NVIDIA Maxwell Speculation Thread

[Picao84]

That's not the first time for a card of this caliber.

However, all I see is a GTX 750 Ti performing worse than a GTX650 Ti Boost.
Here we go for another round of erroneous naming..

Close enough to the Ti Boost, while using less power, on the same 28nm node. True, it might be a bit too much calling it GTX750, but its still impressive.
Further, your link does not show benches for the HD6950 and the GTX650 tested there is barely ahead of an HD7750 (which is already slower than HD6850, nevermind the HD6950), so I dont get your point.

 

[boxleitnerb]

That's not the first time for a card of this caliber.

However, all I see is a GTX 750 Ti performing worse than a GTX650 Ti Boost.
Here we go for another round of erroneous naming..

Rather you don't understand how naming works. 470 is faster than 280. 5850 is faster than 4870. Etc pp - it has been like that for years if not forever. What are you complaining about?

 

[DSC]

It's replacing the GTX 650 Ti and it only has 128bit memory bus, there's no way anyone will mistake it for anything else.

4+2 seems to be refering to the power phase design?

 

[Alexko]

Is it mentioned anywhere that it only uses 75W? I can't find it (then again I don't speak or read Chinese).

 

[Picao84]

Is it mentioned anywhere that it only uses 75W? I can't find it (then again I don't speak or read Chinese).

Well, that is what the rumours say. I don't read chinese either

On a second notice, maybe its just the non-TI that does consume only 75W...

http://wccftech.com/maxwell-gtx-750-nonti-inbound/

A card to compete with Bonaire...

EDIT - Looks like its not true for any? There is talk about 6-pin :s - Or maybe is advertised as a feature for OC, together with the 4+2...

http://videocardz.com/images/2014/01/GeForce-GTX-750-GPU-Picture.png

 

[Picao84]

It's replacing the GTX 650 Ti and it only has 128bit memory bus, there's no way anyone will mistake it for anything else.

4+2 seems to be refering to the power phase design?

Would power phase design be an important selling point of a card of this caliber?
But you are most probably right.

 

[xDxD]

http://wccftech.com/nvidia-geforce-...tured-features-768-cores-1020-mhz-core-clock/

768cc vs.384: simply more smx or more cc in each smx?

 

[itaru]

6pin power?

 

[Picao84]

6pin power?

Yeah, I mentioned on an edit. Could be a feature for OC? With the 4+2 phases and all on a card of this level

 

[itaru]

This will be the vest it? From GTX750Ti see new core Maxwell

Is it less than 120w in furmark?

 

[DSC]

http://www.anandtech.com/show/6289/nvidia-launches-geforce-gtx-650-gk107-with-gddr5

With that said, NVIDIA is equipping the GTX 650 with a 6pin PCIe power connector – strictly speaking it shouldn’t be necessary for stock operation, but since this is a GTX card NVIDIA has professed a concern over power consumption when overclocking.

Probably for that reason.

 

[iMacmatician]

Is it mentioned anywhere that it only uses 75W? I can't find it (then again I don't speak or read Chinese).

A Sweclockers report (posted earlier in this thread) appears to say that neither the 750 Ti nor the 750 need external power connectors, but the translation or the report might be incorrect.

768cc vs.384: simply more smx or more cc in each smx?

Here I will assume that this 768 number and the GPU-Z of the 750 Ti showing 960 CCs are correct. From these values, gcd(768, 960) = 192 implies a very narrow set of possibilities for the number of CCs per SMX (unless Maxwell can somehow disable parts of a SMX). A Kepler SMX has 192 CCs, so unless Maxwell is taking a step back in CCs per SMX, I would guess a Maxwell SMX also has 192 CCs.

 

[McHuj]

With all this stuff (allegedly legit) leaking, is it safe to assume that we're looking at an early February reveal as opposed to closer to the end of the month?

 

[Picao84]

This will be the vest it? From GTX750Ti see new core Maxwell

Is it less than 120w in furmark?

Would that be measuring only the card or the full system?

 

[iMacmatician]

With all this stuff (allegedly legit) leaking, is it safe to assume that we're looking at an early February reveal as opposed to closer to the end of the month?

February 18 is what the reports say.

From this post on AnandTech forums:

I was expecting the die size increase (assuming that number is correct), although I thought it would be a bit bigger.

 

[Picao84]

February 18 is what the reports say.

From this post on AnandTech forums:

I was expecting the die size increase (assuming that number is correct), although I thought it would be a bit bigger.

Yes, it is really strange, given that it has at least double the number of cores (if not more on GTX750Ti), for an increase of only ~32% on die size.

 

[Dave Baumann]

With all this stuff (allegedly legit) leaking, is it safe to assume that we're looking at an early February reveal as opposed to closer to the end of the month?

Bear in mind its Chinese new year now and where most of the board products are likely to be manufactured.

 

[Dave Baumann]

Yes, it is really strange, given that it has at least double the number of cores (if not more on GTX750Ti), for an increase of only ~32% on die size.

Other ratios could have been shifted.

 

[Picao84]

Other ratios could have been shifted.

Well, it seems to keep the number of ROPs and memory bus width, while greatly increasing the number of TMUs from 32 to 80, if rumours are to be believed.

 

[itaru]

Would that be measuring only the card or the full system?

Probably it is the whole system.
By the way, TDP of xeon E3 1230V2 is 69W.