As I understand it, the G80 will

Support FP24 HELP
Support GDDR5 ME
New AA Mode:VCBB SPOCK

humm it could be that they are using the second as the stage between the dedicated gpu/grd and a second gpu that can write to ram space so that the card has a gig to play with 512 dedicated 256 shared and 256 virtual that would reside in ram or swap space ie your hard drives swap file. The only reason this makes any sense is it is very simalar to the way the ramm on the TC chips is used in a far less complex way.

Unified architecture refers to the ability to process texture, vertex shader, pixelshader, in whatever order and however much you want before raster. The xbox 360 uses the x1900 core, and emulates this as the x1900 core is able to move data around the pipe in a non regular fashion. Nvidia may do the same, instead fo making one extremly complex usit that can do both vertex and pixel shading.

water is definitely possible- sapphire has a water cooled x1900xtx. Corsair has a closed look water cooler. There's no reason why another company can't do it. There's no need for a reservoir in a closed loop. The cooler took one slot - that's all! The big benefit was it was quiet.

another possibity for the extra memory - MAY be used for free AA/AF. The xbox 360 has 10MB of ram on the chip in a seperate package for free AA. It's possible the extra ram and chip can provide free high quality AA and AF with almost no hit - there IS a new VCFF mode. Why more than the xbox? well, 10MB can hold a 1920*1200 32bit color image. It takes close to 50MB to hold a 2560x1600 32bpp triple buffered image, super sampling can increase that easily (though only xgi does super sampling nowadays) and I dont have a clue if AF can be done off the chip.

nvidia didnt kill the 7900gt with something cheaper, they killed it off to introduce TWO cards. the 7900gs and 7950gt. The gs is so close there's no difference really, what I think is more important, is it gets rid of the back log stock of defective 7900gt's. Many people had problems with it - some blame the vrm for not pushing enough volts or simply failing. There was so much chatter, evga's (or was it xfx's?) forum had to close. EVGA actually released a fixed version of the cards.

Who spends US$500 on a video card? I mean you could buy an Xbox 360 or a PS3 for that $$.

I have my Xbox 360 and can live with a ~$200 video card for PC gaming. When are the GPU makers going to design low power high performance chip like conroe?

The all makes some sense based on wat I already knew.

nvidia is using a semi unfied shader setup. Basicaly 2 sets of shaders with one set optimized for pixel and vertex openerations and the other set for vertex only. BUT just optimized not dedicated.

By optimizing them they cut size and boost speed for most of the shaders.

The memory size makes perfect sense.

The old cards use 4 64 bit controllers these use5-6. The memory of course makes perfect sense too. Instead of 4-8 chips they use 6-12.

yeah. I thought nVidia previously stated that it would not go the Unified Shader route. Now they've suddenly changed their minds? Is that because of DX10?

Greetings!

# 384-bit memory interface (256-bit+128-bit);
# 768MB memory size (512MB+256MB)

Could it be a dedicated 256MB GDDR3 for Vista graphics and 512MB GDDR4 for games using the 256MB GDDR3 as a buffer/cache thing?
Lot of VRAM management though...

I think the G80 might be really two seperate GPUs if the 384 bitness is true. The main GPU gets a 256bit bus that connects to 512MBs of VRAM. The co-GPU that probably does something very interesting and gets the 128-bit bus which connects to other 256MBs of VRAM. The GPU and co-GPU have their interconnect, probably a special form of SLI.

I suspect the G80's release speed to be around 600-700Mhz and requiring a cooler simliar to X1900XTX.

If the transistor count is right, then I'll be waiting until a 65 nm shrink before buying this.

I guess it's supposed to be implied, but I'm surprised at the lack of mention of DX10. Or am I missing something here?

I take these numbers with a gain of salt.

Memory amount and Memory bus makes no sense since 384 is not a power of 2. Unless, the G80 is acutally two seperate chips on the same board. 1.5Ghz is simply not possible for a GPU that is fab at 65nm. ATI has trouble getting a 16 pipe GPU to go beyond 750Mhz without water cooling.

Hybird air/water cooling is 100% pure BS. Water cooling uses air cooling on the radiator whetever its passive or active!

What is certain is that G80 will use GDDR4 and have DX10 support

It's been speculated the dual memory bus is because of a feature of the xbox. Possible dedicated AA memory.

Guys, I just really would like to know if this thing supports the "Chuck" patch. Cuz "Chuck" is cool.

This kind of information makes the hairs on the back of my neck stand on end.

If its true, its pretty amazing stuff.

I wouldn't put it pas nVidia to produce an uber card to compete for the new series of performance crown tossing. When quad sli cards came out, the improvment on a single card basis was pretty reasuring.

However, I seriously doubt that these specs are even close to whats coming out. It would be to good to have this supposed card competing with the monstrous R600 speculation cerculating. All that can be said is their will be an improvement, and it will probably be a pretty big one.

this is a load of bull. nvidia is not about kill off their very lucrative G71 gpus with something that performs 10x as fast and gobbles 2x the power. Just think people. Expect unified shaders and refinements with about 150% of the current topend performance. These companies are trying to real in their power requirements, not create an oven inside your computer. This reminds of the AMD reverse hyper-threading crap. Its actually nothing but FUD in a subtle way.

Much more efficient than traditional architecture;

That is so bad-ass. My favorite technical specs are those vague, "eat me" specs to previous owners of the technology.

Bulldung I say! Another fanboy ran with the ball trying to score in an endzone (for americans) or a goal (for the rest of us) :P

New AA mode: VCAA
What? Sound interesting.

384-bit memory interface (256-bit+128-bit)
768MB memory size (512MB+256MB)
Should go with a 512-bit interface and keep the 512 MB RAM I think.

Support FP16 HDR+MSAA;
But with that VCAA + HDR, if it is supported, maybe 512 RAM isn't enough?

Unified Shader Architecture
Shader peformance: 2x Pixel/12x Vertex over G71
Even with USA, hahaha, the G80s should have 2 times the pixel and 12 times the vertex power? Am I read this right? That is insane.....

hey look, FP16's back.

I'm eager to see the new mainstream card, which may not show up till spring. But the lineage has been GF4200 -> R9600 -> GF6600 ->GF7600. They have each been particularly impressive, especially the latter two. Bring on the GF8600 / R26K-XTXYZTX

"Unified Shader Architecture"

Didn't even read the remaining but IIRC Unified Shaders is the feature ATI is boasting about for R600 and NVIDIA is playing down for G80 because they won't switch to it just now..

*deleted* by myself... sometimes I'm just too lazy

700M transistors??!!! Time to upgrade your power supplies, boys and girls.