Interesting Info from COO of ATI

[Faeton]

2 development labs

It's sorta funny how people are saying that the R350 is just a bumped up R300.

Both ATI and Nvidia realizes that this is hardball now, and that a company can't get any credibility by doing something like that. ATI want's not only to get the lead but to hold on to it.

Just like Nvidia (and the big chipmakers now), ATI run dual labs (or more) to make these amazing products. It takes much longer than one year to design, test and make these high-tech chips. That's why with dual labs running parallel, they can accomplish it at such a breakneck speed.

The R300 was the first chip made from the new Marlborough lab in Massachusetts, formerly known as ArtX (the Gamecube guys). The R350 is designed at the Toronto lab and in the process of being taped out (not quite done yet, but close).

As far as specs go, I couldn't squeeze anything out of my friend (not surprising), but this is NOT just a OCed R300. It's going to build on it, but of course using the stuff learned from the R350. It wouldn't be surprising to see some quite different innovations due to the difference in labs producing these chips.

And as far as the margine goes, it's been said that a 9700 sold is equal upto 10 9000's sold.

 

[Username]

Re: 2 development labs

As far as specs go, I couldn't squeeze anything out of my friend (not surprising), but this is NOT just a OCed R300. It's going to build on it, but of course using the stuff learned from the R350.

Heh, you mean like a GeForce 4?

 

[olivier]

source rage3d.com


Question: What does ATI feel about DDR II memory?
Answer: Mentioned technology demonstration on Tech TV. Believes that DDR II is real but does not provide enough of a step-up in bandwidth compared to GDDR3. It does provide more headroom and they are ready to use when it ships in volume. ATI did not want to rely on DDR2 to supply the bandwidth needed for R300 so they went to a wider bus. For spring they will have 256bit bus and DDR II.

 

[tamattack]

R9500Pro / R9700 / R9700Pro all share Device ID 4E45. This is because they are the same chip and the driver(s) can behave in the same manner on each of them. The R9500 has Device ID 4E44, so ultimately it must be a little different from the rest of the R300 family else ATI would not have differentiated the Device ID.

R9500 has only 4 pixel pipelines, the others have all 8. From all accounts, it's an R300 with 4 pixel pipelines disabled.

 

[T2k]

Re: 2 development labs

It's sorta funny how people are saying that the R350 is just a bumped up R300.

But the real funny thing is that certain people forgot the official ATI text: every new architecture besed "less_than_a_year"-cycle. (Italic expression taken from last year stockholder info brochure)
And also pay attention what Orton said around last May: R350 will be a refresh part, not a new technology.

Both ATI and Nvidia realizes that this is hardball now, and that a company can't get any credibility by doing something like that. ATI want's not only to get the lead but to hold on to it.

...which can be achieved easily with a simple clock bump. Cheap, fast and still effective move - why do you expect more? Due to GFFX's 128bit, (as of today) an overclock to should be enough to hold the crown.
Imagine this: 4x0+MHz core w/ 256bit 450MHz DDR memory... ~29GB/s!

Just like Nvidia (and the big chipmakers now), ATI run dual labs (or more) to make these amazing products. It takes much longer than one year to design, test and make these high-tech chips. That's why with dual labs running parallel, they can accomplish it at such a breakneck speed.

The R300 was the first chip made from the new Marlborough lab in Massachusetts, formerly known as ArtX (the Gamecube guys). The R350 is designed at the Toronto lab and in the process of being taped out (not quite done yet, but close).

And? Nothing new... and it's not contradict for me.

(BTW, I heard the same thing from the tape-out. )

As far as specs go, I couldn't squeeze anything out of my friend (not surprising), but this is NOT just a OCed R300. It's going to build on it, but of course using the stuff learned from the R350.

From what?

Otherwise IMHO it should be based on R300. That's why it's called R350 and not R400.

It wouldn't be surprising to see some quite different innovations due to the difference in labs producing these chips.

