NVIDIA GeForce 7900GT Pics, Specs

[Kalessian]

Originally posted by: Acanthus

Originally posted by: Steelski

Originally posted by: phantom404
So I guess the gtx will have 32

No. Look at the size of the GPU. it should be at least the size of the original G70 if it were to have 32 pipes. its just too small.

Here are the links for a great size comparison to show this is only a die shrink !!!!!!!!!!!!!!!

G70

G71

And dont anyone say that the GTX will have more as that is the G71 core aswell. they would have another name for it if it was different.

You do realize that going from 130nm to 90nm would be a 100% increase in transistor count for the same surface area? Even if it is smaller you could cram more transistors into that space.

Oh, never thought of that. You and your fancy math.

There may be hope that it's a great card, yet.

 

[Steelski]

Originally posted by: Acanthus

Originally posted by: Steelski

Originally posted by: phantom404
So I guess the gtx will have 32

No. Look at the size of the GPU. it should be at least the size of the original G70 if it were to have 32 pipes. its just too small.

Here are the links for a great size comparison to show this is only a die shrink !!!!!!!!!!!!!!!

G70

G71

And dont anyone say that the GTX will have more as that is the G71 core aswell. they would have another name for it if it was different.

You do realize that going from 130nm to 90nm would be a 100% increase in transistor count for the same surface area? Even if it is smaller you could cram more transistors into that space.

Edit: youre only talking about a ~25% increase in transistor count to go from 24 to 32 pipes. So you could still make the core substantially smaller and still increase the number of pipelines.

its a shame your fancy math does not add up as the G70 process was not 130nm........

DING DING DING DIN>>>>>>>>>ITS 110nm

And this is quoted now to show you how wrong you are .....
DEEP voice "FOREVER"

 

[schtuga]

Originally posted by: Acanthus

Originally posted by: Steelski

Originally posted by: phantom404
So I guess the gtx will have 32

No. Look at the size of the GPU. it should be at least the size of the original G70 if it were to have 32 pipes. its just too small.

Here are the links for a great size comparison to show this is only a die shrink !!!!!!!!!!!!!!!

G70

G71

And dont anyone say that the GTX will have more as that is the G71 core aswell. they would have another name for it if it was different.

You do realize that going from 130nm to 90nm would be a 100% increase in transistor count for the same surface area? Even if it is smaller you could cram more transistors into that space.

Edit: youre only talking about a ~25% increase in transistor count to go from 24 to 32 pipes. So you could still make the core substantially smaller and still increase the number of pipelines.

Actually on several other sites,they are stating that it is a 78/79 pcb but that is the 7600 chip.If you compare the 7600 on vr-zone and the 7900 on dailytech,they are identical,not possible for a 12 pipe and 24 pipe to be the same size.(at least that is what is being stated.)

Metro from chilehardware still states the gtx will be 32 pipes.who knows who to believe with so many contradictory statements being made all of a sudden.

 

[Steelski]

Originally posted by: schtuga

Originally posted by: Acanthus

Originally posted by: Steelski

Originally posted by: phantom404
So I guess the gtx will have 32

No. Look at the size of the GPU. it should be at least the size of the original G70 if it were to have 32 pipes. its just too small.

Here are the links for a great size comparison to show this is only a die shrink !!!!!!!!!!!!!!!

G70

G71

And dont anyone say that the GTX will have more as that is the G71 core aswell. they would have another name for it if it was different.

You do realize that going from 130nm to 90nm would be a 100% increase in transistor count for the same surface area? Even if it is smaller you could cram more transistors into that space.

Edit: youre only talking about a ~25% increase in transistor count to go from 24 to 32 pipes. So you could still make the core substantially smaller and still increase the number of pipelines.

Actually on several other sites,they are stating that it is a 78/79 pcb but that is the 7600 chip.If you compare the 7600 on vr-zone and the 7900 on dailytech,they are identical,not possible for a 12 pipe and 24 pipe to be the same size.(at least that is what is being stated.)

