boylerya


http://www.itproportal.com/2012/02/14/nvidia-kepler-gk104-rumours-continue/
"Some journalists - yours truly included - posted a full rundown of specifications for the next generation graphics processing unit (GPU). Unfortunately the realisation that they might be fake came only after hitting the publish button..."

LoL, at least one of the websites are admitting to their mistake.  Lenzfire sure made themselves a whole lot of new friends in the computer journalism field : ) 

Also, someone needs to learn how to spell:
realisation - a musical composition that has been completed or enriched by someone other than the composer.

Pretty sure that's British English... common usage outside North America, and it's a UK-based website. See Here 
 
Nice to see someone admit the mistake.  Lenzfire had a lot of us 'fired' up.

^-- that. They also spell other things with S where we use Z. Quite common, and a good indicator a person/website might be UK-based.

Then what if they wanted to say realisation with the meaning of a musical composition that has been completed or enriched by someone other than the composer?  Would they then spell it realization?

Something from Nvidia, maybe some mis-information, just to keep people from jumping into the AMD camp.  Time will tell.

Yeah I'll also stay on your side of the courrt michaelcollins for the moment.

michaelcollins


Something from Nvidia, maybe some mis-information, just to keep people from jumping into the AMD camp.  Time will tell.

I'm sorry...what?...
 
was distracted by your avatar

mlittrell


^-- that. They also spell other things with S where we use Z. Quite common, and a good indicator a person/website might be UK-based.

 
The only thing about british accents that annoy me is when a name that ends in A is pronouced as if it ends in er. Drives me up a wall.
 
Example. Jessica
US pronunciation jess-I-cA
English pronunciation jess-I-kER

Personally, I have been looking through a lot of different sites for info on the new nvidia cards. And I've decided that everyone of them is full of bull####. Just wait for Nvidia to finally put out the specs and stop speculating on something like this.

Anyone read any news on what/when Nvidia and Gearbox are showing at PDXlan?

Xarzi


Personally, I have been looking through a lot of different sites for info on the new nvidia cards. And I've decided that everyone of them is full of bull####. Just wait for Nvidia to finally put out the specs and stop speculating on something like this.

Forget even official Nvidia specs. Just marketting. Biased benchmarks can show whatever the company issueing them wants...
 
We have to wait for the card itself. The actual, functioning, working, card....
 
THEN we'll see.

Bruno747

mlittrell


^-- that. They also spell other things with S where we use Z. Quite common, and a good indicator a person/website might be UK-based.

The only thing about british accents that annoy me is when a name that ends in A is pronouced as if it ends in er. Drives me up a wall.

Example. Jessica
US pronunciation jess-I-cA
English pronunciation jess-I-kER

and UK English is suppose to be the proper definitive version of English or so I read

Kepler guys kepler....

kepler <3

Kepler was German...well, 'was' because he's dead, but I imagine his national origin doesn't change when you're dead, so Kepler still is German.
 

Bruno747

Example. Jessica
US pronunciation jess-I-cA
English pronunciation jess-I-kER

Or in other parts of the US they say "Jes-ca" or "Jeza-ka".  But yeah, British add an 'r' wheer there isn't and eliminate when there should be an 'r'.  Example, "Would you like to drink some wata".
 
But yeah back to topic, someone asked how can there be three times as many parallel cores and not be as hot, well if nVidia is eliinating the 2:1 shader frequency to 1:1 just like AMD and Intel, then that would cut the power in half even if the die were to get bigger to compensate for all the cores.

lehpron

But yeah back to topic, someone asked how can there be three times as many parallel cores and not be as hot, well if nVidia is eliminating the 2:1 shader frequency to 1:1 just like AMD and Intel, then that would cut the power in half even if the die were to get bigger to compensate for all the cores.

GK104 is considered to replace 560 Ti

1536 / 384 CUDA Cores
Is four times(4x)

Instead of 48 times 8 we have 96 times 16
That set off my BS Alarm
 
332mm² x 4 = (1328 x 28) / 40 = 929.6mm²
 
GK104 doesn't have hot clocks as it would be pretty hard to have
 
1.2GHz/2.4GHz

seronx

332mm² x 4 = (1328 x 28) / 40 = 929.6mm²

Might want to check your math since lithography is a length unit and die surface is an area, so (28/40)² = 0.49.  So four 332mm² and shrunk to 28nm would be 650.7mm²
Still kinda big but this runs on the assumption of the 1:1 shader-core ratio.

lehpron

Might want to check your math since lithography is a length unit and die surface is an area, so (28/40)² = 0.49.  So four 332mm² and shrunk to 28nm would be 650.7mm²
Still kinda big but this runs on the assumption of the 1:1 shader-core ratio.

GeForce 8800 GTX - 484mm² - 128 CUDA Cores - 575MHz
GeForce 9800 GTX - 324mm² - 128 CUDA Cores - 675MHz
 
x 65 ÷ 90
~350mm²
 
(65/90)²
~253mm²
 
350 - 324 = 26mm²
324 - 253 = 71mm²
 
GeForce 580 GTX - 520mm² - 512 CUDA Cores - 772MHz(+Hot Clock) - GF110
GeForce 680 GTX - ~340mm² - 512 CUDA Cores - 900MHz(If it followed the pattern and had hot clocks) - GK104
 
520 x 28 /40 = 364mm² <-- more closer to the industry estimate more accurate
520 * (28/40)² = 254.8mm²

seronx

lehpron

Might want to check your math since lithography is a length unit and die surface is an area, so (28/40)² = 0.49. So four 332mm² and shrunk to 28nm would be 650.7mm²
Still kinda big but this runs on the assumption of the 1:1 shader-core ratio.

