Interesting Info from COO of ATI

Russ, you're way off on the yield for these particular chips. A01 yield is under 20% as far as I know. I don't know what the yield on A02 is. Of course the eventual production target will be 80+%, but they're nowhere near that at the moment.

 

I don't know the current yields, but I was only estimating with a rule of thumb value.

 

T2k....

I think you are also overlooking the fact that the R350 is not called the R300A, or R301 either.. which is what a tweaked for Overclocking chip would be..

There is a lot more to the R350 than just some faster core clocks..

But, i guess you will just have to wait and see for yourself.. eh?

 

indeed.

 

I thought that the former ArtX/GameCube team, the one that also did R300, was based on the west coast. (team west) while the team that made R200, and is working on R400 is team east. I guess I have it all wrong, please clear this up for me if so.


Also, Nvidia is known to have at least four design teams. ATI has at least two teams, since they acquired ArtX. is that true, ATI only has two teams,
while Nvidia has four?

I guess Nvidia had at least two teams, then in late 2000 or early 2001 when they got 3Dfx and GigaPixel, they expanded to four. sound about right?

btw, where are the majority of the Real3D guys?

 

I agree with these statements.

it is indeed a contradiction. GF256 ==> GF2gts is not that different than
GF3 == > GF4Ti - that is what i expect from R300 ==> R350.

R350 would most certainly not be just a clock increased R300 with DDR2, or otherwise why would they have to produce another chip, another tapeout? the R350 is a new chip (not totally diff arch tho) the 350 designation indicates it's a refresh. ATI's first real refresh that's a performance and core enhancement.

Also, R350 is not the same as RV350. the '350 is not going to be a streamlined '300, like the RV250 is a streamlined R200. R350 is going to be a beefed UP R300, which should include core & memory speed increase, DDR2, on top of core enhancments/improvments & additions.

R350 will neither be a speed bump nor a totally new generation.

 

http://www.digitimes.com/NewsShow/Article.asp?datePublish=2002/12/06&pages=02&seq=2

Taiwan Semiconductor Manufacturing Company’s (TSMC) 0.13-micron processes at Fab 6 have reportedly reached a yield rate of 70%. Fab 6’s biggest customer, Motorola, has seen its 0.13-micron DSP (digital signal processor) production reach 90% yield rates, said sources.

Earlier this year problems in 0.13-micron processing equipment at Fab 6 reportedly resulted in a loss of 3,000 wafers. Yield rates were improved to 50% by September, but in November problems were discovered in the transfer of wafers from one process to the next. With these problems recently resolved, yield rates have broken 70%.

The breakthrough in 0.13-micron yield rates is expected to increase TSMC’s gross margins and pull it further ahead of competitors, sources said. Customers like ATI Technologies and Altera have been eager to advance from 0.15-micron technology up to the 0.13-micron node. Now that TSMC has achieved higher yield rates, the global foundry leader is also reportedly in talks with major customers about purchasing by wafer rather than by die.

TSMC has vigorously expanded production at Fab 6 since 2001 from a monthly capacity of 5,000 wafers to the current 15,000 to 20,000 wafers. Besides Motorola’s DSPs, Qualcomm has chips in 0.13-micron volume production and Nvidia’s NV30 graphics chip has begun small-volume test production at Fab 6.

 

Mmm, but you'd expect their DSP to have less than 125 million transistors! I wonder how high the yield rates for the new stepping of NV30 will be?

 

This is my understanding as well... I'm not sure where the previous post about ArtX being the marlborough team came from. ATI has had the Marlborough (east coast) offices for quite some time... It is the offices that ATI Research (demos/software/etc) are located among others. (not sure if "ATI Research" also covers the west coast offices... I think someone at ATI once mentioned to me that all ATI's US divisions fell under the ATI Research moniker, but I never clarified it so I might not have heard that right.)

the former ArtX teams offices are silicon valley.

Also, there is no "ArtX team" at ATI. When ATI acquired ArtX they mixed their design teams so it is silly to say ArtX designed the R300. Certainly people from that team worked on it, but it more than just that. ArtX people are mixed with other ATI design teams as well.

ATI designed the chip. Former ArtX people are just a part of the team.

The number of design teams is all in how you define what the design team is. ATI has seperate divisions for all the different markets they are in... so it isn't like they have only two teams working on two things at any given time. The mobile division works on their products, integrated on theirs, desktop, workstation, etc.etc. There is overlap in design teams and personel, but I think where the "two design teams" bit for ATI comes from is that there are two primary locations where ATI has teams doing design work.... that doesn't mean they only ever work on two products.

 

I think its important to differeniate between product and design teams. I think the reference to two teams is that there are likely to be two design/architecture teams concurrently working on high end architectural designs, staggered for when they are expected to be introduced. Further product teams take various elements of those high end designs and alter them to fit in the various different markets.

It seems that NVIDIA have been operating something like this for some time and we are starting to see a similar appraoch in ATI's product output.

 

Did I? Why is that supposed to be called R300A or R301 only?

BTW, I love your idea...

 

well basicly, you seem to have been saying that R350 is just going to be a high clocked, high memory configuration of the R300 that's tweaked. that sounds more like "R300A" or "R301" - the R350 is called R350 no doubt because it is more than just a speed bumped R300 - R350 is going to have changes/improvments/additions to the R300 core. what those are, no one outside ATI knows for certain yet.

 

Hehe, I love your last sentence...

 

but there's little doubt that R350 is a new core over R300. as much as GeForce2 GTS was over GeForce256, or GeForce4 Ti was over GeForce 3.

 

I'll be getting a R350 almost for certain. won't even wait for NV35, since that should have originally been out by now, if NV30 had been the spring 2002 product. with 1 delay, NV35 would be spring 2003 if NV30 had been fall 2002, but NV30 was pushed back a 2nd time, so NV35 will be out in late summer at the soonest. just in time to get pummled by R400

 

I think I'll wait for the R400. I work in 1600: it's far enough. I play in 1600 w/ 4xQAA + 8xQAF: it's far enough.

 

Preview of the Fire XG1.. ATI R300 for the Professional crowd.

Perhaps in the wrong thread but the card supports AGP Pro (no need for external power connector) and with 256MB of 3.3ns RAM (lower clockspeed Infineon DDR RAM). Also a daughter which is detachable/upgradeable(?). Dont know what that is for exactly...also advertised is the dual 10-bit per channel 400MHz RAMDAC.

http://www.darkcrow.co.kr/Preview/Preview_encontent.asp?board_idx=69

 

I had thought that all R300's and NV30's were supposed to have 10-bit DAC's?

 

That 'feature' seems to be missing from the drivers currently available from ATi unless I am reading it all wrong and it is already 'enabled'. Boy I hope I don't sound like an idiot (not hard to do).

 

A 10-bit DAC shouldn't be the result of any driver-level feature, or anything on the programming side. All that would be necessary is a front-buffer that has at least 10 bits per channel color (which would likely require at least a 64-bit back buffer).