Metro from chilehardware still states the gtx will be 32 pipes.who knows who to believe with so many contradictory statements being made all of a sudden.

please provide a link. But I got the picture from the Daily tech website which also has a big sighn saying 7900GT GPU underneath it...and they also have a 7600 but with no core pics. I'm sure that they can tell the difference between the cards in question.

http://www.dailytech.com/article.aspx?newsid=908
there is a link in the link to the 7900 GT aswell.
I would imagine that a 12 pipecore will be even smaller

 

[ItSells]

Originally posted by: phantom404
DailyTech

7900 GT
* PCIe native
* 450MHz core frequency
* 256-bit memory interface
* 42.2GB/sec. memory bandwidth
* 10.8B pixels/sec. fill rate
* 940M vertices/sec.
* 24 pixels per cycle
* Built in dual-link DVI support for 2560x1600 resolution

7800 GTX
- Core clock: 430MHz
- DirectX: DirectX 9
- DVI: 2
- Memory Clock: 1200MHz
- Memory Interface: 256-bit
- OpenGL: OpenGL 2.0
- PixelPipelines: 24
- Tuner: None
- TV-Out: S-Video Out
- VIVO: Yes

Difference:
7900 GT / 7800 GTX

* 450 / 430 MHz core frequency
* 42.2 / 38.4 GB/s memory bandwidth
* 10.8 / 10.32 B pixels/sec. fill rate
* 940 / 860 M vertices/sec.
* 1.32 / 1.2 GHz 256MB GDDR3 memory

Thanks to darXoul


Just FYI

the 7900 GT only makes so much sense until the x1900xt/xL 256mb comes out with a similar price

 

[Acanthus]

Originally posted by: Steelski

Originally posted by: Acanthus

Originally posted by: Steelski

Originally posted by: phantom404
So I guess the gtx will have 32

No. Look at the size of the GPU. it should be at least the size of the original G70 if it were to have 32 pipes. its just too small.

Here are the links for a great size comparison to show this is only a die shrink !!!!!!!!!!!!!!!

G70

G71

And dont anyone say that the GTX will have more as that is the G71 core aswell. they would have another name for it if it was different.

You do realize that going from 130nm to 90nm would be a 100% increase in transistor count for the same surface area? Even if it is smaller you could cram more transistors into that space.

Edit: youre only talking about a ~25% increase in transistor count to go from 24 to 32 pipes. So you could still make the core substantially smaller and still increase the number of pipelines.

its a shame your fancy math does not add up as the G70 process was not 130nm........

DING DING DING DIN>>>>>>>>>ITS 110nm

And this is quoted now to show you how wrong you are .....
DEEP voice "FOREVER"

Ok... so that would mean youd still have 66% more space with the same core size... so my argument would still stand.

You could still make a smaller GPU with 32 pipelines going from 110nm to 90nm.

 

[M0RPH]

Originally posted by: Acanthus

Ok... so that would mean youd still have 66% more space with the same core size... so my argument would still stand.

You could still make a smaller GPU with 32 pipelines going from 110nm to 90nm.

Nah, your argument is still poor. If you do a measurement on those core pics you'll find the die area of the G71 to be roughly 70% that of the G70. Using your own math, there's obviously not enough space there for a 25% increased transistor count that 32 pipes would require.

 

[Steelski]

Originally posted by: Acanthus

Originally posted by: Steelski

Originally posted by: Acanthus

Originally posted by: Steelski

Originally posted by: phantom404
So I guess the gtx will have 32

No. Look at the size of the GPU. it should be at least the size of the original G70 if it were to have 32 pipes. its just too small.

Here are the links for a great size comparison to show this is only a die shrink !!!!!!!!!!!!!!!

G70

G71

And dont anyone say that the GTX will have more as that is the G71 core aswell. they would have another name for it if it was different.

You do realize that going from 130nm to 90nm would be a 100% increase in transistor count for the same surface area? Even if it is smaller you could cram more transistors into that space.

