Why are Nvidia's die sizes always much larger than AMDs?

[xsistor]

So you're right about it being less space optimised.

It could be the architecture or it could be because as the design grows larger the layout optimisation algorithm grows with exponential time complexity... Probably a bit of both.

 

[Operation]

So in conclusion, I can just take mmj_uk's explation at face value and get on with my life, yes?

 

[Owenb]

To make things extremely simple: this is precisely the correct answer.

Except that you are wrong. The 280GTX was had 50% more transistors than the 4870 yet it was only ~10% faster and that had nothing to do with extra hardware. The reason it didn't preform better was because it couldn't hit the clocks that Nv wanted to hit, and neither could Fermi.

Of course it isn't as bad as it seems. If even a chip is ~30% bigger but only 10% faster doesn't make it a bad chip. The chip is only a fraction of the cost of the finished card. Lets say that AMD needs to sell a chip for $70 to make a profit and Nvidia $100 but that Nvidia's is 15% faster. AMD's chip looks better value, but once the cost of the board/memory/profit margins are added in the cards could sell for $220 vs $250.

 

[xsistor]

Except that you are wrong. The 280GTX was had 50% more transistors than the 4870 yet it was only ~10% faster and that had nothing to do with extra hardware. The reason it didn't preform better was because it couldn't hit the clocks that Nv wanted to hit, and neither could Fermi.

Of course it isn't as bad as it seems. If even a chip is ~30% bigger but only 10% faster doesn't make it a bad chip. The chip is only a fraction of the cost of the finished card. Lets say that AMD needs to sell a chip for $70 to make a profit and Nvidia $100 but that Nvidia's is 15% faster. AMD's chip looks better value, but once the cost of the board/memory/profit margins are added in the cards could sell for $220 vs $250.

LOL that has nothing to do with why the dies are different sizes. The reasons a chip doesn't hit clocks they wanted it to are fairly complex (2nd order effects in transistor-resistor-capacitor features in silicon, for example). But the bottomline is
a) NVIDIA's chips are bigger than AMD's because of a fundamental architectural difference.
b) This difference is motivated by NVIDIA's emphasis on gpGPU, and AMD's lack thereof
c) there are a host of other possible reasons. see below...

So in conclusion, I can just take mmj_uk's explation at face value and get on with my life, yes?

Yes. I've been speaking to the guy I know from NVIDIA and we came up with a big list of very specific reasons (architectural and electronics related) why NVIDIA chips may be bigger... If you really want more details I can post it here with some brief explanations....

The GF116 is the current version of the GF106 found in the GTX550TI. The GTS450 had part of the memory controller disabled but the GTX550TI is a fully enabled part.

The GF116 has the same number of transistors and the same die size as the GF106. However in games its performance is much lower than an HD6870 which is around 30% to 40% faster.

You can see a similar trend with the GPUs in the HD6970(Caymen) and the GTX560TI(GF114).

Cayman has 2.64 billion transistors and the GF114 1.95 billion transistors. Cayman is 389MM2 whereas the GF114 is around 330MM2. A GF110 is around 3 billion transistors and is 520MM2 in size.

The list also takes a deeper look into this transistor count discrepancy, explains the causes and then finally also shows why it doesn't mean a thing.


... Of course a lot of of the stuff in this list is based on informed knowledge of VLSI and microprocessor design and is more of a list of probable causes. Chances are 90% of them are the actual reasons for these issues. It is impossible to exactly say why without an exhaustive comparative study of both chips from an architectural/design viewpoint down to a VLSI viewpoint... And something like that would also take way too much time and effort.

 

[Yamahahahahaha]

There's a good article over at Anandtech written by some of the insiders of ATI / AMD's team.

http://www.anandtech.com/show/2679/1

tl;dr
Took a chance on making a smaller, higher clocked chip. Higher yields = lower prices, more profits.
Gamble paid off, Nvidia struggled transitioning to other process technologies while AMD didn't.

 

[xsistor]

There's a good article over at Anandtech written by some of the insiders of ATI / AMD's team.

http://www.anandtech.com/show/2679/1

tl;dr
Took a chance on making a smaller, higher clocked chip. Higher yields = lower prices, more profits.
Gamble paid off, Nvidia struggled transitioning to other process technologies while AMD didn't.

Skimming through, that's actually a pretty good link.

" If you have a unit with 10 shader processors, actually build 11 but use one as a spare should there be a defect in any of the remaining processors. The same technique is often used for on-die memory, include additional cells so that a defect in one part of a cache won’t render the entire chip unusable."

In fact this is one likely reason why the GTX 580 chips has less active transistors. Because the architecture has CMOS SRAM requiring more redundancy

 

[wolvers]

What I'd like to know is, are NV's larger dies the driving force of the market prices? Are NV stuck with prices that they have to charge and are AMD just taking advantage of that to sell at a higher price? If not, why aren't AMD able to sell for less and if so what's topping them undercutting NV and taking a larger market share?

 

[Hanuki]

Also; does this mean AMD can make more profit per GPU considering they use less silicon than Nvidia?

Could AMD have a potentially stronger GPU if they used larger die sizes or are they already limited in other ways?

Because size matter : /

 

[xsistor]

What I'd like to know is, are NV's larger dies the driving force of the market prices? Are NV stuck with prices that they have to charge and are AMD just taking advantage of that to sell at a higher price? If not, why aren't AMD able to sell for less and if so what's topping them undercutting NV and taking a larger market share?

The size definitely does make it more expensive. NVIDIAs gpus are basically graphics processors + parallel processors. Therefore they have some advantages (e.g. scientific computation, tesselation, PhysX etc). AMD has a price/performance advantage in traditional graphics.
If AMD could they would probably sell and price their products as aggressively as possible to remove NVIDIA as a competitor. In fact this is probably the reason NVIDIA is slightly more expensive than AMD in pure graphics performance. For some of us the added compute and host of other features (like 3D) make NVIDIA the only choice we can ever make.
But NVIDIA can remain competitive because they can afford to make little to no profit on some of their products and try to make up that profit in other areas, being a bigger company than ATI catering to a broader market. I don't know too much about the economics side of it, but I've heard that NVIDIA makes a big chunk of its profits from the professional sector - AMD has no part in that.

But it's also because NVIDIA is designing products for a broader market segment with a wider set of features that its GPUs are architected differently, and therefore larger dies are necessary.

Ultimately, the new architectural features of NVIDIA gpus are design constraints/goals. And cost is always a design constraint in engineering, and cost-minimization is a very standard no matter what you do.

AMD could make a bigger die, and it would have to if it wanted to improve its gp features. but as it stands AMD cards are targeted exclusively towards gamers and so it doesn't have to. NVIDIA cards, on the other hand, have other additional design goals.