GeForce 8800 GTX - 484mm² - 128 CUDA Cores - 575MHz
GeForce 9800 GTX - 324mm² - 128 CUDA Cores - 675MHz

x 65 ÷ 90
~350mm²

(65/90)²
~253mm²

350 - 324 = 26mm²
324 - 253 = 71mm²

GeForce 580 GTX - 520mm² - 512 CUDA Cores - 772MHz(+Hot Clock) - GF110
GeForce 680 GTX - ~340mm² - 512 CUDA Cores - 900MHz(If it followed the pattern and had hot clocks) - GK104

520 x 28 /40 = 364mm² <-- more closer to the industry estimate more accurate
520 * (28/40)² = 254.8mm²

http://www.youtube.com/watch?v=hEcuVR2vHwM

anyways...
this time around is the gpu suppose to operate independently of the cpu thus no bottlenecking in theory?
thats the idea anyways

SirWaWa


anyways...
this around is the gpu suppose to operate independently of the cpu thus no bottlenecking in theory?
thats the idea anyways

 
Unified Virtual Memory - CUDA 4.0/4.1
GPU Preemption - DX11.1
Autonomous GPU - Kepler
 
There is a reason why Windows 8 will force people to upgrade rofl
(All these things are done in Graphic Core Next they will just be better with Kepler)

but I don't want windows 8
will there be classic win 7 mode in 8?
then maybe
does ms realize or even know no one really wants windows 8 or is even needed yet they continue to push forward
what arrogance

Windows 8 will have Aero for desktops
 
Control Panel -> System and Security -> System -> Advanced System Settings -> Performance/Settings -> Visual Effects
and you can toggle between Metro(touchscreen) and Aero(keyboard/mouse)
With Aero being automatic for keyboard and mouse computers
and Metro being automatic for touchscreens
Windows 8 is very important

Windows Phone 7.5 -> Windows 8 allows for more cores(Windows 7.5 and older only support one single core meaning no Tegra for you)
Windows Home Starter just sucks for tablets and netbooks which Windows 8 fixes that
Windows 8 for desktops/enthusiasts marks the return of hardware accelerations that was lost with Windows Vista
 
Kepler and GCN already have a WDDM 1.2 WHQL drivers sitting around each HQ
 
The reason why Windows 8 will be great for Nvidia is ARM and Multi-core support
 
Windows 8 will be the home platform for Tegra Wayne(which uses Kepler)

SirWaWa

this time around is the gpu suppose to operate independently of the cpu thus no bottlenecking in theory?

That is impossible in the current infrastructure; the CPU (hence Central Processing Unit) runs the programs and tells each other components what to do, i.e. graphics card is a slave co-processor. 
 
The modern graphics card only does graphics (connects the dots); it doesn't decide how your A.I. functions or reacts to you, or understand what your mouse and keystrokes mean, let alone dealing with ethernet and audio controllers.  Furthermore, the modern graphics card does not have direct access to the hard drive, so it cannot derive the coordinates for the colors and textures of the scene it eventually builds, that's still the CPU's job.
 
If the graphics card were to run independently, then it means the GPU is coded with x86 (or ARM) instruction and actually runs programs by itself; we wouldn't need a fast CPU, let alone having a CPU in the computer at all.  Technically the GPU becomes the CPU, plus we wouldn't need the motherboard and separate system RAM because Vram bandwidth is much higher.  We would require the graphics card to have SATA/ethernet/audio controllers and ports to complete the tranformation.  What would we have?  High-end version of a console, where the CPU and RAM are soddered onto the PCB, meaning you can't upgrade.
 
What you would like, in theory, is a fundamental change in the way things are done.  It could happen eventually at the disapproval of AMD and Intel-- who's to say only games would be coded independent of the CPU; what if all programs were?  That is exactly what CUDA and other GPGPU compilers are capable of.  What we'd need is someone risk taking company to build a new OS from the ground up using CUDA, providing nVidia pulls an inverse of Project Denver and puts ARM in their future graphics cards (could be Maxwell).

Looks like Nvidia wussed out at PDX, and showed Tegra 3....AGAIN.....and borderlands 2.
 
Come on, some of these gamers need to stand their ground and DEMAND some info from Nvidia. 
 
If I was at this, I would have started going 
 
BOOOOO!!! WE DEMAND INFO ON KEPLER!!!

GK107 benchmarks
http://www.techpower...788#post2554788
 
 
3DMark11: P3818 (HD7770=3530)
3DMark06: 23893 (720p, noAA)
3DMark05: 31734 (1024x768, noAA)
3DMarkVantage: P16079

RE5 1080p, 16xAA, DX10:
Variable benchmark: 105.6fps 
 
(if this is real)

Xavier Zepherious


GK107 benchmarks
http://www.techpower...788#post2554788


3DMark11: P3818 (HD7770=3530)
3DMark06: 23893 (720p, noAA)
3DMark05: 31734 (1024x768, noAA)
3DMarkVantage: P16079

RE5 1080p, 16xAA, DX10:
Variable benchmark: 105.6fps 

(if this is real)

Scroll down to the post by Maban, will shed some light on things.

that doesn't prove it's not real ...just that it can be fudged(faked)
meaning it could be true regardless that someone could fudge it
 
It heads up.at least - we will see when the product is actually released
 

http://www.fudzilla.com/home/item/26069-kepler-gk104-cards-in-production
 
Cards now in production, not just the chip....finally? 

Too bad even after the cards are released I will have to wait for versions with extra VRAM 
2GB is weak for 1600p 16xq CSAA supersampling.