http://www.guru3d.com/news_story/nvidia_geforce_gtx_880_details_surface.html
 
Memory is back to 256mb bus width???  When you look at the 880 vs 780 / Titan...doesn't look like a dramatic upgrade...

And even if these specs were true, the power per core of the Maxwell arch is 35% more powerful than Kepler. So 3200 Maxwell cores = 4320 Kepler cores. That's nearly double the power of a GTX 780 with less power draw. Plus, Nvidia has addressed the fact that they can use less memory bandwidth with the higher L2 cache in Maxwell with similar performance.

Waiting for 980Ti equivalent >:) 

I posted this a few days ago
http://forums.evga.com/FindPost/2146251
 
I even found an Nvidia release chart 
not much to say.... don't expect the full GM200 until 2015 ( 3 arm chips)
GM204 supposedly has 1 arm chip
 
GM204 is june/july
GM206 is aug/sept
GM200 is slated for later to make sure of higher yields and better silicon (mature 20nm)
 
there will also be an OEM part as well 
 
 
PS: the current Nvidia drivers support GM200,GM204 and GM206 already
as was reported by 3dgeeks by looking at the driver files

http://www.nordichardware.se/Grafik/laeckta-specifikationer-foer-nvidias-nya-maxwell-baserade-flaggskepp-gtx-880.html 
 
Now talking though several sources about what we have to expect from the upcoming flagship GTX 880 which should be launched later this year. 
[<font]The case of a performance video card that will take over from current GTX 780, and the Maxwell-based graphics chip is said to perform even better than a fully featured GK110 chip that we find in the current GTX 780 Ti and GTX Titan Black.
[<font]According to the sources, we can inter alia expect 3200 CUDA cores, compared with the 2880 sitting in the GTX 780 Ti and Titan Black. These consist of a total of 7.9 billion transistors, and will deliver 5.7 teraflops of floating point performance with single precision. The chip is said to have a base frequency of 900 MHz with a boost frequency of 950 megahertz.
 
 
------------------
again take it with salt - same sources cited
Im unaware if they have other sources - they are not citing them

Just too likely to be an April Fools joke.

first of all the architecture is different - so don't compare kepler to it or even the 700 series maxwell cards (because they aren't really maxwell -maxwell lite or maxwell/kepler hybrid)
 
the maxwells released already(low end..ie 750) are called maxwell 1.0.... the 800 series are maxwell 2.0
they are missing a arm core -  so in order to have them perform like ones with a arm core they had to keep some semblance of kepler - a hybrid chip
the are power saving - shutting off unused cores when necesssary - this would reduce memory bandwidth because the cores aren't querying memory or the MB CPU or being fed by MB CPU
 
the real maxwells have an arm core and more L2 cache - of course it would you have a additional controller CPU - which would need have code and prefetch worked and reuseable graphics work
this chip is also power saving - throttling cores - controlled by arm chip - so the card doesn't query system ram or MB cpu going what?? I need work or the MB CPU pester it
the arm does this - and only if there is a demanding work to be done
you have to remember during none peak times it's prefetching work and storing stuff and with more L2 cache you aren't throwing as much away - you are reusing more
this saves bandwidth
 
so maxwell would out perform kepler with less memory bandwidth
 
the rops and TMU count were changed from kepler and even maxwell light - so don't use that as a guideline either
 
if you look at the specs it falls in the line of the 680 or 780 cards 
a 20% increase over previous arch - with 10% power drop (30% increase) im assuming a refresh would be similiar - so nearly double performance over kepler
and we aren't seeing the full silicon yet
 
if a full maxwell(titan 2 or whatever they want to call it) is spec'd with 3 arm cores and if you keep arch the same - that's 9600 cores and 3 arm chips
- im assuming it be more around 6k cuda count....that's a full double over 880 (or titan as it is now)
 
 
Also everyone makes note of the single 6 pin connector - talked about
isn't the PCIe 2.0 convention 150 WATT thru MB so a single plug would offer an additional 75watt for 225 which brings you close to the 230 watt specified
im sure you could probably squeak a modest 5% more out of the MB or connector
and the card is 230 watt max - and probably runs much lower than the max specified
and under PCIe3.0 it wouldn't even be an issue
 
pcie 1.0 is legacy and shouldn't be supported
 
 
 
 
 

Sajin

lehpron

Sajin
ARM Core = Reduced CPU Overhead = More Performance. 

No reason to still choose Intel or overclock? 

I'll always run a Intel CPU, until something better replaces them that is. I'll always overclock too, free performance is good.

