use it as their halo part for this and next series as well.

In the end it hardly matters as these will be sold in the thousands or tens-of-thousands at best. There just isn't a market for desktop graphics cards costing this much, it's a professional card used as a desktop halo part and that's it.

 

Kepler probably wont look outdated by anything on 28nm which might be what NV and AMD are using up through 2014. As for this gpu its doubtful anything will outperform it until 20nm. Unless that is if AMD is serious about making huge 400mm+ dies again which is unlikely.

 

I'm not entirely sure that high-end parts like this one are solely for the purpose of a halo effect. Bear in mind:

1. A significant fraction of the cost of such a part is born in the fact that engineering work in its development is shared among other parts. So the marginal cost of adding this one product to the entire line is not so great.
2. High-end parts are sold at significantly higher margins than lower-end parts, such that they don't need to sell as many to recuperate their R&D costs.
3. The professional version of the card will sell at higher margins still.

 

This card wouldn't exist without Nvidia's grip on the professional market though. This is why AMD doesn't attempt to challenge them with dies so big because for the consumer market there is no way they could make the money back.

 

Where dies this argument come from? nVidia has had big cards like that which are used for both consumers and professionals for a long time Why would this be any different?

 

Thanks Olivion on XS ..

So if we look at it and it is trustable 14SMX, 875mhz Max ( turbo? ), 6Gb.

Spec look similar of a K20 outside the clock speed ( L2 cache etc ) .. http://clbenchmark.com/compare.jsp?config_0=11905561&config_1=14470292

 

GTX680 is 34% faster than Titan in that Physics: SPH Fluid Simulation

 

http://clbenchmark.com/compare.jsp?config_0=11905948&config_1=14470292#environment
I thought it had some larger caches - can't be seen there (apart from the obvious total increase from 8 to 14 SMX).

So at 875mhz compared to 1059 for 680 we got +44% shader/tmu and +23% setup/rop (and likely +50% bandwidth).

 

That puts the final nail in the coffin of those benchmarks showing >60% performance improvement. Something like 35~40% seems more realistic.

 

NV's OCL support state is barely good enough for listing device specific capabilities -- don't look for reliable performance metrics on this API.

 

As i have some doubt about this result i have run it on my setup ( 2x7970 ), so the problem appears when you run SLI or CFX .. i got barely 6000 points, i disable the CFX and run it normally ( single card ), i goes to a normal score 20 000 + .. Only ray tracing is supporting CFX or ( 2 gpu? ) it seems.. ( 554'000pnts )

Huum, i dont know, cant really conclude anything.

 

From Arab PC World: "Exclusive: the final specifications for the card GEFORCE GTX Titan" (original).



The report also mentions a February 18 release.

The title appears to be in two different fonts and colors, so it seems a bit suspicious to me but I won't write it off as fake just yet. The boost clock seems really high to me, given that most rumors have the base clock below 900 MHz. If the core clock is actually on the low side, then maybe the boost has been improved or it shows the max boost. Neither clock matches that of the CLBenchmark's clock of 875 MHz either.