• 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.

"Big" Maxwell GM200 Has Been Spotted!

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Cloudfire777 said:
That GK104 in 28nm is 295mm2 while GM204 is 430mm2.
Not a good way to start off with 20nm to put it lightly...
Click to expand...

Wait...so now a bigger die is bad? I thought people were wanting "Big Maxwell" right out of the gate.

Seriously, you're willing to proclaim feature size based on the chip size? Really?
 
Mand said:
Wait...so now a bigger die is bad? I thought people were wanting "Big Maxwell" right out of the gate.

Seriously, you're willing to proclaim feature size based on the chip size? Really?
Click to expand...
#1
GF104 40nm- 332mm2
GK104 28nm - 295mm2
GM204 ?nm - 430mm2

#2
A 430mm2 20nm GM204 would feature a massive amount of CUDA cores and its not needed because Maxwell cores perform better than Kepler cores. Do you really think Nvidia would release a GPU capable of doubling performance over GTX 780Ti? No they do it step by step, hence why you see all the first high end GPUs sticking to around 300mm2 dies when a new node arrives.

That they increased the die from 295mm2 to 430mm2 is a huge tell that they did it to be able to fit more transistors on the chip because they are still stuck on 28nm
 
Cloudfire777 said:
#1
GF104 40nm- 332mm2
GK104 28nm - 295mm2
GM204 ?nm - 430mm2

#2
A 430mm2 20nm GM204 would feature a massive amount of CUDA cores and its not needed because Maxwell cores perform better than Kepler cores. Do you really think Nvidia would release a GPU capable of doubling performance over GTX 780Ti? No they do it step by step, hence why you see all the first high end GPUs sticking to around 300mm2 dies when a new node arrives.

That they increased the die from 295mm2 to 430mm2 is a huge tell that they did it to be able to fit more transistors on the chip because they are still stuck on 28nm
Click to expand...

That logic doesn't even explain 28nm though. Maxwell cores are better than Kepler cores. If they didn't reduce size, that's still a ton of cores added in that extra 135mm^2 (nearly 50% increase). How is it that your logic doesn't mean that they went back to 40nm?

I mean, it's a nice observation, but it's far from a reason to believe a guess on feature size is going to be accurate.

You're also completely going off of the assumption that they will always try to launch the first one at 300mm^2. What if they didn't? All your assumptions go out the window, at that point.
 
Fake.

And the only way GM200 gets 'more cores' into the same die size as GK110 is with TSMC's 16nm, which is nowhere yet. More gets packed into the same area with a smaller process allowing more transistors in the same space. 28nm GM200 ?

Fake.

The 400mm2 die we've seen for a GM204 is as big as it will likely get on 28nm, it may only allow for 20% or so more performance than a GK110, but what other option is there when they can't afford 20nm TSMC and have to wait to get 20nm with FF (16nm)
 
Last edited:
Mand said:
That logic doesn't even explain 28nm though.
Click to expand...
Read my post again.

If they didn't reduce size, that's still a ton of cores added in that extra 135mm^2 (nearly 50% increase).
Click to expand...
50% increase in die area and a ton of more transistors overall compared to GK104 and even GK110. It would be too fast. Period.

How is it that your logic doesn't mean that they went back to 40nm?
Click to expand...
You go down a node, you double the transistors. You go back a node, you get 1/2 the transistors on the same area.
Why would they go back to 40nm? GM204 could be 40nm but it would be horrible compared to GTX 780 Ti and would not sell. Why are you referring to "logic" several times in your post? Its not logic, its common sense

You're also completely going off of the assumption that they will always try to launch the first one at 300mm^2.
What if they didn't? All your assumptions go out the window, at that point.
Click to expand...
I just gave you a reason why they would.
A 430mm2 Maxwell GPU in 20nm that is a successor for GK104 would maybe be 2.5-3x faster than GK104. When was the last time that happened?

Never. Because that would be a pretty stupid strategy financial wise. Its the same reason why we didnt get GK110 first and why GM200 comes after GM204. You can`t just release the biggest Maxwell chip you can build, price it at $1000 and have nothing to sell to the owners of that chip 2 years down the road.
 
Last edited:
Deders said:
http://www.kitguru.net/components/g...0nm-process-technology-jump-straight-to-16nm/

Nvidia may skip 20nm altogether
Click to expand...

Yup that is what will happen.

Rumors had it Nvidia already made 20nm designs for their GPUs, but production availability was too low, only one process (LP - SOC) is being used at TSMC which may not have been ideal for AMD/Nvidia and their GPUs. So Nvidia had to port their designs to 28nm which is why we have seen GM204 and GM200 samples flying back and forth between USA HQ and GPU HQ in India the last 3 months. Most likely in a hurry to make a deadline.

AMD said they would do 20nm for upcoming GPUs but they are in a different situation that have a partnership with GloFo which is said to use HPM for 20nm designs. So that could work for them and be just as efficient as 16nm FinFET from TSMC. Even AMD said they would concentrate on 28nm for GPUs this year so I think that gives Nvidia a peace of mind when they are now using 28nm for GM204 and GM200
 
Last edited:
You must log in or register to reply here.

TRENDING THREADS

Back
Top