NVIDIA Maxwell Speculation Thread

DDs80CT.jpg
 
Some of those slides are a bit more "marketing" than I care for, but the benchmarks between 760M and 860M are better than what other alternatives they could have used (percentage increase as related to competitor, blah blah.)
 
Which GTX860 were they using in that graph? Apparently there will be some GTX860s with Maxwell and some Kepler... yay.
 
See, they still have problems running the ancient already 1080p. Guess after how many years in the future we will be able to enjoy 4K or even 8K... That will happen probably in 2030!

I am not going to buy anything until they severely improve my 1600 X 900 on my laptop which is pretty damn close to 1080p and from the ancient year 2008, with hardware components dating even further back to 2007!

Lack of innovation, that's it!
 
Sure...

Back in realityland Maxwell in laptops is a great thing. Even GM107 can handle most games at 1080p without toasting your nutsack in the process.
 
I find the 840M the most interesting, i'd love to see a price sheet of these new parts.

Nevermind, it uses DDR3.
 

Given that 860M and 850M are fully enabled GM107 chips, I suspect 840 and 830M may rather be harvested GM107 parts with 1 or more SMM's disabled.
Shocking..if these numbers are correct then Intel is way behind on efficiency..even with a significat process advantage on their side! Would be interesting to see a comparison with Maxwell as well.
Which GTX860 were they using in that graph? Apparently there will be some GTX860s with Maxwell and some Kepler... yay.

Indeed..as if rebranding wasn't bad enough...

So now you have two GTX860M's:-

1. Kepler based with 1152 "cores" at 797 mhz base clock with 2.5 ghz 128 bit GDDR5
2.Maxwell based with 640 "cores" at 1029 mhz base clock with 2.5 ghz 128 bit GDDR5

Interestingly..Maxwell based 860M has a higher base clock than the desktop 750Ti.
Maxwell in laptops will truely shine once they move it to the next process node. If we are lucky- next year ;)

I expect the biggest gains to come with the process node after the next one, i.e. 16nm Tri-Gate/FINFET
I find the 840M the most interesting, i'd love to see a price sheet of these new parts.

Nevermind, it uses DDR3.

Good thing you added the edit..was about to ask what exactly was so interesting about it :???:

I seriously doubt its going to beat 740M's performance. And given the 64 bit interface, it has roughly one-third the memory bandwidth of 850M. Not surprising then that 850M beats it by 70% :rolleyes:
 
Indeed..as if rebranding wasn't bad enough...

So now you have two GTX860M's:-

1. Kepler based with 1152 "cores" at 797 mhz base clock with 2.5 ghz 128 bit GDDR5
2.Maxwell based with 640 "cores" at 1029 mhz base clock with 2.5 ghz 128 bit GDDR5

Interestingly..Maxwell based 860M has a higher base clock than the desktop 750Ti.

That is a bit odd, but at least on the desktop, modern NVIDIA GPUs really never run at the base clock. In fact they often operate above the rated boost clock :smile:

My HD7950 also maintains its boost clock all the time. Almost wonder why they bother calling it a boost clock.
 
Last edited by a moderator:
Shocking..if these numbers are correct then Intel is way behind on efficiency..even with a significat process advantage on their side! Would be interesting to see a comparison with Maxwell as well.

Why don't they show comparisons to AMD's cards?
AMD cards usually offer better performance/ price ratio (and overall characteristics) but for some unknown reasons OEMs flood the market with exactly the opposite.

I expect the biggest gains to come with the process node after the next one, i.e. 16nm Tri-Gate/FINFET

Mhm, 16 nm is next after 28 nm when they skip 20 nm altogether.
 
Why don't they show comparisons to AMD's cards?
AMD cards usually offer better performance/ price ratio (and overall characteristics) but for some unknown reasons OEMs flood the market with exactly the opposite.
not unknown at all. Main reason for nvidia success in laptops is Optimus. In fact, it has been said many times that nNidia is a software company and the launch of Maxwell for laptop is another giant proof. Look at the software ecosystem on the green team, the GeForce experience package (auto tuning, shadowPlay, GameStream) + battery boost, it's all great ideas, it mostly works great out of the box and it adds value to the end-user. AMD is so ridiculously outpaced in this field that it's pathetic. Enduro never really worked and still today, 3 years later after Optimus launch, it lags immensely behind.
Moreover, OEMs don't only buy silicon but also peace of mind. I don't know if its true, but one US OEM, said that it gets 10 times less support calls with Nvidia on their laptops than with AMD. It's surely exaggerated (NV propaganda ?) but I will not be surprised at all that the truth is not too far from this hard statement...
 
That's pretty much exactly what the Anandtech article concluded: Optimus just works. Better additional software products etc.
I don't think the mobile Kepler generation was much better than AMD's offering in terms of power (unlike higher end), but unless there's some unknown mobile AMD silicon lurking that we haven't heard about, I don't think they will stand a chance against Maxwell this cycle.

Maybe their APUs are competitive one way or the other, but if the general public has the same (irrational!) distrust against AMD CPUs that I have, that's not going to help them.
 

Not too surprising since Maxwell is "mobile first" to begin with. What is surprising is that GTX 850M has more CUDA cores than GTX 750, and that GTX 860M has higher base GPU clock operating frequency than GTX 750 Ti (without giving up any functional units either, albeit with a bit lower mem. bandwidth in comparison).
 
Last edited by a moderator:
So now you have two GTX860M's:-

1. Kepler based with 1152 "cores" at 797 mhz base clock with 2.5 ghz 128 bit GDDR5
2.Maxwell based with 640 "cores" at 1029 mhz base clock with 2.5 ghz 128 bit GDDR5

The performance of Kepler 860M and Maxwell 860M should be very very similar considering that the Maxwell variant has ~ 30% higher GPU base clock operating frequency and ~ 35% more perf. per CUDA core in comparison to the Kepler variant (with the same memory bandwidth to boot). I can't think of a good reason to have two different variants unless certain customers would rather market the variant with more "cores" (at the expense of higher power consumption most likely). Note that it wouldn't help to call one 865M and the other 860M if they are performance equals.
 
The performance of Kepler 860M and Maxwell 860M should be very very similar considering that the Maxwell variant has ~ 30% higher GPU base clock operating frequency and ~ 35% more perf. per CUDA core in comparison to the Kepler variant (with the same memory bandwidth to boot). I can't think of a good reason to have two different variants unless certain customers would rather market the variant with more "cores" (at the expense of higher power consumption most likely). Note that it wouldn't help to call one 865M and the other 860M if they are performance equals.


GK104 is bigger than GM107 on the same process, so Nvidia probably makes more money selling the GM107 variant.
 
Back
Top