R360 != 0.13 process ?

[Seiko]

Having read a number of threads/articles etc I suspect that the R360 will still use the 0.15 process but wanted to understand why ATI couldn't have shifted it to 0.13 and reaped the rewards of a higher clocked 9800? If the new R360 is indeed 0.15 based then aren't we really looking at a very small tweak here or there? If so, this really is letting Nvidia catch up as the NV40 looks like it will debut before the R4xx and thus put Nvidia right back on top?
In short, just what does it take to get an existing chip onto a smaller process and dont you think ATI have squanded their R300/R350 technology lead with these very minor speed bumps?

Thx inadvance,

Seiko

 

[Joe DeFuria]

Having read a number of threads/articles etc I suspect that the R360 will still use the 0.15 process...

That is true with as much certainty as we can have short of an official announcement.

but wanted to understand why ATI couldn't have shifted it to 0.13 and reaped the rewards of a higher clocked 9800?

The short answer is cost and higher risk of delays to market. It takes a lot of cash and resources to "shift" a product from one fab process to another.

If the new R360 is indeed 0.15 based then aren't we really looking at a very small tweak here or there?

Yes. Likely just a "faster" R350, with perhaps a tweak here or there.
This is also what we are expecting of nVidia's NV38.

If so, this really is letting Nvidia catch up as the NV40 looks like it will debut before the R4xx and thus put Nvidia right back on top?

That all depends on how successful each company is in getting their parts in volume production. R360 will certainly be here well before the NV40. One would guess that NV40 has an advantage vs. R420 in time to market, but it's way to early to tell. A lot will be known once "first silicon" makes it back from the fabs on each of these parts.

In short, just what does it take to get an existing chip onto a smaller process

It's almost like designing a brand new chip. You have do re-do much of the most expensive and time consuming type of work.

and dont you think ATI have squanded their R300/R350 technology lead with these very minor speed bumps?

No, they just made a business decision. If you keep churning out brand new cores every year or so, you don't give yourself time to recoup R&D. The short of it is, the decision that ATI made is one that they feel will be most profitable.

Everyone thought a year ago that Nvidia had a clear advantage of NV30 being on 0.13, while R300 was still on 0.15. we know how that turned out.

The best possible outcome for ATI would be the following:

1) R360 and NV38 come out about the same time, and R360 is perceived as the best part.

2) R420 and NV40 come out at about the same time, and R420 is perceived as the best part.

That situation would give ATI the most time to recoup dev costs, while still maintaining "technology leadership." It also gives R420 better chance at utilizing more advanced / stable 0.13 process (lower cost), and availability of more advanced memory technology, etc.

 

[Seiko]

The short answer is cost and higher risk of delays to market. It takes a lot of cash and resources to "shift" a product from one fab process to another.

The problem I have with that Joe is that ATI have already tweaked the basic R300 twice now. That alone is costing time and resource. Add to the fact they then created the RV350 on 0.13 and it seems ATI is spoilt for choice.

Lets suppose they had opted not to produce the RV350 (9600pro etc) and instead created the R350 on 0.13. Although risky I honestly believe they could then have dropped the older R300 cores into the $200 price range and killed NVidia off completely. Sure margins would be small on the R300 but by simply flooding the market they'd surely recoup due to quantity. I'd love to see some internal figures on this theory but I think it may have panned out albeit risky?

The thought of this potential ATI line up makes me drool

ATI 9800 PRO/Non Pro using 0.13 process Top end

ATI 9700 PRO/Non Pro using 0.15 process Mid range

ATI 9500 PRO/Non Pro using 0.15 process Bottom end.

Although the ATI 9500 couldn't really drop into the sub $100 market place it certainly would have come close and helped grey the DX8/DX9 issue ATI currently have with regard to not having a DX9 chip in the lowest segment.

I know this is all probably impossible as the margins wouldn't have panned out but I'd love to see just how close it was! Imagine a 500/500 ATI 9800 Pro dominating the benchies and the 9700 Pro controlling the mainstream section.

Now with only 3 cards (6 including the none pro variations) ATI have simplified their protfolio, reduced these expensive tweaks and could then concentrate on their next big chip. When that chip arrives the whole section moves down a notch. At the momemt its extemely painful for a customer trying to note just what to purchase. With the 9600Pros not really cutting it compared to the older top end GF4s or in future DX9 titles its a little difficult to encourage people on anything equal to or higher than a GF3 to upgrade. Sure you may gain an extra level of AF or even FSAA but it will hardly change your gaming experience. Now the 9700Pro and above do. These cards allow you to switch things to the max and enjoy every game out there but unfortunately for a massive $300 it's hardly mainstream?

