No GTX 780 Before March 2013 (VR-Zone)

[3DVagabond]

LINK

Interesting that they say not before march '13, not that it will actually be out by then.

 

[BD231]

When it comes to Nvidia, the company is heavily engaged on increasing the yields for its products.

So the real reason Kelper sucks ass is because of the cost cutting measures not because the 680 was supposed to be the 660. Are we going to be stuck with a meager 20-30% improvement on these new underwhelming GP-GPUless offerings moving forward? Cost cutting hasn't matriculated down to buyers in the least.

 

[ViRGE]

That's a reasonable estimate. I wouldn't expect any 2nd generation 28nm GPUs until then. Otherwise NV/AMD will have nothing new until the middle of 2014.

 

[Anarchist420]

Man, I was hoping that nvidia would bring out something that didn't have crippled DP performance and that could do full speedrgba16 blending with 32 rops or have 64 ROPs and non crippled DP performance.

"compute performance vs. gaming performance" is a false dichotomy as double precision shaders will be necessary or at least desirable for games in the future.

 

[poohbear]

i don't see any reason to release anything in March 2013, the majority of games can't even push a 660ti @ 1080p, so why release a new card so soon?

 

[Arzachel]

That's a reasonable estimate. I wouldn't expect any 2nd generation 28nm GPUs until then. Otherwise NV/AMD will have nothing new until the middle of 2014.

Three generations per node. If it won't happen on 28nm then definitely on 22/20nm.

 

[Homeles]

Otherwise NV/AMD will have nothing new until the middle of 2014.

And what would be the problem with that? It's not like a year and a half is a terribly long time. 20nm is supposed to be available next year anyways.

Three generations per node. If it won't happen on 28nm then definitely on 22/20nm.

I don't even see that being remotely plausible on 28nm. I don't see it happening on 14nm either. TSMC's 14nm node is supposed to be ready in 2015. Yeah, projected cost per wafer is high right now, but there's plenty of time for that to be negotiated.

 

[tviceman]

VR-Zone is not always known for it's accuracy, but that is an interesting read. I don't believe GK106 has been abandoned or never existed. There is just too much of a gap between GK104 and GK107. Also, Nvidia is going to have to move to faster vram if it wants to extract a noticeable performance gain out of it's 2nd generation Kepler parts. I've also read some articles that indicated Tegra 4 would have Kepler-GPU shaders, which was at least part of the reason for Tegra 4's delay. If that ends up being true, then it will actually be this generation that will incorporate top-to-bottom same GPU functionality.

Tegra 5 sounds like it is going to be a beast though (relative to what is available in the mobile space now).

 

[Spjut]

i don't see any reason to release anything in March 2013, the majority of games can't even push a 660ti @ 1080p, so why release a new card so soon?

Because Crysis 3 will be out by then😉

But yeah, I agree. I'm personally not planning on upgrading before at least 2014, by which time I expect the next-gen consoles to be out and have upped the bar.

 

[ViRGE]

Three generations per node. If it won't happen on 28nm then definitely on 22/20nm.

There's been nothing to indicate it will happen on 28nm. But you're right in that it will probably need to happen on 20nm.

And what would be the problem with that? It's not like a year and a half is a terribly long time. 20nm is supposed to be available next year anyways.

Just like the rest of the computer industry, NV and AMD businesses are based on having yearly product refreshes. They can't sell customers new products if they have nothing new to sell, which means they need to pace themselves.

I don't even see that being remotely plausible on 28nm. I don't see it happening on 14nm either. TSMC's 14nm node is supposed to be ready in 2015. Yeah, projected cost per wafer is high right now, but there's plenty of time for that to be negotiated.

The first rule of TSMC's projections is that TSMC's projections are always vastly overstated. 20nm will run about 2.5 years behind 28nm, and 14nm at least another 2.5 years behind that, if not more due to the fact that they'll be switching to 3D transistors at the time.

 

[boxleitnerb]

And what would be the problem with that? It's not like a year and a half is a terribly long time. 20nm is supposed to be available next year anyways.

As available as 28nm was in 2011 with all the supply constraints? :whiste:
Realistically, 20nm will take off after Q1 2014 at the earliest.

 

[Grooveriding]

Too bad, so sad. Sounds about right though, I would of figured February. No better time to launch the 780 than with a Crysis 3 bundle and some benchmarks to go with it. 😛

It is a long stretch for nvidia as they haven't left themselves any room for a refresh with GK104 like AMD had with the 7970GE. GK104 as GTX680 is a horse that is being flogged and run roughshod, there is no gas left in it with the 680 to make a refresh seem possible.

GK104 came on strong but feels to be slowing down fast against better performing and better priced cards from AMD now. I think we'll see some prce drops on 680/670/660 this time out and doubt we will see anythin new from nvida until the GK110 consumer card launches.

 

[tviceman]

It is a long stretch for nvidia as they haven't left themselves any room for a refresh with GK104 like AMD had with the 7970GE. GK104 as GTX680 is a horse that is being flogged and run roughshod, there is no gas left in it with the 680 to make a refresh seem possible.

