Thoughts on die size of Radeon 7000 and GeForce 600 Series

[kambon]

I have just made a small comparison between Radeon 7000 and
GeForce 600 series graphics cards:

http://gpus.herokuapp.com

The factors taken into account:
die size, manifacturing process and
the market segment targeted by each gpu.

I do not know how good I distributed the different
chips to their respective segment. Correct me
if I am wrong.

Conclusions:
It is mix of gpus made in old 40 nm process
with the new one 28 nm. The lowest segment is occupied
by 40 nm parts and the rest of the segments mainly by 28 nm parts.
GeForce has smaller die size of its top chip, GK104, compared
with Radeon top chip, Tahit. That allowed NVidia to target
lower segments with the same chip. In this way NVidia
reduces the producting of a separate product line.

Questions:
I am just wondering if the integrated gpus takes more
market share, will we see just 1-2 new chips per
generation instead of 3-4 before?
What do you think?

[Mathragh]

Regarding your questions, I think that if and when the performance of integrated solutions gets closer, the dedicated cards would either have to: Get more expensive/powerfull, or be released more often. I think this because they need to keep the gap between iGPUs and dedicated ones big enough so people will keep wanting to buy new cards instead of relying on the intergrated stuff. Also, at some point most people will perhaps be content with the power that is provided by iGPUs alone, regardless of how much more powerfull dedicated cards are. I still see enough graphical power(or plain parallel power for that matter) needed in the near future for people to want dedicated solutions though.

I think that as soon as consoles only use integrated GPUs, the market for dedicated GPUs might be in trouble.


Regarding your site: I think this kind of thing can be a powerfull tool. However, in order to reach any real cool conclusions, you should think of including as many specifications as you can, like poweruse, clocks, amount of transistors, Gflops, price,and perhaps even some real world benchmarks.

That way, you can compare a lot more things and come to cool conclusions, like, efficiency per transistor, or flops per area.

Just a thought

Curious as to what you might find!

[kambon]

Regarding your questions, I think that if and when the performance of integrated solutions gets closer, the dedicated cards would either have to: Get more expensive/powerfull, or be released more often. I think this because they need to keep the gap between iGPUs and dedicated ones big enough so people will keep wanting to buy new cards instead of relying on the intergrated stuff. Also, at some point most people will perhaps be content with the power that is provided by iGPUs alone, regardless of how much more powerfull dedicated cards are. I still see enough graphical power(or plain parallel power for that matter) needed in the near future for people to want dedicated solutions though.

I agree with you.

Regarding to your site: I think this kind of thing can be a powerfull tool. However, in order to reach any real cool conclusions, you should think of including as many specifications as you can, like poweruse, clocks, amount of transistors, Gflops, price,and perhaps even some real world benchmarks.

I tried to include up to now factors that are undisputable:
die size, core count, manifacturing process. They are
more oriented to the cost rather than the income. I think
in Techpowerup.com there is a good set of reviews that
include enough benchmarks. The price of the products
is usefull. Unit sales is great factor. Unfortunately, only
the manifacturers know the exact number of sales.

[Brusfantomet]

Don't forget GK 100, it is not for consumer cards, but still works as a GPU. There are stil 3-4 chips per generation tho. (Tahiti, Pitcairn and Cape Verde from AMD and GK100, GK104, GK106 and GK107 from Nvidia) but the performance of integrated solutions is eating into the lower end cards.

Consoles starting to use integrated GPUs will not change a thing in my opinion, as they already get outpaced rather fast by desktops anyway. 360 uses a modified ati x1900 and ps3 a modified nvidia 7800, both of these cards are underpowered today, and was mid range round launch of the consoles. If the integrated solutions start getting into mid range territory they could get included in consoles.

Another interesting thing with GPUs and consoles is that with both vendors pushing a working multi GPU system, GPU upgrades for the possible integrated chips can be done (the N64 had a ram upgrade available that some games required) this would extend the life of the console, and the limited hardware con-figs possible (with or without GPU upgrade) would make it easier get most of the performance form the multi GPU solutions

Just my thoughts on the subject