NVIDIA Kepler speculation thread

You have to factor that they are targeting all segments of the market regardless the price point of an individual product. This hasn't really changed over the years aside from the low end where integrated has all but eliminated discrete products. You can actually get some idea of numbers if you look at market analyst (mercury or Jon Peddie) numbers which breakdown units moved for discrete and integrated.

 

It is true that they target all segments, and thus can keep a customer active if they can manage shifting him to a relatively lower performance segment at the same price. But that is a dubious proposition, and as you mentioned, the lower end of the market is eaten by integrated graphics which, to 80+%, is Intel, and those are lost customers to both AMD and nVidia.

 

Yes that's right (xx/yy)² of course being identical to (xx²/yy²). Add something for the parts which don't shrink well (i/o?) and you're on the right track. So a shrunk GF100/110 would be around 300mm² most likely. (Something like gate-last vs. gate-first on 28nm HKMG is said to have a density difference of 10-20% according to AMD so this is really just a rough estimate.)

 

As I recall, the 55nm node was just an optical shrink of the 65nm one, meaning you couldn't get a big improvement unless perhaps you were willing to go through a more significant design effort. I think TSMC advertised a 10% linear improvement at the time, or 21% in area.

I believe GT200 went from 576mm² to 485mm² or something like that, so it was more than 10%.

Well, there are people far more knowledgeable about these things here, but that's what I gathered.

 

Some other folks have touched on it already but there are only three real options.

#1. Pricing reflects competitive performance versus Tahiti.
#2. Pricing reflects competitive performance versus an incoming GK110.
#3. nVidia is just taking the piss to fuck with AMD's sales (unnecessary IMO).

If it's #2, I'm sorry but AMD will be in bad shape. Being capped at $299 is no fun and dual-GPU cards won't help much. Die size isn't very relevant either. nVidia has historically been willing and able to sell their big dies for less than AMD and still turn a profit.

Question of the day: where are the volumes for a $300 part coming from?

 

Not really.

Tahiti cant be worse than R600 And AMD has also survived a cycle in exactly that situation: top end competing in sub $300 segment and dual gpu solutions for higher segments.

It could also be that Nvidia is color coding the leaks and Charlie's sources' last days could be up.

 

there is also a lot of performance left on the table for a 7980, but the reality is we still know sweet f all about Kepler.

 

Depends on if you consider R600 just "bad shape" They will of course survive but I didn't say otherwise. Just that it won't be fun.

If we step back from fantasy for a bit then we come up with a troubled and delayed GK1x0 and a GK104 that can't quite romp with the big boys but is priced aggressively. That's probably closer to reality than some of the magical claims out there.

 

You missed supply and demand then. If you sell X cards at $500 you sell X+Y at $300. The price point you choose is not just based on what you can sell it for. It is based on what is the optimal price to maximize profits.

 

They aren't selling only 1 product. And x+ y has a limitation of the number of people interested in spending $300+ on a graphics card and you no longer benefit from those willing to spend $500 as you've effectively just given them a $200 discount by moving down the high end option.

 

It will be #2. nVidia will release the GK110 (single GPU) around mid year. It is a FACT that this GPU will be produced as it will be used in the HPC/Workstation market and already has a design win (Cray).

As a stop gap a dual GK104 may be produced to combat the dual card coming from AMD before the GK110 is available.

It does look like it will be a painful 2012 for AMD.

 

Please remove the OFF topic posts that are polluting this thread

This thread titled: Kepler speculation thread and has been derailed big time.

Why are all the off topic posts on AMD products/positioning/prices allowed? What exactly do they they have to do with what Kepler is/will be.

I expect a title of a thread "Kepler speculation thread" to be about Kepler.

There are other threads here for the AMD parts so please post to those threads.

Moderators please clean up this thread of the off-topic posts.
Thanks

 

Oh, there's no way in hell nVidia is abandoning the $500 segment if they can help it. If GK104 goes for $299 it's either because it can't justify a higher price or because they have something even faster that does. What's for sure is that they won't sell it cheap just for kicks.

 

If so, the strategy sounds like what AMD was doing at 4870.
One thing differenct is HD7970 is exceedingly not a big chip, means 7990 can come up soon.

 

Uh... 8800 GT anyone?

 

GK104 replaces GF114 in the NV line-up- 299$ perfectly matches a GF114 2GB at launch. Also consider the timeframe. If we believe that GK104 was planed for early Febrauary, but is now delayed until late March, then GK110 could still beready for late April or ealry May. No point in launching the GK104 @ 500$, when the real 500$ chip is right around the corner.

AMD should have no problems at all. Tahiti is just way too expensive for a performance chip at the moment.

 

You are effectively assuming a fixed number of market participants willing to pay at a given price point. That isn't necessarily true. You could convince a lot of people where were only willing to buy a $200 GPU to get a $300 one instead if it was the equivalent of a $500 GPU. They could always make something higher or the dual GPU, SLI stuff to pick up the top end buyers who want to throw more money.

All I am saying is the market is more nuanced than "Charge the maximum a consumer is willing to pay." If they did that then GPUs would be far more expensive and they would only sell one card to a person willing to pay the most

 

AMD already have a lot of problems. From the looks of all of it it'll certainly be a harder time in GPUs for AMD than it was during RV770->Cayman period. So I would say that they'll have even more problems soon. But this thread is really not about AMD -- I think there is a thread for that somewhere in 3D Industry forum.