[nelg]

I came across this today.

Originally Posted by Digitimes

ATI Technologies will begin sourcing flip-chip substrates from Taiwan-based Phoenix Precision Technology (PPT) for its 0.11-micron graphics chip, with first shipments of 500,000 units scheduled for next month, according to sources.

[Ostsol]

<-- Lots of that from me if their RV4xx part is .11 micron.

[jvd]

If this is true then i hope its the xt-pes that are .11. I really don't want to spend the money and find out a few days later there is an xt-pe- ultra coming out for the same price and has another 20% performance

[Chalnoth]

Given that ATI has in the past used the smaller processes for their low-end products, the suggestion that this will be a RV4xx is more likely. The 500,000 number may be an even better indication. I'm not sure it would make sense to run off 500,000 high-end chips in a first-run. If anybody has actual data on typical sales numbers these days, we might have a better idea.

[Fodder]

It's been known for some time that RV370 would be 0.11u, and that should be here real soon now. Is there any compelling reason to assume it's something else?

[Dave Baumann]

(Some pointlessness stripped from the thread)

Originally Posted by jvd

If this is true then i hope its the xt-pes that are .11.

Remember, 110nm is a "cost optimised" process, not a performance process. The major benefit from 110nm is a reduction in die size.

[anaqer]

Originally Posted by DaveBaumann

Remember, 110nm is a "cost optimised" process, not a performance process.

From what I see that would be just as good as getting another +50MHz, possibly even better.

[jvd]

Originally Posted by DaveBaumann

(Some pointlessness stripped from the thread)

Remember, 110nm is a "cost optimised" process, not a performance process. The major benefit from 110nm is a reduction in die size.

and they should run cooler which i like to hear .

[Fodder]

Originally Posted by anaqer

From what I see that would be just as good as getting another +50MHz, possibly even better.

But remember, TSMC don't offer Low-K on their 0.11u lines, so any benefit would be negated by lower achievable clocks. It seems like the sole purpose is to cram more budget chips per wafer.

[radar1200gs]

Originally Posted by jvd

and they should run cooler which i like to hear .

What makes you think the chips will run cooler??? It is literally the 0.13 micron process physically shrunk. Nothing other than physical feature size changes significantly.

The dies will be harder to cool because more heat is in less physical space than a 0.13 micron die.

There is no noticeable performance benefit, the only benefit is to ATi in the form of more dies per wafer which may possibly result in slightly lower prices for the consumer. Consumers otherwise don't benefit heat or performance wise.

[digitalwanderer]

Originally Posted by DaveBaumann

(Some pointlessness stripped from the thread)

Quality smart-ass remarks like that are not pointless, they're annoying. 8)

[Geo]

"Phoenix Precision Technology"?? How/where does a "substrate provider" fit in between foundry and fabless semi company?

[Joe DeFuria]

Haven't we all been though this before?

I thought it was fairly common "speculation" already that the first 0.11 chips are likely to be RV370...Which is most likely a shrunk RV350? This would possibly enable ATI to hit the "low end" discrete market with a 4 pipe DX9 chip? (Finally pushing the 9000/9200 chips out).

[CyFactor]

Originally Posted by Joe DeFuria

Haven't we all been though this before?

I thought it was fairly common "speculation" already that the first 0.11 chips are likely to be RV370...Which is most likely a shrunk RV350?(Finally pushing the 9000/9200 chips out).

If anything, RV370 is a shrunk RV380. It's native PCI-E. And the 9550 series will most likely be booting the 9000/9200 chips out of there in the AGP low-end. Just my opinion.

[Xmas]

Originally Posted by geo

"Phoenix Precision Technology"?? How/where does a "substrate provider" fit in between foundry and fabless semi company?

Packaging.
http://www.dayshine.com.tw/product/p...chFunctionID=3

[Mulciber]

Originally Posted by radar1200gs

What makes you think the chips will run cooler??? It is literally the 0.13 micron process physically shrunk. Nothing other than physical feature size changes significantly.

The dies will be harder to cool because more heat is in less physical space than a 0.13 micron die.