Have you thought about what reduced CPU overhead ultimately means?  It means the use of a co-processor to better feed the GPU cores = less CPU utilization, that is the whole idea.  What good is overclocking (let alone using Intel, or multi-core) if nVidia forces GPU-intensive and GPU-bounding through this architecture into future games?  If nVidia hadn't tried this, then it would be more reason to get an 8-core i7 as future games will be more multi-threaded to feed all the parallel processor cores of a future GPU using every CPU thread available-- but that isn't happening.  
 
 
Maxwell is the beginning of an attempt to pull the bottleneck out of the CPU reducing the need for overclocking, let alone using a better CPU; but you still praise your choices as if you have trained yourself to blow off and ignore any discussion that suggests otherwise.

lehpron

Sajin

lehpron

Sajin
ARM Core = Reduced CPU Overhead = More Performance. 

No reason to still choose Intel or overclock? 

I'll always run a Intel CPU, until something better replaces them that is. I'll always overclock too, free performance is good.

Have you thought about what reduced CPU overhead ultimately means?  It means the use of a co-processor to better feed the GPU cores = less CPU utilization, that is the whole idea.  What good is overclocking (let alone using Intel, or multi-core) if nVidia forces GPU-intensive and GPU-bounding through this architecture into future games?  If nVidia hadn't tried this, then it would be more reason to get an 8-core i7 as future games will be more multi-threaded to feed all the parallel processor cores of a future GPU using every CPU thread available-- but that isn't happening.  
 
 
Maxwell is the beginning of an attempt to pull the bottleneck out of the CPU reducing the need for overclocking, let alone using a better CPU; but you still praise your choices as if you have trained yourself to blow off and ignore any discussion that suggests otherwise.

I could still use the Intel CPU for better benchmark scores. 

GTX880 would be going after GTX780 and nothing else, this isn't an elephant chain where one follows the other.  If we go back to fall 2005, nVidia frst made a 130nm G70 die calling it GeForce 7800GTX with 20 pixelpipes, yet the flagship 90nm G71 "GeForce 7900GTX" didn't appear for 6 months with higher frequencies with 24 pixel pipes.  When 8800GTX came in November of that year, while everyone compared it to the 7900's, it actually succeeded the 7800's.  Granted, we never saw an "8900GT/GTX", AMD's R600 delay prevented the need-- sounds familiar?  R600 was late and Hawaii was late; both times nVidia was allowed enough time in a deFacto-monopoly to overcharge without drastic improvements, i.e. 8800 Ultra versus Titan.
 

Sajin
 
I could still use the Intel CPU for better benchmark scores. 

I won't argue that, benchmarking is largely prescripted rendering with no user interaction.  This should make them less CPU-bound, but most popular presets are done with CPU-bound resolutions, yet high details so graphics cards have something to do.  In the end, the scores are just formulas manipulating frame rate math; which adds to CPU utilization.

I would keep my eyes peeled for a paper launch next month
 
there is an official Nvidia announcment already posted on the web saying 2H (early) - you can find it from a story a few months ago
Ive also seen a few posts saying june/july and may (since TSMC is ahead on the fabs)
 
Nvidia already has 20nm silicon - from initial 2 fabs producing silicon - it is a limited supply - 1000's of units
 
3dgeeks already spotted gm200(gm210),gm204 and gm206 in the latest drivers
 
the key is Nvidia won't launch until TSMC production is ramped up - 2H when the new fabs come online
otherwise there will be a shortfall as demand for the product hits and find that TSMC can't meet demand - because of low production and low yield
the silicon Nvidia has on hand will be enough for an initial launch(month or so)...but not for a prolonged season
 
don't expect gm200 until 2015 - Nvidia has Server/mainframe demands(commercial interests) - tesla to fill at $4k a card rather than fill our little greedy hands with them
at least not until yields are better and production is enough to fill commercial interests as well as ours
 
How credible is the info?
some of it is true  - silicon supply is accurate
the official word from Nvidia is spot on
Drivers already support GM 200, 204, and 206 - so silicon is incoming soon
 
dates are floating - TSMC has announced early production so dates are moving
Ive seen gm204, june/july - gm206 Aug/sept
but recently Ive seen May as a release date for gm204...so the rumors floating around could be spot on the Specs
 
they also may play a waiting game - push out tidbits - see what AND brings to bear and bring out a counter-punch
 
Tweaktown had a recent Visit at Nvidia headquarters this month - they even had a visit in the "secret area" - where all the new tech under NDA is kept
other than tweaktown reporting a top secret military case (which we seen before) they aren't saying much
 
And I may have found a early glitch from an Asian retailer spouting GTX880 - no avail prices yet - no stock either
 
 

like I said...more likely incoming
 
more news

 
http://videocardz.com/50301/nvidia-preparing-maxwell-gm204-kepler-gk210-gpus 
 
 
What I did find is 20nm silicon yields are pretty good  - at least from Apple's standpoint  - over 50% and as high as 65% - with yields improving
 
