The Fermi Thread - Part 3

[***Deimos***]

well if deimos could clarify his original point that would help. are you talking about rebrands or just different sku's based off of g92? there is a difference all you have to do is look

I think the pattern is fairly obvious. All except probaby GT330, use "G92 and/or G92b".. ie the same chip.

Is 8800GS a rebrand? 9800GX2? I guess you could say GTS250 is a rebrand, but it has different PCB, memory, heatsink. box!! etc.

=============

G200 eol a good thing?
Imagine you bought a GTX275 or worse, a GTX295 yesterday. Week later, your "investment" resembles those GM stocks you thought would be a good idea.

============

Those arguing not to worry if nVidia loses money on GTX470.
So how do they recuperate several $100m of R&D expense?
Uber super cool and brand new DX10 G92 products?
Those high margin non-existent DX11 cards that haven't launched?
Perhaps the millions of Tesla products you see every teen in school buying?

What is my point?

GTX 280 launced with MSRP of $650. Because of AMD's 4870, it was reduced a week later to $500. Lets pretend board has fixed cost of $300. That's difference of $350 vs $200 for profit... or almost half as much.

Fastforward a 2 years later, and nVidia is still hasn't learned. Boards and chips are bigger, hotter and more expensive to make.

Everything else being equal, consumers would probably buy:
- cooler/lower-power product
- higher performance
- nVidia if it still represents great brand name image - questionable?

So, you can't just say 10% higher performance, so we'll make MSRP 10% higher - need to factor in cons of power/heat/availability. Things would be rosy if a 384 bit board, 50% more DRAM chips, and 50% bigger die with low yields only added 10% more to costs. Very likely a lot lot more.

With 0 competition, AMD has sold millions of DX11 products and made mucho $$$. nVidia is at $0.00 And, people who already own a DX11 card, are unlikely to "upgrade".

So, just like AMD with Phenom(1), how do you sell the product?:
- lower price? -> temporary gain in marketshare at expense of less profits.
- lower power? -> trouble enough as it is getting working dies
- higher clocks? -> power is already very high
- lower costs? -> would you buy more expensive 768MB card? What if "some assembly required"?
- PhysX marketing? -> trouble is, its nothing new since all GF8/9/200 have it.
- DX11 promotion? -> sucks you spent last 6 months saying its not a big deal.
- CUDA GPGPU apps? -> its nice bonus. People don't buy $500 video cards to accelerate their Adobe PDF Reader.
- EyeInfinity? -> Don't got none.

The desperate "solution" is the same as it was in fx era... its the only thing you can really change...
questionable driver optimizations at the expense of rendering quality and driver stability.

Ofcourse like G80, about a year later they're gonna have a 32/28nm 448SP 256bit 1GB GDDR5 board. They're gonna call it "480GT"