0,13 Low-K vs 0,11 - What's going to be nV's answer to X850?

[azopi]

What are the advantages and disadvantages of each technology?
What's the reason of the lack of 0,11 in the case of High-End GPUs, and the success of the same technology on the Mainstream/Low-End market?
ATI is releasing one more High-End product line (R480) produced on the good, old 0,13 Low-K just as the previous flagships, 9800 and X800, and its mainstream product line - The R430 is based on the shiny new 0,11.
I'm a bit confused.

On the other hand, the other player, nVidia still hasn't leaked the specs of it's refresh parts. It's unknown whether it will use 0,13 or 0,11.
Using of Low-K seems to be unlikely since nV has no experience with this technology. But sticking with 0,13 non Low-K is possible.
What's your bet? Will nV go on 0,11 with the successors of 6800NU, 6800GT and 6800U since it managed to get pretty good yields with the 6600 line?

It must be much cheaper than the 130 nano tech and much higher clockspeeds can be achieved on the same transistor count what can make ATI run for its money.

But why is the Canadian GPU maker sticking with 0,13 Low-K instead of jumping to 0,11

 

[Fodder]

110nm is still new, and ATI don't seem to have had as much success with it as NVIDIA (see X700 struggling to hit 500MHz).

 

[Dave Baumann]

X700 is running a lower voltage than 6600 GT.

110nm is primarily a lower cost option and as such, it is not offered with a low-k option; low-k can very much be an aid to higher clockspeeds, and the majority of 90nm processes appear to have it in place. So, at the moment 130nm low-k is sold as the high performance option by TSMC, whilst 110nm is sold as the "cost optimised" process (and has the same inherant properties as their 130nm FSG process).

Of course, whilst 110nm isn't necessarily sold as a high performance (clockspeed) process it does buy die space, so manufacturers have the option to make smaller parts for cheaper or the same size parts (relative to 130nm) that will have more transistors and using more transistors is another way of gaining performance as opposed to pure clockspeed.

 

[azopi]

110nm is still new, and ATI don't seem to have had as much success with it as NVIDIA (see X700 struggling to hit 500MHz).

So it will be a challenge for ATI to get the 16 pipe X800XL and even the 12 pipe X800 clocked at 400 MHz on 110nm, won't it?

So why don't they use 130 low-k on mid-range parts as well? Why do they gamble with 110 nm? They're trying to remain still competitive beside nVidia's coming 110 nano GPUs as 130 nm Low-K process seems to be far too expensive to make it to the mainstream market.

ATi must face a hard quarter, I suppose.

 

[azopi]

X700 is running a lower voltage than 6600 GT.

110nm is primarily a lower cost option and as such, it is not offered with a low-k option; low-k can very much be an aid to higher clockspeeds, and the majority of 90nm processes appear to have it in place. So, at the moment 130nm low-k is sold as the high performance option by TSMC, whilst 110nm is sold as the "cost optimised" process (and has the same inherant properties as their 130nm FSG process).

Of course, whilst 110nm isn't necessarily sold as a high performance (clockspeed) process it does buy die space, so manufacturers have the option to make smaller parts for cheaper or the same size parts (relative to 130nm) that will have more transistors and using more transistors is another way of gaining performance as opposed to pure clockspeed.

So does it mean that if nVidia want to get their GPUs clocked on higher speeds, they must go 130 nm Low-K, if they go 110 nm they can only get them cheaper?

So the only ways to increase the performance of the NV4X line is putting into them more transistors (more pipes), or moving it to 130 Low-K and clock them higher?

 

[Dave Baumann]

So it will be a challenge for ATI to get the 16 pipe X800XL and even the 12 pipe X800 clocked at 400 MHz on 110nm, won't it?

Looking at the maximum speeds something is released at doesn't really tell you where the majority yield is, the higher end parts usually have the smaller volume; minimum clocks for the X700 range is 400MHz, which means that they expect the vast majority of parts to be hitting 400MHz or greater. If ATI were to announce another part with only one clockspeed in the range then they will only do it if they feel that the vast majority of those cores are hitting those speeds - much like anyone else would do the same.

So why don't they use 130 low-k on mid-range parts as well? Why do they gamble with 110 nm?

Its not a gamble. But you don't want to use 130nm in the mid end as cost is more important here - the more die you can get per wafer the better the price of the parts will be.

 

[Dave Baumann]

So the only ways to increase the performance of the NV4X line is putting into them more transistors (more pipes), or moving it to 130 Low-K and clock them higher?

Given the clocks they are running at the moment I should imagine there is still a fair bit of wriggle room on the current NV40 process with some silicon tweaks, so they probably can still release a clock boosted NV40 (in fact, the PCIe Ultras they sent out for SLI reviews were running at 437MHz, rather than 400MHz). They may just opt for a small boost now and go for a larger change at a later date, dependant on when they feel NV50 is scheduled.

 

[madshi]

Re: 0,13 Low-K vs 0,11 - What's going to be nV's answer to X

ATI is releasing one more High-End product line (R480) produced on the good, old 0,13 Low-K just as the previous flagships, 9800 and X800, and its mainstream product line

9800 was 0.15.

 

[hovz]

they hardly need an answer