Geforce 8800 GTX und GTS Preview

[AthlonXP1800]

PC WELT has 7 pages of G80 preview, it in german:

http://www.pcwelt.de/news/hardware/video/61081/index.html

G80 spec is much the same as other sources, 700 million transistors are confirmed.

Here the slight translated spec sheet from german:

Geforce 8800 GTX

Code name: G80
700 million transistors
90nm process
Chip clock: 575MHz
128 Streaming processors
Streaming processors clock: 1350MHz
Theoretical pixel filling rate: 36800 MPix/s
768MB GDDR3
12 memory chips
900MHz memory clock
384 bits Memory interface
86.4GB Memory bandwidth
Shader model 4.0
DirectX version 10
OpenGL version 2.0
SLI capable

Street price: 650 euro

Geforce 8800 GTS

Code name: G80
700 million transistors
90nm process
Chip clock: 500MHz
96 Streaming processors
Streaming processors clock: 1200MHz
Theoretical pixel filling rate: 24000MPix/s
640MB GDDR3
10 memory chips
800MHz memory clock speed
320 bits Memory interface
64GB Memory bandwidth
Shader model 4.0
DirectX version 10
OpenGL version 2.0
SLI capable

Street price: 500 euro

[|MaguS|]

WoW, What kind of preview was that? No pictures or benchmarks... All the information given was basically the same **** thats been floating around lately.

[agentkay]

GTX cooling is surprisingly quiet.
GTX needs at least a 450w PSU with 30amps at 12v (apparently it uses 200w under full load)
GTX length is 26cm
GTX weight is 800grams

[SH64]

GTX cooling is surprisingly quiet.

Did they mention anything about water/air hybrid cooling ?

[AthlonXP1800]

I noticed Nvidia changed the "Shader" phrase to new word "Streaming"? Here another word "Gigathread Technology" I think originally from "Gigapixel Technology" :D

[agentkay]

Did they mention anything about water/air hybrid cooling ?

No, they didn´t mention any hybrid cooling and I´m pretty sure they had mentioned it if it came with one. :)

Edit: I quickly read the article again, GTX and GTS are both air cooled.

[SH64]

No, they didn´t mention any hybrid cooling and I´m pretty sure they had mentioned it if it came with one. :)
Man i wish its just regular air cooling at least for the refrence 8800GTX (dont care about overclocked versions) .. the idea of water-cooling involved is giving me nightmares :|

[Zelda_fan]

Man i wish its just regular air cooling at least for the refrence 8800GTX (dont care about overclocked versions) .. the idea of water-cooling involved is giving me nightmares :|

I hope it doesn't ship standard with water cooling. Not because I'm afraid to water cool, but because that will mean that the thing runs HOT! Way, way to hot for a reference part. I do plan on water cooling it anyway, but if it temps at 70C with water cooling, that is way way too hot.

[Morrow]

The thing which is interesting in this German preview is again the fact they say that the G80 architecture is unified shader architecture where thou every hint till now said the contrary.

[SpiffMistroII]

The thing which is interesting in this German preview is again the fact they say that the G80 architecture is unified shader architecture where thou every hint till now said the contrary.



This German Artical also states that G80 is 30% faster than R580+. And the GTS is the same speed. :thumbdwn:

[fivefeet8]

This German Artical also states that G80 is 30% faster than R580+. And the GTS is the same speed. :thumbdwn:

In dx9 with pre-release drivers. (pirate)

[Redeemed]

The thing which is interesting in this German preview is again the fact they say that the G80 architecture is unified shader architecture where thou every hint till now said the contrary.

Dude, where have you been- if you don't mind my asking?

All released specs of the G80 have had unified shaders, now being called "streaming shaders" I believe.

You were aware that G80 will have 128 unified shaders, right? Or did you somehow miss the specs of G80 completely?

[Razor1]

I noticed Nvidia changed the "Shader" phrase to new word "Streaming"? Here another word "Gigathread Technology" I think originally from "Gigapixel Technology" :D


gigathreaded cause it can do a sh*t load of threads at the same time :)

[Morrow]

You were aware that G80 will have 128 unified shaders, right? Or did you somehow miss the specs of G80 completely?

