GF104 core size revealed

[tbone8ty]

"Early information has claimed a surface area of 366mm^2 with GF104, which is confirmed by these pictures. You can compare this to the 334mm^2 of Cypress"

http://www.nordichardware.com/news/7...-revealed.html

[Manicdan]

i wonder if it still works....

[STEvil]

If they had taped it off: yes.

Since they didnt..... i'm gonna say 60% possibility yes.

[flippin_waffles]

Charlie was right yet again!

[Luka_Aveiro]

So, would a full enabled GF104 consuming at load the exact same power of an HD5870 be a better performer or a worse performer???

[STEvil]

So, would a full enabled GF104 consuming at load the exact same power of an HD5870 be a better performer or a worse performer???

uh.. that would depend on performance and power used, no?

[nn_step]

So, would a full enabled GF104 consuming at load the exact same power of an HD5870 be a better performer or a worse performer???

that depends entirely on the task being performed; since both chips are optimized in different manners.

[=SOC= Admiral]

Charlie was right yet again!

Any predictions on when the world is going to end?

[Sn0wm@n]

Any predictions on when the world is going to end?

2012 obviously


ohh and charlie was right about the shape of the die if the pic is real

[hurleybird]

So I guess just confirms again ATIs architectural lead in Perf/mm2 over Nvidia. With that being said, a fully enabled GF104 with good cooling and high clocks should be able to get within spitting distance of HD5870.

Also brings the question on how much money Nvidia is really making on GF104. My guess would be not very much, and that GF104 is positioned to capture/hold market share from ATI so Nvidia can make more money in the future.

Things are going to get interesting for Nvidia really soon, with southern Islands on the way soon, and good integrated-in-the-cpu graphics coming from both AMD and Intel. If worst comes to worst, Nvidia will lose the ability to compete in the high-end and the low-end, while their mid-range might make very little money due to the perf/mm2 disadvantage over ATI. They still have the professional market nicely (for the most part) to themselves though.

[saaya]

hmmm thats bigger than i thought... and what a weird shape huh?
seems to be about the size of amds first a64 chips and later dualcore chips?

[Russian]

Charlie was right yet again!

About? I don't really read his site.

[antiacid]

This shape is a sign of things to come.

Next up, the U-shaped silicon, it is built with watercooling in mind and comes with a pre-built block that works as an IHS as well! You read it here first! :P

Just wait until we start hitting the cool shapes like octagon or enneadecagon!

[STEvil]

This shape is a sign of things to come.

Next up, the U-shaped silicon, it is built with watercooling in mind and comes with a pre-built block that works as an IHS as well! You read it here first! :P

Just wait until we start hitting the cool shapes like octagon or enneadecagon!

I made this many many years ago... I feel so ashamed of hiding my prototype silicon engineering wonder for so long!!!!

[-Boris-]

hmmm thats bigger than i thought... and what a weird shape huh?
seems to be about the size of amds first a64 chips and later dualcore chips?

A64 debuted at 193mm²
A64 X2 at 199/147mm² depending on cache. So no, GF104 is almost double the size.

[xdan]

This is a picture posted two days ago here on Nvidia section. Nordic hardaware doesn't say anything about height and width, which are essentials.
From this picture if we calculate 13.7*24.2= 331.54mm2 which is a thiny smaller than Cypress which is 334mm2.
So don't rush with wrong presumptions. I think that is truths let say that GF104 is equal to Cypress otherwise i don't think Nvidia would have such low prices for GTX 460 and such good temperatures and power consumption.

[Iconyu]

About? I don't really read his site.

If you're serious, Charlie claimed GF104 was over 360mm^2 and that nvidia had released the the 460 at it's current price to maintain market share.

[Johnny87au]

gee its huge!

