What problems are ATi running into with 110nm?

[Karma Police]

What problems are ATi running into when transitioning from 130nm to 110nm? Specifically, why would the X700 be released at 475Mhz while the 6600GT, which is on the same transistor process, be released at 500Mhz & run cooler as well?

I guess I am curious what difficulties a company would run into that would typically be fixed with a re-spin of the same architechure.

 

[digitalwanderer]

None, it's going just swimmingly...thanks for asking though.

 

[Dave Baumann]

Yields at for particular clockspeeds are not just dictated by the process and design, but the voltage you choose to run at. Alternatively, you may consider that the lower end designs is where you get more volume and what the respective clock speeds of the lower end of the boards are.

 

[Karma Police]

I thought ATi had difficulties that limited them to 475, while the 6600GT was at 500, is that not correct? The X700 also seems to run hotter, soooooo........... why would this be? They're both at .11.

Are all 110nm not built the same?

 

[Karma Police]

Yields at for particular clockspeeds are not just dictated by the process and design, but the voltage you choose to run at. Alternatively, you may consider that the lower end designs is where you get more volume and what the respective clock speeds of the lower end of the boards are.

But if it's only the voltage that is limiting the core, why not just change the voltage & bump up the core? I understand what you are saying about the lower-end designs, though.

What I am more curious about is the R420 move to .11 with the R480 as the result. What problems could ATi run into?

 

[Dave Baumann]

If I were to take a stab then I would say don't expect anything faster, in fact, expect something lower in performance. If this R480 is what is purported to be then I would epxect that this will merely be a way of offering 16 pipelines at a lower price (but lower performance than the high end as well).

 

[Pete]

Are you talking about lowering prices aside from a die shrink? And when you say lower performance, are you talking about something other than a clock speed reduction? Will a 16-pipe R480 cause ATi to drop the X800P or XT/PE from its lineup, or both? Will the R480 replace the R420 altogether, with 16- and 12-pipe variants?

Lot of possiblities if the R480 is a "cheaper" R420, but I'd think ATi wouldn't want two 16-pipe chips produced at the same time, no? Or is R480 PCIe...?

 

[Karma Police]

Yeah, this is why I don't seem to think the R480 will be any better thatn the R420. The 130nm low-k tech tops out at ~540Mhz, while the 110nm regular tech tops out at ~500Mhz. When Ati goes from R420 to R480 the top speed will be ~500Mhz but with the same feature set, right? So, unless they put higher speed memory (700Mhz DDR3, etc), it would be only as fast as a X800XT non-PE, which also has a core of 500Mhz, is that right?

 

[Bjorn]

while the 110nm regular tech tops out at ~500Mhz.

I wouldn't say that it tops out at 500 MHz. I'm basing that on the rather good overclocks from the 6600 GT.

 

[Killer-Kris]

Yields at for particular clockspeeds are not just dictated by the process and design, but the voltage you choose to run at. Alternatively, you may consider that the lower end designs is where you get more volume and what the respective clock speeds of the lower end of the boards are.

But if it's only the voltage that is limiting the core, why not just change the voltage & bump up the core? I understand what you are saying about the lower-end designs, though.

If I understand the point that Dave is trying to make it's not that simple. ATI wants to ship as many identical chips as possible. So with that in mind if 98% of all the dies are capable of running at the specified speed and voltage, or they can chose say 80% able to run at 500MHz at a higher voltage. (I'm just making up numbers out of thin air to make a point)

So what happens now is that the business people sit down with the info from the engineers, calculate how many sales will be lost to nvidia for having lower speed, and then compare that to how many fewer dies they will have to toss out. They then pick the option that they feel will provide the greatest $ return.

With that in mind ATI could probably up the voltage and bump the speed up in order to perform on par w\ the 6600, but it would cost to much to be worth it.

 

[chavvdarrr]

if X700 runs on lower voltage , then 6600 would be hotter with bigger heatsink. But we have the opposite - X700 is hotter.... so either ATi rushed X700 too early, either they have 0.11 related problems.
Consider also that overclocking X700 seems hard, chips barely able to pass 500, while 6600 goes much higher

 

[PatrickL]

Yes I think ATI would have wanted to have another respin before releasing that chip. The fact they had too much things going on at same time being Nintendo chip, Xbox Chip, regular 3d chips, Hdtv products an so on seems to have a bit hard to overcome with the ressources they had.
But, if work on Xbox2 chip is really done, they may have more engeering ressources free to work back on regulars 3d chips. The next six months should tell us how much ressources were diverted on the R500 contract.

 

[Richthofen]

what i think happend is that they went from a 9600XT on 0.13 low-k to a X700XT on 0.11 without low-k.
The X700 is still based on the principle RV3xx design as the R4xx is too.

I think they have a problem with 0.11. The x700xt runs hotter than the GF6600GT and does not overclock that much. It has also a lower default clock speed. Missing low-k on 0.11 is the problem if you ask me.

 

[PatrickL]

Could be, we will see that after the the A2 core ?

 

[Evildeus]

It's already there A12 :|

 

[gokickrocks]

if X700 runs on lower voltage , then 6600 would be hotter with bigger heatsink. But we have the opposite - X700 is hotter.... so either ATi rushed X700 too early, either they have 0.11 related problems.
Consider also that overclocking X700 seems hard, chips barely able to pass 500, while 6600 goes much higher

or it could just be that the 6600 has a bigger heatsink, which may or may not be paired with a larger die size compared to the x700, which can account for it running cooler (i really dont know as i have yet to see either board)

 

[chavvdarrr]

according to reviews i read X700's cooler is bigger&heavier

 

[{Sniping}Waste]

What is the most effecent HSF. I have used HSF that are bigger and heavier but a smaller and lighter one does a better job. Both were AL HSFs.

 

[bigz]

according to reviews i read X700's cooler is bigger&heavier

and a lot noisier