Edit: youre only talking about a ~25% increase in transistor count to go from 24 to 32 pipes. So you could still make the core substantially smaller and still increase the number of pipelines.

its a shame your fancy math does not add up as the G70 process was not 130nm........

DING DING DING DIN>>>>>>>>>ITS 110nm

And this is quoted now to show you how wrong you are .....
DEEP voice "FOREVER"

Ok... so that would mean youd still have 66% more space with the same core size... so my argument would still stand.

You could still make a smaller GPU with 32 pipelines going from 110nm to 90nm.

NO!!!!!!!!!!!!!!!!!!!!!! there is not 66% more space. its more like 33% at the most.
please do your math right. each full step there is a 50% size reduction. as that would be a half step then it is a 25% reduction (this is all rough)
thats a core size of 75% of the original. (if this smaller core is seen 100% then you would need 33.3% extra to make it the same size as the original core.
as we both see that this core is smaller we can assume that the 33% pipe increase has not been implemented.

And yes it is a 33% increase as 8 pipes from 24 do not equal 25%.....they are 33.3% of the 24, therefore having an 8 pipe increase is not a 25% increase from 24 pipes..(25% is what the overall percent is from 32 pipes.....not 24(by your calculations a 16 pipe card has 50% more pipes than an 8 piped card)

Simple as that.

 

[ElFenix]

this thing had better be priced under $400 street. maybe even lower. ati has an x1900xl somewhere in there.

 

[crazydingo]

Originally posted by: ElFenix
this thing had better be priced under $400 street. maybe even lower. ati has an x1900xl somewhere in there.

March 23.

 

[Matt2]

The fanboism is oozing from this thread.

 

[M0RPH]

Originally posted by: Matt2
The fanboism is oozing from this thread.

By calling people fanboys, you're just adding fuel to the fire. This forum would be so better off if everyone just stopped using that word altogether. I stopped using it long ago.

 

[Steelski]

Originally posted by: M0RPH

Originally posted by: Matt2
The fanboism is oozing from this thread.

By calling people fanboys, you're just adding fuel to the fire. This forum would be so better off if everyone just stopped using that word altogether. I stopped using it long ago.

If he was refering to me then i dont think it was even warrented.
I just get steamed when people dont really accept facts and make things up.

By quoting morth aswell it makes me feel like i'm protecting myself like someone hiding behind their dad...... it wasent what i originally intended, but thats the feeling i got just before i was gonna post..

hehehe

 

[Acanthus]

Originally posted by: Steelski

Originally posted by: Acanthus

Originally posted by: Steelski

Originally posted by: Acanthus

Originally posted by: Steelski

Originally posted by: phantom404
So I guess the gtx will have 32

No. Look at the size of the GPU. it should be at least the size of the original G70 if it were to have 32 pipes. its just too small.

Here are the links for a great size comparison to show this is only a die shrink !!!!!!!!!!!!!!!

G70

G71

And dont anyone say that the GTX will have more as that is the G71 core aswell. they would have another name for it if it was different.

You do realize that going from 130nm to 90nm would be a 100% increase in transistor count for the same surface area? Even if it is smaller you could cram more transistors into that space.

Edit: youre only talking about a ~25% increase in transistor count to go from 24 to 32 pipes. So you could still make the core substantially smaller and still increase the number of pipelines.

its a shame your fancy math does not add up as the G70 process was not 130nm........

DING DING DING DIN>>>>>>>>>ITS 110nm

And this is quoted now to show you how wrong you are .....
DEEP voice "FOREVER"

Ok... so that would mean youd still have 66% more space with the same core size... so my argument would still stand.

You could still make a smaller GPU with 32 pipelines going from 110nm to 90nm.