[qcmadness]

This is a picture posted two days ago here on Nvidia section. Nordic hardaware doesn't say anything about height and width, which are essentials.
From this picture if we calculate 13.7*24.2= 331.54mm2 which is a thiny smaller than Cypress which is 334mm2.
So don't rush with wrong presumptions. I think that is truths let say that GF104 is equal to Cypress otherwise i don't think Nvidia would have such low prices for GTX 460 and such good temperatures and power consumption.

yes this estimate is more reasonable

GF104: 1.95b transistor count
RV870: 2.15b transistor count

GF104 should not be bigger than RV870 for more than 10%

[DeltZ]

It's a rectangle

[ohnoitseddy]

yes this estimate is more reasonable

GF104: 1.95b transistor count
RV870: 2.15b transistor count

GF104 should not be bigger than RV870 for more than 10%

Keep in mind that ATI makes more 'dense' chips than nVidia, they're able to put more transistors in the same space.

GF100: 3.0b transistors in 529 mm^2: 5.7m transistors / mm^2
Cypress: 2.15b transistors in 334 mm^2: 6.4m transistorts / mm^2

With the GF100's density in mind, you'd say it would be 344 mm^2 (this is probably how Charlie got his estimate), which is bigger than Cypress. That said, I guess 331 mm^2 still makes more sense than 366 mm^2.

As for why it's a rectangle - Perhaps that way nVidia could engineer it more easily? From the die shot we know that in GF100 there's four square-ish blocks that house the shaders, since GF104 has probably has two of those with more shaders each, it might not have been worth the trouble making it a perfect square again.

[xdan]

GF100: 3.0b transistors in 529 mm^2: 5.7m transistors / mm^2
Cypress: 2.15b transistors in 334 mm^2: 6.4m transistorts / mm^2

If i calculate by this numbers 1.95b/5.7=342mm, and count that GF 104 has less L3 cache and Polymorf units than GF 100.
And also i think that because of larger SM units, 48SP instead of 32SP some space is also gain due the lower numbers of SM modules, so lower numbers of SM "walls" modules if you understand.
So my opinion is that GF104 is = Cypress = 334mm2 and point. All other speculation are useless.

[LordEC911]

This is a picture posted two days ago here on Nvidia section. Nordic hardaware doesn't say anything about height and width, which are essentials.
From this picture if we calculate 13.7*24.2= 331.54mm2 which is a thiny smaller than Cypress which is 334mm2.
So don't rush with wrong presumptions. I think that is truths let say that GF104 is equal to Cypress otherwise i don't think Nvidia would have such low prices for GTX 460 and such good temperatures and power consumption.

Wrong... the original article I saw in Chinese was trying to hide the real die size. B3D worked out the math and found out the length and width on the picture is wrong.

It's a rectangle

Pad limited. Rectangle is usually the best way to fit the most I/O.

With the GF100's density in mind, you'd say it would be 344 mm^2 (this is probably how Charlie got his estimate), which is bigger than Cypress. That said, I guess 331 mm^2 still makes more sense than 366 mm^2.

Charlie basically confirmed that he has some sort of access to an x-ray inspection measurement device, see link.
http://forum.beyond3d.com/showpost.p...postcount=6394

I said it quite awhile ago. GF104 is larger than Cypress, larger than G92 and I was guessing around 350mm2. Turned out it was even larger.

[Boissez]

So I guess just confirms again ATIs architectural lead in Perf/mm2 over Nvidia. With that being said, a fully enabled GF104 with good cooling and high clocks should be able to get within spitting distance of HD5870.

Also brings the question on how much money Nvidia is really making on GF104. My guess would be not very much, and that GF104 is positioned to capture/hold market share from ATI so Nvidia can make more money in the future.

Things are going to get interesting for Nvidia really soon, with southern Islands on the way soon, and good integrated-in-the-cpu graphics coming from both AMD and Intel. If worst comes to worst, Nvidia will lose the ability to compete in the high-end and the low-end, while their mid-range might make very little money due to the perf/mm2 disadvantage over ATI. They still have the professional market nicely (for the most part) to themselves though.

Well said

Unless SI flops I have a hard time seeing Nvidia making much money in the near and the distant future.

OT: This actually makes me wonder: How come there aren't any "Jensen RAUS!"- threads? Like back when Ruiz was slowly killing AMD with his mismanagement.

[xdan]

I still don't buy what Charlie says or others fans off......
This guy totally proved that is against Nvidia.
See this also...
http://forum.beyond3d.com/showpost.p...postcount=6381