I think the same, or even worse things, was said about GF100 and the gtx480. Being way too power hungry to make a refresh possible, and stuff like that. Turned out entirely different though. A refresh for GK104 is not only entirely possible but very likely, especially since 28nm production is only 8 months old. The problem lies in it's memory bandwidth; without significantly faster vram a refresh won't introduce much of any performance gain.

 

[tviceman]

I was hoping for december, but March does sound realistic. I still think the first iteration of GK110 won't be fully functional. 13 SMX's (2,496 cuda cores) @ 900mhz should give it a 40-50% improvement over the gtx680.

 

[AnandThenMan]

The first rule of TSMC's projections is that TSMC's projections are always vastly overstated.

So true. TSMC consistently under delivers, but they have little competition. GloFo has been promising promising ever since AMD cut the cord, still waiting.

 

[bunnyfubbles]

boourns

 

[Borealis7]

whoops!!
i'm tired...my second derp of the day

 

[Lonbjerg]

When it comes to Nvidia, the company is heavily engaged on increasing the yields for its products.

So the real reason Kelper sucks ass is because of the cost cutting measures not because the 680 was supposed to be the 660. Are we going to be stuck with a meager 20-30% improvement on these new underwhelming GP-GPUless offerings moving forward? Cost cutting hasn't matriculated down to buyers in the least.

You are trying to rewrite the english language? 😵

 

[RussianSensation]

When it comes to Nvidia, the company is heavily engaged on increasing the yields for its products.

So the real reason Kelper sucks ass is because of the cost cutting measures not because the 680 was supposed to be the 660. Are we going to be stuck with a meager 20-30% improvement on these new underwhelming GP-GPUless offerings moving forward? Cost cutting hasn't matriculated down to buyers in the least.

What cost cutting? 28nm wafers went up in price. That means NV has no cost cutting to pass onto to us, but instead it has to figure out a way how to NOT increase prices beyond a reasonable $400-500 level ==> die size reduction to improve yields and new focus on performance/watt for which GTX480 was criticized. When 28nm node matures and yields improve, they can add another 25-40% more with GTX780 part. By that point that GPU will be way faster than your card but it'll still cost $500+.

The only way it was viable for AMD and NV to sell you 500-600mm^2 die chips was to either wait until 28nm node matures, which would likely improve yields and allow the graphics card maker to stay profitable selling you such large chips at $500, or sell them at minimal profits which really isn't what shareholders want to see. The alternative could have been to increase prices well beyond the $500 threshold ==> $800-1000 for high-end and $500 for mid-range GTX680. Given that GK104x2 in GTX690 costs $1k, that gives you an idea that 600mm^2 of 28nm transistors costs an arm and a leg at TSMC right now. Instead of passing on these exorbitant prices (especially in this economy) beyond the historical $500-550 mark, both companies are just delivering what they can given their profitability and margins expectations.

Sooner or later an NV GPU will be 2x faster than your card but it won't be $250 for a while for that level of performance increase. You probably would need to wait until 2014 before there is a $250 Maxwell part that is 2x faster than a GTX480 OC. At the same time if you only paid $200-250 for your 480, chances are you bought it near the EOL (end of 2011?). It's not exactly fair to expect a $200 brand new GPU that's 2x faster. HD7950 OC will give you 40-50% faster performance at $310 or so. The improvement is there but it's slower than it has been historically. The power envelope has also limited that performance increase since no modern GPU consumes 270W of power like a stock 480 did.

I was hoping for december, but March does sound realistic. I still think the first iteration of GK110 won't be fully functional. 13 SMX's (2,496 cuda cores) @ 900mhz should give it a 40-50% improvement over the gtx680.

Very reasonable. I am glad you didn't join the 70-80% faster 2880 Cuda cores 1Ghz part bandwagon. I am going with 12 SMX 2304 SP. 😛

 

[tviceman]

Very reasonable. I am glad you didn't join the 70-80% faster 2880 Cuda cores 1Ghz part bandwagon. I am going with 12 SMX 2304 SP. 😛

GTX785 4-6 months later, fully unlocked at maybe 50mhz higher. :biggrin:

 

[moonbogg]

GTX785 4-6 months later, fully unlocked at maybe 50mhz higher. :biggrin:

And it never ends. Looking forward to the 780's because I have fond memories of my 7800gtx's and the name is similar so that means I will like them.

 

[OCGuy]

So the real reason Kelper sucks ass is because of the cost cutting measures not because the 680 was supposed to be the 660. Are we going to be stuck with a meager 20-30% improvement on these new underwhelming GP-GPUless offerings moving forward? Cost cutting hasn't matriculated down to buyers in the least.

If Kepler "sucks ass" , I hate to hear what you think of the rest of the cards on the market!

 

[aaksheytalwar]

The will probably release 2-3 months after 8970 I guess. 🙂

 

[biostud]

and in the mean time nvidia decides to slash prices on GTX 670/680 to match 7950/7970 *Wishes REALLY hard* 😛

 

[aaksheytalwar]

Unlikely. Nvidia doesn't believe in price cuts.