NO!!!!!!!!!!!!!!!!!!!!!! there is not 66% more space. its more like 33% at the most.
please do your math right. each full step there is a 50% size reduction. as that would be a half step then it is a 25% reduction (this is all rough)
thats a core size of 75% of the original. (if this smaller core is seen 100% then you would need 33.3% extra to make it the same size as the original core.
as we both see that this core is smaller we can assume that the 33% pipe increase has not been implemented.

And yes it is a 33% increase as 8 pipes from 24 do not equal 25%.....they are 33.3% of the 24, therefore having an 8 pipe increase is not a 25% increase from 24 pipes..(25% is what the overall percent is from 32 pipes.....not 24(by your calculations a 16 pipe card has 50% more pipes than an 8 piped card)

Simple as that.

pipes are the only thing in a gpu now? Yeah, theres no logic, or even a friggin video decoder in there, no caches, no memory controllers (4 of em), nothing but pipes right?

And yeah youre right, 110nm to 90nm is a ~34% decrease.

 

[coldpower27]

Originally posted by: Acanthus

pipes are the only thing in a gpu now? Yeah, theres no logic, or even a friggin video decoder in there, no caches, no memory controllers (4 of em), nothing but pipes right?

And yeah youre right, 110nm to 90nm is a ~34% decrease.

Interesting I just did the math now, and surprisingly the die size of G71 would be ~ 224mm2 by my calculations, if the transistor density had increased in a linear fashion from 110nm to 90nm with an increase to 149.14% assuming 110nm is 100%.

A die size of 224mm2 would be in the range of not requiring a silver frame of the the package which is what the picture of the G71 die currently, shows but if that is the G71 GPU imagine the die size of the G73 GPU on 90nm, it could easily be 60% as it is basically half of G71, however something will still stay constant like the video logic, so 60% is a reasonable estimate.

I think it's obvious Nvidia focused on attainable yeilds and good margins, on product rather then on pure speed, I don't believe this card will overtake X1900 XTX unfortunately, an 18% increase in Pixel Shader Power over 7800 GTX 512, is not enough to win as there already some situation where the X1900 XTX beats the 7800 GXT 512 by margins of more then 20%. It will maintain the OpenGL crown though as the 7800 GTX 512 already beats the X1900 XTX in that though.

I think G80 is when Nvidia will try to go for the juggular in performance again, this time around better margin then expensive 315mm2 R580 die is what Nvidia is focusing on.

 

[Acanthus]

Originally posted by: coldpower27

Originally posted by: Acanthus

pipes are the only thing in a gpu now? Yeah, theres no logic, or even a friggin video decoder in there, no caches, no memory controllers (4 of em), nothing but pipes right?

And yeah youre right, 110nm to 90nm is a ~34% decrease.

Interesting I just did the math now, and surprisingly the die size of G71 would be ~ 224mm2 by my calculations, if the transistor density had increased in a linear fashion from 110nm to 90nm with an increase to 149.14% assuming 110nm is 100%.

A die size of 224mm2 would be in the range of not requiring a silver frame of the the package which is what the picture of the G71 die currently, shows but if that is the G71 GPU imagine the die size of the G73 GPU on 90nm, it could easily be 60% as it is basically half of G71, however something will still stay constant like the video logic, so 60% is a reasonable estimate.

I think it's obvious Nvidia focused on attainable yeilds and good margins, on product rather then on pure speed, I don't believe this card will overtake X1900 XTX unfortunately, an 18% increase in Pixel Shader Power over 7800 GTX 512, is not enough to win as there already some situation where the X1900 XTX beats the 7800 GXT 512 by margins of more then 20%. It will maintain the OpenGL crown though as the 7800 GTX 512 already beats the X1900 XTX in that though.

I think G80 is when Nvidia will try to go for the juggular in performance again, this time around better margin then expensive 315mm2 R580 die is what Nvidia is focusing on.

What math? Nvidia doesnt tell us exactly how much space everything takes up, not even rough estimates. I was just saying that it could fit in theory in a smaller core. Nothing even close to exact.