There is no noticeable performance benefit, the only benefit is to ATi in the form of more dies per wafer which may possibly result in slightly lower prices for the consumer. Consumers otherwise don't benefit heat or performance wise.

wouldn't going to .11u allow them the possibility to use slightly lower voltages for the same clock

[Dave Baumann]

Originally Posted by Joe DeFuria

I thought it was fairly common "speculation" already that the first 0.11 chips are likely to be RV370...Which is most likely a shrunk RV350?(Finally pushing the 9000/9200 chips out).

http://www.theinquirer.net/?article=16174

RV370 is ATI's first 0.11µ chip, will be called the X300, will sit in the sub $100 market. We'll get back to you on clock speeds and some more details when we have them, but but both of the cards are ATX based.

[Joe DeFuria]

Originally Posted by CyFactor

If anything, RV370 is a shrunk RV380. It's native PCI-E.

Sure....though RV380 (0.13 low-K I believe) would likely be designed for significantly higher clock-rates. So in effect, the RV370 would not be a "true shrink" of anything. You could consider it either:

1) RV350 core, but modified to support PCI-E native instead of AGP

or

2) RV380 core, but designed to be built on a non-low K process and with lower clocks. (This is NOT a "physical shrink" RV380, because low-k requires a different design than non-low K.)

No matter how you look at it, RV370 would appear to me to be "effectively" RV350 with native PCI-E. I expect similar (or if anything, possibly even lower) clocks on the RV370 relative to the RV350.

[WaltC]

It's important to recall that as processes dip below .13, current leakage problems accelerate (Intel's .09 Prescott being an extreme example of the principle.) So it's entirely possible that an .11 micron shrink could well affect cost parameters (more chips per wafer) without improving either voltage requirement or clock speed to any degree, but much depends on several other factors including transistor density, as well as the clockspeed and thermal targets you set. The targets for Prescott at .09 were exceedingly ambitious, so modest clock and voltage improvements for these .11 micron chips over their .13 counterparts can't be ruled out, I would think. Also, it's not uncommon for chip manufacturers to specify a given process for a chip when the reality is the process used is actually a hybrid--that is that only parts of the chip get shrunk with the rest remaining at the original process size. If that's the case here, then I'd expect that these .11 micron chips would achieve mainly cost benefits. Lots of variables to consider, in other words.

[Joe DeFuria]

Originally Posted by DaveBaumann

http://www.theinquirer.net/?article=16174

Right...that's what I thought.

[Joe DeFuria]

Originally Posted by Mulciber

wouldn't going to .11u allow them the possibility to use slightly lower voltages for the same clock

No.

0.11u should have the same power requirements per transistor. As mentioned above by someone, (simplifying) it's actually manufactured using 0.13 "rules", but they are just packed a bit closer together, allowing for a smaller die size all else being equal.

If anything 0.11 chips will be more difficult to clock higher than 0.13 chips, because you will have less surface area per transistor to dissipate heat. This is why 0.11 is mostly for cost reasons, not performance. If you have a chip that has very little power / heat issues (like RV350), this is an ideal candidate to "shrink" down to 0.11. You take advantage of the cost savings, will having little risk of running into significant heat issues. Still, it wouldn't surprise me to see at least some RV370 variants to be clocked lower than RV350.

[Geo]

Originally Posted by Xmas

Packaging.
http://www.dayshine.com.tw/product/p...chFunctionID=3

ThankS!

[Kaizer]

So in a quick summary I would know from this that there will be acceptably performing DX9-parts to be had for less than $100 in very little time?

Way to go for pushing DX9 into the definitive mainstream markets!

Kjetil

[Bjorn]

Originally Posted by Kaizer

So in a quick summary I would know from this that there will be acceptably performing DX9-parts to be had for less than $100 in very little time?

Way to go for pushing DX9 into the definitive mainstream markets!

Kjetil

I only hope that we won't see any 64 bit mem bus versions. That goes for all IHV's.

[Joe DeFuria]

Originally Posted by Kaizer

So in a quick summary I would know from this that there will be acceptably performing DX9-parts to be had for less than $100 in very little time?

That should be the case....assuming core clock speeds and memory bandwidth are at least "respectable."

It's likely that memory bandwidth will be significantly reduced from what we see with today's RV350 / RV360 parts to maintain the sub $100 pricepoints. However, this should not impact the shading performance of the cards too much. So, for example, there's a good chance that these sub $100 cards will be capable of running Half-Life2, for example, with all eye candy turned on. It will just be at a lower resolution and AA settings compared to say a 9600 or 9600 XT.