AMD: R9xx Speculation

Doesn't look that way

 

True, but the higher the number of ALUs the more complex the design, the more likely errors will occur.

 

possible, but thats more of chip design methodolgy, I don't think that will effect yields directly.

 

Let's see for GTX580
US price (newegg) ~$525 excluding VAT = ~400€
French price (surcouf) ~450€ including VAT = ~$595
French VAT = 19.6%, so French price excluding VAT = ~375€ = $495

So US price excluding VAT ~$525, French price excluding VAT = $495

 

First prices for France seems to be 379€ and 479€ which are around the GTX 570 and 580 prices...




http://www.comptoir-hardware.com/ac...t-hd-6970-referencees-en-france-les-prix.html

 

Indeed, but a 2000 ALU chip is going to have significantly lower yields than an 8 core CPU.

 

That is totally untrue. The 1:1 ratio is with taxes included, at least these days (granted taxes are not the same everywhere but I think in most countries using euro somewhere between 15-20%).

 

How do you figure that?

mczak > 19.6% in France.

 

hmm doesn't work that way , complexity of chips is ALU's alone, if a chip utilizes more transistors it could be more complex to design, it might have lower yields, if the transistors are needed for more cache, it might be easier to design, and yields could be higher. There can be many things in the middle. that could effect yields, but not just because of ALU count.

AMD's gpus have more ALU's then nV's but nV's designs are more complex, so that negates what you are saying.

 

It depends on the amount of redundancy, but I will go out on a limb and say a 2000 ALU GPU is significantly harder to get decent yields on than an 8 core CPU. Which is why Fusion designs have been delayed on the 32nm High-K process and the vanilla CPUs haven't.

 

The sildes are ok for 24 and more simds. Maybe there wasnt any 1920 SP part.

 

http://www.digitimes.com/news/a20101208PD229.html

 

I'm starting to get the feeling that with the failure to advance processes as time goes on, AMD's small die strategy is going to start to backfire because they're aiming too low

edit: that being said, the final performance is unknown so we have to see what happens

 

I don't understand you... In fact they are aiming higher and higher.

HD3870: ~65% performance of 9800GTX
HD4870: ~75% performnace of GTX280
HD5870: ~85% performance of GTX480
HD6970: ......... performance of GTX580 (please fill in by your own)

 

It's too early to say, but if there is a general slow down in process shrink then ATi will follow Nvidia in making larger more power hungrly chips. It gives Nvidia an advantage because they are already experienced in making large hot chips work. ATi are geared towards shrinking their chips on time and delays will work against them because of this. I suppose you can say JHH was smart to bet against the tehcnological prowess of TSMC and build a 40nm product that pushed the limits, but I get the feeling he lucked into that!

 

I wonder if this increase parallels the die size ratio between AMD/Nvidia chips.

HD3870: ~65% die size of 9800GTX
HD4870: ~75% die size of GTX280
.
etc.

Well, I guess I can check this myself.

 

Some GTX570 available for 300€... perhaps AMD discovered some magic/alien tech after all?

 

These transitions are all made with timely process shrinks, the one time there hasn't been one(this time) ATi might not increase their relative performance this time around.

 

U forget the dual card. If the new cards will show same scaling as the 68xx cards in crossfire than nvidia is second again.

Also the money is not in the high end cards. And game developers usualy dont care a shit about nvidia 580gtx or radeon 5970 cards.

 

Pretty solid source.
So when AMD says "we'll have the fastest card for most of 2011" they're talking 5970 and 6990 obviously.