It would be very-very suprising: do you seriously think they didn't work together? They are not working from same resources, sketches, everything?

C'mon...

Moreover this is contradiction in terms: if they really work separately then it's quite impossible to design any kind of (refresh-)part based on the last core...

And as far as the margine goes, it's been said that a 9700 sold is equal upto 10 9000's sold.

Dunno but it's possible.
Actually I think the sales ratio could be something similar...

PS: Actually I do love your prediction...

 

[T2k]

source rage3d.com


Question: What does ATI feel about DDR II memory?
Answer: Mentioned technology demonstration on Tech TV. Believes that DDR II is real but does not provide enough of a step-up in bandwidth compared to GDDR3. It does provide more headroom and they are ready to use when it ships in volume. ATI did not want to rely on DDR2 to supply the bandwidth needed for R300 so they went to a wider bus. For spring they will have 256bit bus and DDR II.

Hmmm... sounds cool...

But I still doubt it...

 

[Nagorak]

Of course, the question wasnt about investor sentiment, it was about returning to profitability. Since when has share price reflected market reality?

When is Microsoft's console segment going to BEGIN to show profitability, never mind return to it?

 

[Username]

When is Microsoft's console segment going to BEGIN to show profitability, never mind return to it?

Who cares? What does that have to do with an ATI conference call? Lets not get too pathetic with our fanatical devotion to a company, we already have enough of those floating around.

 

[Doomtrooper]

A pot calling the kettle black username...the 1st one to Nvidias defence...with childish cheap shots and little to no content posts

 

[Mr.huang]

I picked up some info yesterday night... in a club (if you were over there you know what I'm talking about ): it seems R350 is just an overclocked R300 with no ddr2.
Dunno but previously I expected ddr2 - well, we'll see...

This would be in line with what I have suspected all along--what's the point in making a significant jump in the spring, and then another significant jump in late summer--when you don't have to? Or, as someone else suggested, R350 could be a value part for notebooks (that might or might not fit in with what was said about R350.) I guess I have to concede at this point that we do have some decent evidence that "R350" exists (not that I doubted you, HB, I was just looking for some confirmation from ATI and this would *appear* to be that--except for the nature of the R350, of course.)

But let's say yields have improved and pipelines get tweaked, etc., and you come up with a better cooling solution (not like nv30's--just a lot more efficient than at present)--I don't think 400MHz-425MHz is out of the question for R300. The whole point of the 256-bit bus is to be able to put off the move to DDR II until you have the GPU power to fuel such a bus and you've given the II technology a bit of time to mature. If they stick to DDR but move to 2.2-2.3ns DDR, the combination would significantly enhance current R300 performance (enough to easily steal nv30's thunder--assuming nv30 actually hits 500MHz in quantity), and then save your .13 micron ~500MHz R400 beastie, coupled with your ~40 gig/sec DDRII/256-bit bus for the late-summer/fall release--at which time that product will easily keep up with and probably surpass whatever nVidia might wish to do with its nv30 fall refresh.

But I'm not placing bets... ATI surprised the heck out of me with R300--won't be making glib projections about them again...

maybe ATI would use infineon's Graphics RAM , if so, i willl be very impressive, but i am curious, why infineon can put the DDR speed more closer to 500MHZ.

http://www.infineon.com/cgi/ecrm.dll/ecrm/scripts/prod_ov.jsp?oid=13531&cat_oid=-8007

 

[Nagorak]

When is Microsoft's console segment going to BEGIN to show profitability, never mind return to it?

Who cares? What does that have to do with an ATI conference call? Lets not get too pathetic with our fanatical devotion to a company, we already have enough of those floating around.

 

[RussSchultz]

Re: 2 development labs

And as far as the margine goes, it's been said that a 9700 sold is equal upto 10 9000's sold.

I doubt that statement is true.

The R300 likely costs around 30 dollars to make; they likely sell it for 60. (for 50% margin). This means $30 of gross profit. The 9000 would have to mean 3 dollars of gross profit.