Ah well, if only I had a billion to play with

 

[Joe DeFuria]

The problem I have with that Joe is that ATI have already tweaked the basic R300 twice now.

Including the R360, correct.

That alone is costing time and resource.

But it's minimal compared to a new design.

Add to the fact they then created the RV350 on 0.13 and it seems ATI is spoilt for choice.

Don't get what you mean here. ATI chose to get acquainted with the 0.13 process with a "modest size" part.

Lets suppose they had opted not to produce the RV350 (9600pro etc) and instead created the R350 on 0.13.

Then we'd probably still be waiting for the R350 to ship. In which case not only would nVidia probably be dominating the mid-end with the NV-31, they would have the high-end with the NV35 against the R300.

Although risky I honestly believe they could then have dropped the older R300 cores into the $200 price range and killed NVidia off completely. Sure margins would be small on the R300 but by simply flooding the market they'd surely recoup due to quantity.

Nothing is for sure.

The thought of this potential ATI line up makes me drool

ATI 9800 PRO/Non Pro using 0.13 process Top end

I'd like that too...but we may still be waiting for the part!

ATI 9700 PRO/Non Pro using 0.15 process Mid range

ATI 9500 PRO/Non Pro using 0.15 process Bottom end.

Of course I'd like to see that too. Again, the problem is cost. I trust ATI knows their financials, fab capacities and yields better than either of us.

I know this is all probably impossible as the margins wouldn't have panned out but I'd love to see just how close it was!

Yeah, it would be interesting to know!

[quote...]With the 9600Pros not really cutting it compared to the older top end GF4s or in future DX9 titles its a little difficult to encourage people on anything equal to or higher than a GF3 to upgrade. [/quote]

From a Geforce3? A 9600 Pro is a very decent upgrade IMO. Note that you are talking about upgrading what was once top of the line (GeForce3), with a current "mid-range" product.

For a A GeForce2 MX or a GeForce3 Ti 200...(The mid-range chips of that era)...a 9600 Pro is a GREAT upgrade.

Ah well, if only I had a billion to play with

Heh...just keep investing in ATI stock.

 

[Natoma]

ATI Confirms R420 Made Using 0.13 micron

http://www.xbitlabs.com/news/video/display/20030812141326.html

 

[Joe DeFuria]

ATI Confirms R420 Made Using 0.13 micron

http://www.xbitlabs.com/news/video/display/20030812141326.html

So, tell us something we don't know.

 

[Natoma]

Given that the 420 is going to be .13, and is scheduled for Q1 04, it doesn't seem likely that the 360 is going to be .15, when it's supposed to be released roughly Q4 03. Granted this is a guess. However, considering the time frames between the two releases, an educated one.

 

[RussSchultz]

I'm not sure what time correlation has to do with process decision.

 

[Natoma]

Given that .13 production has to be ramped up heavily for the 420's product release, it would seem an easy decision to include the 360 in that ramp. The 360 is a fairly complex chip, and ATI has no experience outside of the RV350 with .13. My guess is that given the release timeframe of the 360, they will opt to have the 360 at .13.

It's not a major product release, merely a respin of the current core to achieve higher clock speeds. Sounds like a good enough guinea pig to me for testing purposes with a complex core.

 

[PatrickL]

If R360 is just a R350 respin still at 0.15, what would be the performance gap between the 2 cards? Around 10 % faster ?

 

[Joe DeFuria]

Given that the 420 is going to be .13, and is scheduled for Q1 04, it doesn't seem likely that the 360 is going to be .15, when it's supposed to be released roughly Q4 03. Granted this is a guess. However, considering the time frames between the two releases, an educated one.

??

Well, first of all, I expect the usual 6 month time frame between product launches. Sept. for R360, and Feb/March for R420.

Second, what does this have to do with anything?

 

[surfhurleydude]

If R360 is just a R350 respin still at 0.15, what would be the performance gap between the 2 cards? Around 10 % faster ?

Based off of credible sources, it's supposed to be around 23% faster than the Radeon 9800 Pro, at worst case scenarios for the R360.

 

[Joe DeFuria]

Given that .13 production has to be ramped up heavily for the 420's product release, it would seem an easy decision to include the 360 in that ramp.

How so? Why is it "easier" to develop and tape out a second completely new chip, vs. just respinning an existing one?

The 360 is a fairly complex chip,

Agreed.

and ATI has no experience outside of the RV350 with .13. My guess is that given the release timeframe of the 360, they will opt to have the 360 at .13.

You just can't "opt" to have one chip at a given process. You make that decision very early on, likely about 18 months before you get the first product. Your whole development cycle is based on that initial decision.