Yeah, and those specs you are talking about still have not been confirmed (--> rumor)

I recall some nvidia marketing spokesman saying a few months ago that their next generation (G80) would NOT have a unified architecture and that they would continue using a dedicated architecture for now.

[SpiffMistroII]

Yeah, and those specs you are talking about still have not been confirmed (--> rumor)

I recall some nvidia marketing spokesman saying a few months ago that their next generation (G80) would NOT have a unified architecture and that they would continue using a dedicated architecture for now.



I also find it funny that tech sites are starting to say "streaming" instead of "Unifide". I too remember Nvidia spokes person saying that G80 will not be Unifide, and that it has no advantage over triditional arc.

hmmmm....


700m also sound way out of hand. Since it's still on a 90nm die.

[Redeemed]

nVidia never claimed that they would NOT be using USA. I know the article you speak of.

I have to leave for work now, but tomorrow morning when I get back I'll see if I can find it and show that they never once claimed that G80 will not use USA.

[Redeemed]

I also find it funny that tech sites are starting to say "streaming" instead of "Unifide". I too remember Nvidia spokes person saying that G80 will not be Unifide, and that it has no advantage over triditional arc.

hmmmm....


700m also sound way out of hand. Since it's still on a 90nm die.

If you include a 384 bit memory bus, 128 "streaming pipes", and a Quantum Effects Physics Processor, 700 million doesn't sound too unrealistic.

[SpiffMistroII]

If you include a 384 bit memory bus, 128 "streaming pipes", and a Quantum Effects Physics Processor, 700 million doesn't sound too unrealistic.


The bus has nothing to do with the trannsitor count.

128 "Streaming ALU's" still sounds funny on how it's not being called unifide anymore. Weird how this happens right after editors day.

Also that quantium PPU is not a PU in it self. It's a layer with in G80 that dose Physics calculations. It runs off of the same trannsitors that G80 runs off. In other words, it share the resources that G80 already has.

[Redeemed]

The bus has nothing to do with the trannsitor count.

128 "Streaming ALU's" still sounds funny on how it's not being called unifide anymore. Weird how this happens right after editors day.

Also that quantium PPU is not a PU in it self. It's a layer with in G80 that dose Physics calculations. It runs off of the same trannsitors that G80 runs off. In other words, it share the resources that G80 already has.

In regards to the text in bold: do you have any means to prove this? How do you know that there isn't a PPU on the same die as the GPU? Kind of a "dual-core" setup, though they'd be completely different cores? Or maybe they're completely seperate dies? Can you prove the contrary? Afterall, it is name the "Quantum Effects Physics Processor". Now, if it wasn't a processing unit unto itself, then nVidia chose a very misleading name for it. But, due to the high count of transistors, what uses up all 700 million? Just the GPU? How and why? is it all those pipes? No, I find it quite easy to believe there is a second processing unit included, and it is the Quantum Effects Physics Processor.

As to the article where nVidia talked about USA:

"We will do a unified architecture in hardware when it makes sense. When it's possible to make the hardware work faster unified, then of course we will. It will be easier to build in the future, but for the meantime, there's plenty of mileage left in this architecture."


http://www.bit-tech.net/bits/2005/07/11/nvidia_rsx_interview/4.html

That quote gives the impression that G80 would not have a unified shader architecture, but it does not downright state that it wont.

I'm thinking nVidia had some strong marketing going on.

Look at this from the marketing perspective:

Would nVidia really want to allow ATi to have a card that sported USA, a very popular (albeit not necessary) feature supported by DX10? No. Instead, my opinion is that nVidia led everybody to believe specific things about G80, which weren't entirely true. After all, this would be great to keep people in the dark until release day. Which is exactly what nVidia has been trying to do.