 

[coldpower27]

Originally posted by: Acanthus
What math? Nvidia doesnt tell us exactly how much space everything takes up, not even rough estimates. I was just saying that it could fit in theory in a smaller core. Nothing even close to exact.

I don't need to though that really, the die size of G70 itself is known, which is 334mm2 for 302 Million Transistors, on 110nm which works out to a transistor density, 904K gates/mm2, not going from 110nm to 90nm is a 49.14% increase in transistor density.

110^2 /90^2 = 1.4914 So you increasing your transistor density per mm2 to 149.14%

So if it is indeed a straight shrink, 1348K gates/mm2 is the new density on 90nm, take the 302 Million divide by new transistor density and walla 224 mm2 die size for an optical shrink to 90nm.

 

[Acanthus]

Originally posted by: coldpower27

Originally posted by: Acanthus
What math? Nvidia doesnt tell us exactly how much space everything takes up, not even rough estimates. I was just saying that it could fit in theory in a smaller core. Nothing even close to exact.

I don't need to though that really, the die size of G70 itself is known, which is 334mm2 for 302 Million Transistors, on 110nm which works out to a transistor density, 904K gates/mm2, not going from 110nm to 90nm is a 49.14% increase in transistor density.

110^2 /90^2 = 1.4914 So you increasing your transistor density per mm2 to 149.14%

So if it is indeed a straight shrink, 1348K gates/mm2 is the new density on 90nm, take the 302 Million divide by new transistor density and walla 224 mm2 die size for an optical shrink to 90nm.

Ok, so in short you determined a die shrink can make the die smaller :roll:

 

[pcmodem]

Originally posted by: phantom404
DailyTech
Difference:
7900 GT / 7800 GTX

* 450 / 430 MHz core frequency
* 42.2 / 38.4 GB/s memory bandwidth
* 10.8 / 10.32 B pixels/sec. fill rate
* 940 / 860 M vertices/sec.
* 1.32 / 1.2 GHz 256MB GDDR3 memory

I'm loving this...

A single slot solution that's likely 5% to 10% faster than a 7800GTX, depending on the individual 3rd party manufacturer's spec's and OC'ing.

60 FPS on AoE III sounds feasible.


Cheers,
PCM

 

[coldpower27]

Originally posted by: Acanthus
Ok, so in short you determined a die shrink can make the die smaller :roll:

What I determined is that if it is indeed a simple die shrink the cost advantage will be in Nvidia's court for sure as 224mm2 will yield much better then the R580 cores 315mm2.

 

[Avalon]

Originally posted by: Acanthus

Originally posted by: Steelski

Originally posted by: Acanthus

Originally posted by: Steelski

Originally posted by: Acanthus

Originally posted by: Steelski

Originally posted by: phantom404
So I guess the gtx will have 32

No. Look at the size of the GPU. it should be at least the size of the original G70 if it were to have 32 pipes. its just too small.

Here are the links for a great size comparison to show this is only a die shrink !!!!!!!!!!!!!!!

G70

G71

And dont anyone say that the GTX will have more as that is the G71 core aswell. they would have another name for it if it was different.

You do realize that going from 130nm to 90nm would be a 100% increase in transistor count for the same surface area? Even if it is smaller you could cram more transistors into that space.

Edit: youre only talking about a ~25% increase in transistor count to go from 24 to 32 pipes. So you could still make the core substantially smaller and still increase the number of pipelines.

its a shame your fancy math does not add up as the G70 process was not 130nm........

DING DING DING DIN>>>>>>>>>ITS 110nm

And this is quoted now to show you how wrong you are .....
DEEP voice "FOREVER"

Ok... so that would mean youd still have 66% more space with the same core size... so my argument would still stand.

You could still make a smaller GPU with 32 pipelines going from 110nm to 90nm.

NO!!!!!!!!!!!!!!!!!!!!!! there is not 66% more space. its more like 33% at the most.
please do your math right. each full step there is a 50% size reduction. as that would be a half step then it is a 25% reduction (this is all rough)
thats a core size of 75% of the original. (if this smaller core is seen 100% then you would need 33.3% extra to make it the same size as the original core.
as we both see that this core is smaller we can assume that the 33% pipe increase has not been implemented.

