G80 De-Mystified; GeForce 8800GTX/GT

[retsam]

Don't worry - The P4 is going to reach 10ghz any day now, too.
hahahah....:D

[walterman]

Well yea, 2.5GHz actual clockrate is a little high I think.

Especially since the GDDR4 memory finally has passing yeilds of 1GHz, 2.5GHz seems a little unlikely.

Exactly, but we all know that the real change is the new memory bus width.

:jumping:

[SH64]

Exactly, but we all know that the real change is the new memory bus width.

:jumping:
Yeah perhaps wider bus & less clock rate. (like 1.6Ghz-2Ghz)
that will still result in a huge memory bandwidth.

[shabby]

I managed to save the slideshows :)

http://img82.imageshack.us/img82/8248/g801tb3.jpg
http://img82.imageshack.us/img82/5425/g802yf2.jpg
http://img82.imageshack.us/img82/285/g803pu7.jpg

[retsam]

so maybe just maybe the g80 is a gpgpu?

[jAkUp]

yipes. well if that is true and the R600 is true, looks like the G80 will have higher memory bandwidth.

[a12ctic]

Don't worry - The P4 is going to reach 10ghz any day now, too.
dude it already has, THE GOVERMENT IS KEEPING INTEL FROM RELEASING IT TO THE PUBLIC!!!

*puts on tin foil hat*

[Redeemed]

I don't like that whole water cooling idea for the GTX. I do hope that it is optional.

Regardless, I'll probably be going with two 8800GTs anyways.

[nekrosoft13]

yipes. well if that is true and the R600 is true, looks like the G80 will have higher memory bandwidth.

when will evga be getiing their first samples?

[lee63]

when will evga be getiing their first samples?I dont think he be able to tell you that, even if he knew(pirate)

[nekrosoft13]

he probably has them in his rig already

[SH64]

I managed to save the slideshows :)

Thanks! didnt know they came in slides.

[PeterJensen]

Club3d have samples....

[havox]

Rumors = BS

[chipguy]

not at these densities there not getting up to 800 mhz's. you have to remember that the 80nm node is just a shrink and there is not lowk like there is at 90nm.

At these densities, you can get easily get to 2 GHz (see initial P4's running a 2GHz in 180nm).

It has nothing to do with better fabs or higher quality processes, but all with how much time and manpower you're willing to invest. It's really that simple. What's more, the smaller your process nodes, the harder it becomes to reach high frequencies with non-custom design techniques. For custom design, the effort remains more or less the same.

[retsam]

At these densities, you can get easily get to 2 GHz (see initial P4's running a 2GHz in 180nm).

It has nothing to do with better fabs or higher quality processes, but all with how much time and manpower you're willing to invest. It's really that simple. What's more, the smaller your process nodes, the harder it becomes to reach high frequencies with non-custom design techniques. For custom design, the effort remains more or less the same.
90nm node has a theretical speed of 800-900 MHZ's with low-k dialectrics at 230 million trannies(beyond3d had a discussion about it). so your telling me that 80nm without low-k with 500 million trannies has double that. compairing the p-4 which half of its die size is cache( isnt cache memory auto routed?) with a gpu asic is like comparing apples to oranges. the p-4 only had 42 million trannies.


ftp://download.intel.com/pressroom/images/events/moores_law_40th/Microprocessor_Chart.jpg

[lowdog]

Club3d have samples....


Off ???

[PeterJensen]

Off ???

Take a wild guess :)

[Fotis]

Take a wild guess :)
R600?:D :p

[chipguy]

90nm node has a theretical speed of 800-900 MHZ's with low-k dialectrics at 230 million trannies(beyond3d had a discussion about it).
After 15 years of chip design experience, I'm confident to say: BS.

Please provide me some physics constant that limits the speed to 900 MHz.
The only real theoretical constraint is the switching speed of a transistors which is somewhere around 200 GHz in 90nm and the propagation signal time, which depends somewhat on the dielectric (though not much), but mostly just on the RC of your metal wires.

so your telling me that 80nm without low-k with 500 million trannies has double that. compairing the p-4 which half of its die size is cache( isnt cache memory auto routed?) with a gpu asic is like comparing apples to oranges. the p-4 only had 42 million trannies.

The P4 in 90nm goes up to, what, 4 GHz? The amount of transistors is irrelevant: whether you have 100 or 700 million of them doesn't change the laws of physics. It only changes the cumulative power consumption of your die, whichs makes cooling harder. But as long are you're able to keep the junction temperature (not: package temperature!) of your transistors below 125C, you're good to go.

Let me repeat my earlier statement: the maximum speed depends on the amount of time you can spend. 95% of the semiconductor industry uses standard cell design, which are small prebuild logic blocks that make it easy to automate tools and crank out design (fairly) quickly. Only if you want the absolute maximum speed, once can go to custom placement of standard cells, which can improve your speed by, say, 100% over your non-custom placed design, or so-called full custom design, which mean you design each and every transistor individually. Intel and AMD can afford the time/manpower and money to do the latter. Most others don't.
That all there is to it.

Most arguments at beyond3d are made by better informed people than, say, here :), but that doesn't mean everything they say is gospel.

[PeterJensen]

R600?:D :p

No ffs :) This thread is about the G80...

[nemecb]

Most arguments at beyond3d are made by better informed people than, say, here :), but that doesn't mean everything they say is gospel.
Heresy! There are no better informed people than here!:D

[NoWayDude]

Michael Hara continued to explain that nVidia about it think, the new „Unified Shader “- architecture, in which pixel and Vertex Shader are summarized, to use also for physics computations. By the unification of these two computation units it is very simple to compute also physics within the GPU. With nVidia one believes firmly in this concept and sees also opposite processors, which will have four cores, an achievement advantage.;)

http://www.computerbase.de/news/hard...eschleunigung/

Thank you geo @ B3D

http://www.beyond3d.com/forum/showthread.php?t=33576&page=21

[slaWter]

;)

http://www.computerbase.de/news/hard...eschleunigung/

Thank you geo @ B3D

http://www.beyond3d.com/forum/showthread.php?t=33576&page=21

http://www.nvnews.net/vbulletin/showpost.php?p=996067&postcount=44

^^