It's not a major product release, merely a respin of the current core to achieve higher clock speeds.

Which is precisely why it's a major hassle to try this on 0.13. There's no such thing as a "respin" on a new process. It's an entirely new product that would have to undergo a nearly complete development cycle.

Sounds like a good enough guinea pig to me for testing purposes with a complex core.

I understand what you're saying...but that's just not the way it works. You really have to "redesign" a chip for a new process. The best guinea-pigs for a new process, are less complex designs.

 

[Joe DeFuria]

If R360 is just a R350 respin still at 0.15, what would be the performance gap between the 2 cards? Around 10 % faster ?

What's the performance difference between the R300 and R350?

 

[PatrickL]

I was also considering that we had almost 9 months between R300 and R350 but only 3 months between R350 and R360 and so less time to optimize a lot of things.

 

[surfhurleydude]

What's the performance difference between the R300 and R350?

I made this post at NVnews. It'll probably be helpful here too:

SurfMark Benchmark.

Radeon 9700 Pro: 50 FPS
Radeon 420: 100 FPS

Wow, so far, we have concluded that the R420 is 2x the speed of the Radeon 9700 Pro, and therefore 100% faster!

However, some more things are thrown into the mix, such as Radeon 9800 Pro and R360. We don't know what numbers they score in surfmark, but we do know how much faster or slower they are than the other cards.

Radeon 9800 Pro = 25% faster than Radeon 9700 pro
R360 = 30% slower than R420.

With this in mind, we will use simple mathematical formulas to figure out what they score in SurfMark.

Radeon 9800 Pro = (Radeon 9700 Pro = 50 FPS) * .25 + (Radeon 9700 Pro) = 62.5 FPS!

Now, to double check this, we will use another simple mathematical formula to find out what % the Radeon 9800 Pro is of the Radeon 9700 Pro! We will divide the frame rate scored by the Radeon 9800 Pro by frame rate scored by the Radeon 9700 Pro.

62.5 FPS/50 = 1.25 or 125%. Thus, the Radeon 9800 Pro is 125% of the Radeon 9700 Pro or, 25% faster.

So now we have 3 of the 4. We know that the R420 is 130% of the R360, so we will make an algebraic equation here.

100/1.3 = X

Well, by dividing 1.3 we get 76.9

So we have all 4.

Radeon 9700 Pro = 50 FPS
Radeon 9800 Pro = 62.5 FPS
R350 = 76.9 FPS
R420 = 100 FPS

So from here it is easy.

Based off of these facts, we can roughly expect (in BEST case scenarios)...

Radeon 9800 Pro to be 25% faster than the 9700 Pro.
R360 to be 23% faster than the Radeon 9800 Pro.
And of course, R420 to be 30% faster than the R360.

So essentially what we have here is that the R360 is going to be the midpoint between the Radeon 9700 Pro and the R420. After mathematically proving this, I hope I am wrong here and MuFu's facts are incorrect, as A LOT of people are going to be disappointed in the R420's performance (even though they shouldn't be, I mean hey 30% is a nice jump) mainly because the Radeon 9700 Pro had no "Radeon 9800 Pros and R360s" in between it and the Radeon 8500.

 

[Joe DeFuria]

I was also considering that we had almost 9 months between R300 and R350 but only 3 months between R350 and R360 and so less time to optimize a lot of things.

?? We had roughly 7 months between the R300 and R350. And assuming R360 launches in Sept, it will be 6+ months between R350 and R360.

 

[Prometheus]

According to Anands latest benchmarks(Quake3,UT2003,JKII 4xfsaa/8xQaniso) Radeon 9800pro is about 15% faster than Radeon 9700pro.

 

[Natoma]

I was also considering that we had almost 9 months between R300 and R350 but only 3 months between R350 and R360 and so less time to optimize a lot of things.

?? We had roughly 7 months between the R300 and R350. And assuming R360 launches in Sept, it will be 6+ months between R350 and R360.

Eh? Wasn't the R350 released in April/May?

 

[Joe DeFuria]

Radeon 9700 Pro: 50 FPS
Radeon 420: 100 FPS

What is surfmark, and where did that R420 number come from?

[snip the "math"]

I dont' get the point of that whole exercise. We know that the 9800 is roughly 15-20% faster based on clock-speeds alone. (Actual increase will depend on specific situation.)

My point was, the R350 was "only 15-20%" faster, and it's a completely viable product.

The R360 might "only" be 15-20% faster than the R350, and it will also be a completley viable product.

Some people seem to be spoiled by the "step change" in performance that was ushered in essentially by the advent of the 256bit bus. 100% performance increases in one product cycle is the exception...not the rule.