And yes it is a 33% increase as 8 pipes from 24 do not equal 25%.....they are 33.3% of the 24, therefore having an 8 pipe increase is not a 25% increase from 24 pipes..(25% is what the overall percent is from 32 pipes.....not 24(by your calculations a 16 pipe card has 50% more pipes than an 8 piped card)

Simple as that.

pipes are the only thing in a gpu now? Yeah, theres no logic, or even a friggin video decoder in there, no caches, no memory controllers (4 of em), nothing but pipes right?

And yeah youre right, 110nm to 90nm is a ~34% decrease.

Acanthus, if Steelski's math is right and going from 110nm to 90nm is a 34% decrease, then I don't see how adding 8 pipes would be feasible, as it would take up the majority of the free space earned from the die shrink. Plus, other things they need to add to the card. I'm not sure, I could be wrong. Thoughts?

 

[coldpower27]

Originally posted by: Avalon
Acanthus, if Steelski's math is right and going from 110nm to 90nm is a 34% decrease, then I don't see how adding 8 pipes would be feasible, as it would take up the majority of the free space earned from the die shrink. Plus, other things they need to add to the card. I'm not sure, I could be wrong. Thoughts?

Well you got to think about it how much would adding 8 Pipelines cost....

Adding 8 Pipelines from NV40 to G70 was 80 Million Transistors, but the pipelines were also enhanced as well, also 2 more vertex units were added, enahnced buffer for 20x15 operation, if we be generous and give that amount for 8 pipeline that would bring the total to 382 which ~ the same level as R580 so it is possible to produce for Nvidia, if ATI can rpoduce a 384 Million Transistor GPU on 90nm why can't NV do the same?

 

[Regs]

Originally posted by: coldpower27
Adding 8 Pipelines from NV40 to G70 was 80 Million Transistors, but the pipelines were also enhanced as well, also 2 more vertex units were added, enahnced buffer for 20x15 operation, if we be generous and give that amount for 8 pipeline that would bring the total to 382 which ~ the same level as R580 so it is possible to produce for Nvidia, if ATI can rpoduce a 384 Million Transistor GPU on 90nm why can't NV do the same?

You are assuming that they engineer their own dies on their own TSMC process, which is wrong. Remember the whole 5800 Ultra fiasco when they had the die shrink? They may build their engine and puzzle the parts together just like any car company would for instance, however they farm out the actual design of the substances that make up the puzzle pieces.

 

[coldpower27]

Originally posted by: Regs

Originally posted by: coldpower27
Adding 8 Pipelines from NV40 to G70 was 80 Million Transistors, but the pipelines were also enhanced as well, also 2 more vertex units were added, enahnced buffer for 20x15 operation, if we be generous and give that amount for 8 pipeline that would bring the total to 382 which ~ the same level as R580 so it is possible to produce for Nvidia, if ATI can rpoduce a 384 Million Transistor GPU on 90nm why can't NV do the same?

You are assuming that they engineer their own dies on their own TSMC process, which is wrong. Remember the whole 5800 Ultra fiasco when they had the die shrink? They may build their engine and puzzle the parts together just like any car company would for instance, however they farm out the actual design of the substances that make up the puzzle pieces.

What are you talking about, this is hardly the 5800 Ultra at all, Nvidia does have experience with 90nm GPU with the 7300 and IGP 6100/6150's, to me Nvidia isn't any less capable then ATI on 90nm and considering how long they worked on it should be albe to reach the same transistor counts give or take on the same process, since they are both using from the same person TSMC. Nvidia has played it smart with the 90nm procss and taken their time, now it is the right time to introduce high end GPU with die size similar to the 110nm parts Nvidia has brought because 90nm is now fairly mature.

 

[Regs]

It was a loose reference in stating the fact that they only work with what they can get.