NVIDIA GF100 & Friends speculation
- Thread starter Arty
- Start date
NathansFortune
Regular
So closer to the original estimate then?
Charlie said 367mm², actually.
NathansFortune
Regular
If it is 340mm2 then it will have higher yields (along with more dice per wafer) giving more room for Nvidia to profit and make price cuts.
There is also a mini-flamefest here in the background over whether Charlie can be trusted or not, if the die is conclusively measured at ~340mm2 then he should no longer be a trusted source of information because of this article. If it is measured closer to what he says it vindicates his position.
On the size itself, I would have thought the good heat characteristics and low power consumption would have pointed to a smaller die size than the article pointed out, if it were as big as Charlie says we would be seeing much fancier custom coolers.
It's been demonstrated time and again that Charlie's information is spotty at best. He expresses certainty about many things, a number of which turn out to be flat wrong. It's clear that he does have some inside information, but it's never clear which of his information is true beforehand. Thus it's best to consider him unreliable.
Are you really correlating die size and power dissipation? Even if that that would be true, a 8% larger die wouldn't result in a massively larger cooler.
He's without doubt much to biased in his views to be creditable.
I don't think we really need to discuss that after the "GF100 doesn't scale" bla bla and the bumpgate article revival just in time for the GTX 460 launch.
He's without doubt much to biased in his views to be creditable.
Last edited by a moderator:
Silent_Buddha
Legend
Has anyone bothered to check if his heatspreader measurements are even correct?
Regards,
SB
The PCIe connector seems to be ~85mm in size: http://www.interfacebus.com/PCIe_16x_Connector.gif
That gives a spreader size of ~45.7mm * 35mm and a die size of 364mm². Seems like Charlie was right after all.
IMHO; price cuts are not possible for the GTX460 regardless of the actual die size. IMHO, you don't need any inside informations to see this, you only have to look at the actual configuration of the GTX460 cards:
Due to: GPU price ---> memory <----card price / indended market price
The price for the GF104 GPU seems to be very high for the intended market price. Therefore the OEMs and Nvidia had to use the strange 192bit/768MB memory configuration for the baseline GTX 460 to make the intended market price of the GTX 460 of ~US$ 199 possible.
IMHO, in the end not even Nvidia will be happy with this market position because the margins for the GF104 will be rather low too.
aaronspink
Veteran
The PCIe connector seems to be ~85mm in size: http://www.interfacebus.com/PCIe_16x_Connector.gif
That gives a spreader size of ~45.7mm * 35mm and a die size of 364mm². Seems like Charlie was right after all.
Considering he all but confirmed he has an xray inspection measurement with <.01 mm accuracy, of course he's right. Charlie isn't an idiot, if he is giving a hard number, he has some information to backup that number and if you have either the right connections or a board and the right equipment, you can measure these things pretty reliably.
DavidGraham
Veteran
xbitlabs made a fairly good cover-up on tessellation performance between ATi and Nvidia in current DX11 games :
http://www.xbitlabs.com/articles/video/display/hardware-tesselation.html
The difference between the two is between 5~10% at best , with nvidia being the one that has the upper hand .
however what really caught my eyes is that :
1-GTX480 has a 45% advantage over HD5870 in STALKER COP @ 1920x1080 , decreasing to 30% @ 2500x1600 .
2-GTX480 has a 68% advantage over HD5870 in AVP @1920x1080 , increasing to a whooping 81% @ 2500x1600 !
3-GTX480 has a 27% advantage over HD5870 in Dirt 2 @ 1920x1080 , decreasing to 13% @2500x1600 .
4-HD5870 has a 3% advantage over GTX480 in Metro2033 @ 1920x1080 without AA or AAA , with the situation being reversed @ 2500x1600 (GTX480 has a 3% lead) .
All games are maxed out with 4XAA and 16XAF except for Metro2033 , with 258.96 and 10.6 drivers .
I think there is something wrong with this piece , since when does GTX480 enjoy such a massive lead ? , I though it was merely 15% faster .
http://www.xbitlabs.com/articles/video/display/hardware-tesselation.html
The difference between the two is between 5~10% at best , with nvidia being the one that has the upper hand .
however what really caught my eyes is that :
1-GTX480 has a 45% advantage over HD5870 in STALKER COP @ 1920x1080 , decreasing to 30% @ 2500x1600 .
2-GTX480 has a 68% advantage over HD5870 in AVP @1920x1080 , increasing to a whooping 81% @ 2500x1600 !
3-GTX480 has a 27% advantage over HD5870 in Dirt 2 @ 1920x1080 , decreasing to 13% @2500x1600 .
4-HD5870 has a 3% advantage over GTX480 in Metro2033 @ 1920x1080 without AA or AAA , with the situation being reversed @ 2500x1600 (GTX480 has a 3% lead) .
All games are maxed out with 4XAA and 16XAF except for Metro2033 , with 258.96 and 10.6 drivers .
I think there is something wrong with this piece , since when does GTX480 enjoy such a massive lead ? , I though it was merely 15% faster .
DavidGraham
Veteran
they mentioned using only default CCC settings , (except High Quality Mip-Maps) .. same for Geforce .
I'm not sure, if we talk about the same thing...
http://www.xbitlabs.com/articles/video/display/hardware-tesselation_4.html#sect0
http://www.xbitlabs.com/articles/video/display/hardware-tesselation_4.html#sect0
DavidGraham
Veteran
I see , thanks for the clarification , what a weird thing to do , why would they do that ?!
Similar threads
- Deleted member 2197
- Architecture and Products