You're suggesting that the 9000 costs $9 to make, and they sell it for $12 (for a 25% margin). (If you inflate the 9000 cost too much more, the margin on the 9000 falls to a point where no company wants to participate)

I seriously doubt the 9000 cost $9 or less to make, knowing the cost to manufacture our much smaller chip.

 

[Bambers]

Hmmm... sounds cool...

But I still doubt it...

It came from a summary of the Q&As session of david ortons talk at CSFB conference

link:

http://www.rage3d.com/articles/csfbnotes/

Theres also a link to where the full thing can be listned to at the top of that.

 

[jandar]

Re: 2 development labs

And as far as the margine goes, it's been said that a 9700 sold is equal upto 10 9000's sold.

I doubt that statement is true.

The R300 likely costs around 30 dollars to make; they likely sell it for 60. (for 50% margin). This means $30 of gross profit. The 9000 would have to mean 3 dollars of gross profit.

You're suggesting that the 9000 costs $9 to make, and they sell it for $12 (for a 25% margin). (If you inflate the 9000 cost too much more, the margin on the 9000 falls to a point where no company wants to participate)

I seriously doubt the 9000 cost $9 or less to make, knowing the cost to manufacture our much smaller chip.

He might mean that it is 1 9700 GPU sold for every 10 9000 GPUs sold, which to me does make sense.

Of course, Ati might be willing to take little profit on the 9000 cards simply to flood the marketplace with DX 8.1 chips so more and more ISVs are willing to target those, helping Ati in the end like the GFMX series and nVidia.

But thats just a wild off the wall, lack of caffiene guess.

 

[antlers]

Re: 2 development labs

And as far as the margine goes, it's been said that a 9700 sold is equal upto 10 9000's sold.

I doubt that statement is true.

The R300 likely costs around 30 dollars to make; they likely sell it for 60. (for 50% margin). This means $30 of gross profit. The 9000 would have to mean 3 dollars of gross profit.

You're suggesting that the 9000 costs $9 to make, and they sell it for $12 (for a 25% margin). (If you inflate the 9000 cost too much more, the margin on the 9000 falls to a point where no company wants to participate)

I seriously doubt the 9000 cost $9 or less to make, knowing the cost to manufacture our much smaller chip.

You're implying, then, that the margin on the 9000 is probably much less than $3? Which would mean that the gross on each 9700 would be far more than 10x as much?

 

[RussSchultz]

I'm implying they aren't making 10x the money per 9700 sold vs. 9000 because I know their chip cost isn't less than $9, and I suspect they aren't taking less than 25% margin on any product they have. (High tech investors don't like numbers like that AT ALL)

 

[ram]

Re: 2 development labs

The R300 likely costs around 30 dollars to make; they likely sell it for 60. (for 50% margin).

Wow. Some months ago, IIRC back in June/July, there have been estimations from PVR that a R300 costs over 100 USD to make in 0.15u. Of course prices have come down since then, but by more than 60% in two quarters??

 

[RussSchultz]

In one of these threads was a my guestimate cost of the NV30, but here it is in short:

1 wafer = $3000
131 chips per wafer (from pictures of the wafer)
80% yield (generally acceptable OK yield)

Plus packaging puts it at about 30-35$. (I'm guessing their fancy packaging doesn't cost more than 7$).

The R300 is using a smaller design, on a larger process. I believe the numbers end up being approximately a wash, so I'm calling them essentially equivalent.

But pick any number, and its very tough to have 1 chip make 10x the gross margin as another if they're roughly the same product, and cost structure.

 

[antlers]

Well, ATI is a board vendor as well as a chip vendor. From your analysis it sounds like it would be much easier for a complete board to have 10 x the gross margin. Is it possible he was talking about margin per board?

 

[RussSchultz]

Work out the math, it gets even worse as your costs go up. The only way you can get the 10x equation and still make happy margins on both parts is when one costs about 5x more than the other.