• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

NVIDIA Pascal Thread

Page 83 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Eye-balling GM204 to the size of memory module and GP104 to the memory module, it look like GP104 is about the same, or even larger.
I don't have anything on this PC to do measurements.
 
You can just MS paint.

It's not that hard, and eye balling, GM204 is ~20% larger.

as rough as it is, comparison based on pixels between die and pcb resulted in 27% for GM204 and 25% of area.
Which means GP104 is 92% of GM204.
92% x 398= around 365mm sq

*if the substrate pcb is the same size.
 
as rough as it is, comparison based on pixels between die and pcb resulted in 27% for GM204 and 25% of area.
Which means GP104 is 92% of GM204.
92% x 398= around 365mm sq

*if the substrate pcb is the same size.

Easier to compare to GDDR5 since that's a known quantity.

IIRC, GDDR5 modules are 14mm x 12mm.

Eye balling, it looks like GP104 is 18mm x 16mm approximate, ~288mm2.
 
Last edited:
Btw, if true, these prices are ridiculous. lol

They just keep on creeping up higher and higher for mid-range.

Its not going to be any different with AMD. 14/16nm isn't cheap, neither in production or design. You are going to pay big premiums for the electrical benefits, aka perf/watt.
 
GTX 1080 - GDDR5X, 4 GPC, 2560 CUDA cores, 64 ROPS, 1480 MHz boost clock, around 200W TDP.
GTX 1070 GDDR5, 4 GPC, 2048 CUDA cores, 64 ROPs, 1480 MHz boost clock, 165W TDP.

Why not lower power consumption? New process allows 60% lower power consumption in the same thermal envelope OR 30% higher core clocks in the same thermal envelope.
 
I see so much mention everywhere, including wccf etc., referring to 970 and 980 replacements and sometimes a 1080ti is mentioned but almost as an afterthought. Any bets on the release being 1080Ti, 1080, 1070, with the 1070 being slightly faster than a 980, a 1080 being slightly faster than a 980Ti, and a 1080Ti being significantly faster, say 40-50%?
 
I see so much mention everywhere, including wccf etc., referring to 970 and 980 replacements and sometimes a 1080ti is mentioned but almost as an afterthought. Any bets on the release being 1080Ti, 1080, 1070, with the 1070 being slightly faster than a 980, a 1080 being slightly faster than a 980Ti, and a 1080Ti being significantly faster, say 40-50%?

There was a rumor that the second cut die is only roughly 970 performance and is for OEMs only.
 
NVIDIA-Pascal-GP104-vs-GM200-vs-GM204-GPU-Comparison.jpg



I hope not. Is there any reason to trust that chiphell source?

Regarding pricing, they have a mixed bag record, remmember Geforce GTX 980 Ti $800 pre-launch rumours. Let's wait and see.
 
Here is cross-section of both GP104 and GM204 based on this img (horizontal and vertical):
5S2Andv.jpg

xhOiDhf.jpg


PCBs are different size. Can't estimate die size based on that.
 
Last edited:
Shot are in line with cooler mounting holes, which can change from generation to generation. Explains the difference in white lines and why it doesn't match.
It is pretty big chip ^^
 
Btw they also confirmed DDR5 for 1070 and DDR5X for 1080.
1070 will sucks.
Its not 1070 its renamed GTX1060TI.
 
Last edited:
I get ~321mm²...

R9qYuEk.png


Edit - It looks like the shots are lined up with the white lines around the outside of each die.

Apropos the white lines, you can't necessarily trust those, since the GP104 picture has been clearly photoshopped. The die picture has been copypasted into another picture, which is quite obvious from the fact that the die looks perfectly clear, whereas the underlying PCB of the card is a mess of compression artifacts.

Basically videocardz took this image and unskewed it so that they got this image, they then took the unskewed image and copypasted into this image. There's quite a lot of room for error in this process.

For what it's worth videocardz estimated the die size at 332.9 mm2.

source
 
Last edited:
Shot are in line with cooler mounting holes, which can change from generation to generation. Explains the difference in white lines and why it doesn't match.
It is pretty big chip ^^

I see that now (the holes lining up), but the white lines do line up horizontally. I see you checked vertically and they are different.

Apropos the white lines, you can't necessarily trust those, since the GP104 picture has been clearly photoshopped. The die picture has been copypasted into another picture, which is quite obvious from the fact that the die looks perfectly clear, whereas the underlying PCB of the card is a mess of compression artifacts.

Basically videocardz took this image and unskewed it so that they got this image, they then took the unskewed image and copypasted into this image. There's quite a lot of room for error in this process.

For what it's worth videocardz estimated the die size at 332.9 mm2.

source

Yeah, last time I did this I was a little off (I think I got around 620mm² for Fiji, when it is actually 596mm²). Lots of room for error, but it is all for entertainment anyways.
 
Last edited:
Will be interesting to see how close to the Geforce GTX 980 Ti the GDDR5-based GP104 will be perf-wise, considering GM200's ~31% bandwidth advantage.
1070 looks like crap.
1080-max bandwidth 512GB/s
1070-256GB/s

1070 will be 20-25% cut + slow memory.i see Gap in 1080 vs 1070 25-30% minimum.Even more in games that are bottleneck by memory bandwidth.

Its really not 1070 at all.Its 1060TI selling for 1070 price you know.
Nv strikes again with dirty tricks
 
Back
Top