By now, with any previous GPU, there'd have been slides and such leaked all over the net. Not with G80. We have a few sites (such as DailyTech) claiming to have review samples and supplying some information, but that is it and it isn't concrete yet. nVidia is keeping G80 very secret. And it'd be a big slap to ATi if nVidia managed to release G80 with the currently "rumored" specs.

http://www.nvnews.net/vbulletin/showpost.php?p=1009625&postcount=1

There are the supposed specs of G80, and a link to the DT article regarding it. With g80 wieghing in at 700 million transistors, I find the specs quite beleivable. And if they do turn out accurate, then ATi better make R600 uber-kick@$$ and it better be a hard launch.

I'm sure ATi can pull it off. ;)

[Morrow]

Would nVidia really want to allow ATi to have a card that sported USA, a very popular (albeit not necessary) feature supported by DX10?

USA is really not a feature :) It's not that it offers better or more complex effects, it's just a way the rendering is done internally and really not interesting to the end-consumer. In fact, I have no doubt the nvidia's marketing department could make USA look worse than a dedicated architecture if they really wanted to.

USA or not, it's really meaningless this generation.


X-bit labs just released an article (http://www.xbitlabs.com/news/video/display/20061023142704.html) in which they express their doubt that anyone not under NDA really knows what the G80 is going to be:

Specifications of the G80 chip are not clear. Some sources indicate that Nvidia’s first DirectX 10 chip will incorporate 48 pixel shader processors and an unknown number of vertex shader/geometry shader processors. Other, however, claims that the G80 has 32 pixel and 16 vertex and geometry shader processors. Yet another source has indicated that the G80 will have unified shader architecture and will consist of 700 million transistors.

So, if the G80 sports a USA or not, still needs to proven/confirmed.

[SpiffMistroII]

In regards to the text in bold: do you have any means to prove this? How do you know that there isn't a PPU on the same die as the GPU? Kind of a "dual-core" setup, though they'd be completely different cores? Or maybe they're completely seperate dies? Can you prove the contrary? Afterall, it is name the "Quantum Effects Physics Processor". Now, if it wasn't a processing unit unto itself, then nVidia chose a very misleading name for it. But, due to the high count of transistors, what uses up all 700 million? Just the GPU? How and why? is it all those pipes? No, I find it quite easy to believe there is a second processing unit included, and it is the Quantum Effects Physics Processor.


"IF" thier was a ppu in the same die as G80, that actully bad for G80.

1) no games support QEP.

2) the specs on G80 would be very misleading and will be a marketing ploy.

why? Becuase that means those 700m tranns, are not all G80's. Those 128 ALU's would not be all G80's, so on and so forth.

[Redeemed]

"IF" thier was a ppu in the same die as G80, that actully bad for G80.

1) no games support QEP.

2) the specs on G80 would be very misleading and will be a marketing ploy.

why? Becuase that means those 700m tranns, are not all G80's. Those 128 ALU's would not be all G80's, so on and so forth.

I still don't see how that is bad. Even if G80 doesn't have all that to itself, it'd probably get most of it. Are you really going to complain about having only 96 shaders? I wouldn't. :p

Then again, you could be right, and I guess that would make sense as well in regards to Havok and such.

We'll all find out before too much longer. :)

[SpiffMistroII]

I still don't see how that is bad. Even if G80 doesn't have all that to itself, it'd probably get most of it. Are you really going to complain about having only 96 shaders? I wouldn't. :p

Then again, you could be right, and I guess that would make sense as well in regards to Havok and such.

We'll all find out before too much longer. :)


by no means am I down playing G80 though. :o

It's going to perform great for sure. It dose indeed have a seperate chip on board, but not for physics.

I'm warning you though, R600 is going to be monster. That's all I can say about it for right now. I will say their are plenty of hints out there, so you can put the puzzle together, at most 25% of the way.

[Xion X2]

R600 won't be much stronger than G80, if at all. Just look at the trend over the last year or two. The cards are always neck and neck.

[SpiffMistroII]

R600 won't be much stronger than G80, if at all. Just look at the trend over the last year or two. The cards are always neck and neck.



my lips are shut.:)