I don't believe the story that it's 28nm
how do you go to 3200 cuda cores and advance memory support.. more l2 cache and an arm chip and still maintain the same size(socket size) on 28nm? - doubtful
20nm production is ongoing since last year - ramped up to 2 more fabs by may/june
 
TSMC made some news this week as well - record profits this Quarter (23% increase) based on 28nm and new 20nm - with revenue increasing next Quarter
the march news that TSMC had material problems on 20nm was temporary(less than a week) and didn't impact production or timelines (there own words)
so news there was going to be a delay in march was wrong
 
 
assume 1-2 months for stocks to build for release -puts everything at may/june/july
 
 
AMD story 
 
No 20nm GPUs this year
http://videocardz.com/50299/amd-20nm-gpus-year
 
take that with a grain of salt - they said all their product is in 28nm - that would be correct
however adding 20nm late in the year or half year would not make that false
 
you would still need to produce 28nm cards for awhile - until total line replaced
 
they said 20nm is in design and will enter production this year and then finfet which is next year
 
 

Will have to wait and see what the 800 series will bring. But for now not worth arguing over, or getting panties up in a bunch, or speculating until we see some solid proof. ATM pretty happy with the GTX 780 TI first time in awhile I have been able to slam up all settings in any game at 1920x1080 and not take some type of performance hit that is slightly noticeable. Just wish it was as powerful in directcompute, PhysX, gpu rendering, gpu real time rendering, and the massive amounts of gpu power and vram for hair simulations as a Titan Black. That is my biggest complaint but I didn't spend 1k+ to get those things so really don't have a leg to stand on to complain. Just take it as it comes guys don't get excited.

updated news
 
Recently, Videocards comes with something really interesting a latest flowchart, that GM204-A01 chips are now in testing phase, and will equip with 8Gb memory. The memory is currently higher than Gtx 780/Gtx 780Ti. If the 8Gb memory rumor is true, then the Gtx880 has 256 or 512bit memory interface, but this is still we can say a barrier because due to the current ram technology Higher is not possible.
 
“GM water Cooled plunger” which is listed on the Spec Sheet, So are the new Graphic card from Nvidia are based on Factory Based Water Cooling??
 

http://tech4gamers.com/gtx-880-could-be-equipped-with-8gb-of-memory/ 
 
 
 
Nvidia already testing GM204-based graphics cards internally
http://www.kitguru.net/components/graphic-cards/anton-shilov/nvidia-already-testing-gm204-based-graphics-cards-internally/ 
 
 
 
Wei Feng network May 6 news, latest news online a few days ago,
 
Nvidia's new flagship GeForce GTX 880 engineering sample (ES) has started trial production. 

[<font]　　Sources said, GeForce GTX 880 engineering samples Nvidia has completed the development of R & D centers in India, has said the precious parcel packing, shipping by China through pilot production plant, will further test and complete the final version. If nothing else, the new flagship GeForce GTX 880 will be released before the end of the year is bound.
 
GeForce GTX 880 graphics card is still using TSMC's 28nm process, spread out from the list of specifications, the card will build on the flagship Maxwell GM204 core architecture, the current revision of the trial production of "GM204-A2", while also supporting DX 12, standard capacity up to 8GB of GDDR5 memory.
　　Nvidia GM204 GK104 core successor, reportedly has 3200 CUDA cores, also includes 256-bit memory interface. There is no doubt that if Nvidia really put on a 8GB integrated PCB, bound with a very high density flash memory chips.
　　As for pricing, India spread its structure is about 47,000 rupees, about 4880 yuan. The current generation of the current market pricing flagship GTX 780 TI also about 5,000 yuan, so the performance is between 880 and GTX 10% higher than the 780 TI, whose final price will be higher.
 
 
 
 
looks untouched - hell if I know if it's real
 
http://tech.feng.com/2014-05-06/Nvidia_s_new_flagship_GTX_880_pilot_light_seconds_cash_8GB_Memory_583817.shtml 
 
GM204-B2 = GTX970  -20nm  due 2015

Associates claim that NVIDIA will present new graphics solutions for the desktop segment even before the exhibition Computex 2013, which will start in late May
xtreme.com
 
google search link for videocards.com even says that Nvidia may have something to say within next week or two
 
Chinese online retailer broke NDA
 
also found a site that says cards are being assembled for final production
 
rumors abound
 
even if this is true - stock would be relatively low -  ie Paper launch
 
Id put more stock in a least 1 more month for testing before production
But Nvidia had stated they wanted to keep to their Release timelines ..so heaven knows it could be true
 
 
There is also info that 980 is scheduled for June 2015
so refresh 1 year after 800 series launch
 

tech news austrailia