Not surprising: nVidia Market share declining

[ZoinKs!]

I have a hard time believing that ati will bring out new a performance mainstream part at .15; that seems like a step backwards. Especially since they've gotten good yields with rv350. On the other hand, .15 at umc should be cheap. But they'd probably have power and heat issues.

It makes sense for ati to eventually move to a smaller process with a relatively simple mainstream part. But it seems too soon for that.

I'm assuming the $200 r9700pro equivalent chip is the rv360. I also assume the rv360 is a respin or refresh of rv350. Are these assumptions correct? If not, what parts are wrong?

 

[PatrickL]

Hum RV360? i knew of course for R360 but we will get a RV part too for the mainstream segment ?

 

[WaltC]

IMO "DX9" when applied to the 5200 is little more than a marketing bullet, and ought to be treated as such. The 5200 is a pretty rotten "DX9" part, very nearly as rotten a DX9 part as ATi's low-end DX8 parts could be said to be "DX9" parts (simply because they'll include DX9-compatible drivers.)

I mean, nVidia's high-end nv35 is a fairly poor "DX9" part when contrasted to R350, and the 5200 is just much worse.

To me the phrase "low-end DX9" is an oxymoron of sorts....

 

[ZoinKs!]

Yeah, the rv360 is supposed to be a faster, cheaper, bug fixed respin of rv350. So there shouldn't be much new about it. But there was also some talk of an rv380 somewhere further down the line.

 

[WaltC]

I have a hard time believing that ati will bring out new a performance mainstream part at .15; that seems like a step backwards. Especially since they've gotten good yields with rv350. On the other hand, .15 at umc should be cheap. But they'd probably have power and heat issues.

It makes sense for ati to eventually move to a smaller process with a relatively simple mainstream part. But it seems too soon for that.

I'm assuming the $200 r9700pro equivalent chip is the rv360. I also assume the rv360 is a respin or refresh of rv350. Are these assumptions correct? If not, what parts are wrong?

nVidia's had plenty of heat and power issues with its .13 micron chips for the last year--not to mention yield issues. I think it's more important to look at processes in terms of economics as opposed to power consumption, heat, and performance. It could well be that a .13 micron R350--right now--might provide considerably lower yields, run hotter and consume more power than at .15 microns. I think that by and large the generally assumed benefits for die shrinks were apparent for .25 to .18 and then for .18 to .15. Things get dicier going to .13, and even more problematic going to .09, because of physical engineering/manufacturing barriers of a type not a problem at the larger processes.

I think ATi is looking at moving to .13 completely when the economics get right, and that performance, heat, and power consumption are less the concern than the economics of it at the moment. Much depends on the architecture and the circuit design, and so while RV350 is doable economically at .13, R350 isn't. Yet.

Meanwhile, comparisons between R350 and nv35 are positive proof that a process does not a vpu make...

 

[Fred da Roza]

I have a hard time believing that ati will bring out new a performance mainstream part at .15; that seems like a step backwards. Especially since they've gotten good yields with rv350. On the other hand, .15 at umc should be cheap. But they'd probably have power and heat issues.

It makes sense for ati to eventually move to a smaller process with a relatively simple mainstream part. But it seems too soon for that.

I'm assuming the $200 r9700pro equivalent chip is the rv360. I also assume the rv360 is a respin or refresh of rv350. Are these assumptions correct? If not, what parts are wrong?

I assume your definition of performance mainstream is 5200. I can't imagine a 9600 manufactured at .13 being offered at 5200 prices at this time. DaveB did say soon. The 5200 is manufactured at .15 and nobody said it was a step backwards.

Why would a .15, 9600 have power and heat issues. The .13, 9600 seems to have a lot of headroom and a .15, 9600 wouldn't have to clock as high. It only has to beat a 5200. Since NV34 is manufactured at .15, at TSMC to boot, I would assume .15 is still quite a bit cheaper than .13 per chip, and that's what counts.

 

[ZoinKs!]

The market can be roughly divided up something like this: value, mainstream, performance, enthusiast. "Performance mainstream" is about $150-200 range, the same place the 9600pro is right now. The 5200 is a value or mainstream card.

I'm assuming ati's next performance mainstream product will be based on rv360, which is just a fast, economical rv350. If this is the case, it'd have to be .13, but DaveB has implied it might not. So that's got me curious... Anything other than .13 for that market segment is a surprise.

Which is fine with me. I like surprises.

 

[just me]

The .13, 9600 seems to have a lot of headroom and a .15, 9600 wouldn't have to clock as high. It only has to beat a 5200.

Au contrare, IMO.

It has to be comparable to a 5200nu price & beat the performance of the 5200's & be w/in -10%max of a 5600nu. The 9600Pro has to compete w/the 5600Ultra in price/performance. The 5600nu is in the 9600np class, not the 9600Pro class afterall. 8)

.02,

 

[Fred da Roza]

The market can be roughly divided up something like this: value, mainstream, performance, enthusiast. "Performance mainstream" is about $150-200 range, the same place the 9600pro is right now. The 5200 is a value or mainstream card.

I'm assuming ati's next performance mainstream product will be based on rv360, which is just a fast, economical rv350. If this is the case, it'd have to be .13, but DaveB has implied it might not. So that's got me curious... Anything other than .13 for that market segment is a surprise.

Which is fine with me. I like surprises.

In another thread Joe DeFuria wrote:
http://www.beyond3d.com/forum/viewtopic.php?t=7417

Just some blurbs from the conference presentation above:

ATI stated:

We do intend going forward to have a multiple-foundry strategy as we have now. We will probably not take part of the high end graphics and send it somewhere else (like apparently our competitor).

DaveB has already said:

You'll see 9600's working their way down soon.

So what part will ATI have UMC manufacture?

 

[Dave Baumann]

I think people are confusing the path RV350 will take with the path RV360 will take.

 

[Joe DeFuria]

I think people are confusing the path RV350 will take with the path RV360 will take.

I think Dave is purposely trying to confuse people on the RV350 and RV360.

 

[NocturnDragon]

I think people are confusing the path RV350 will take with the path RV360 will take.

Rv350 take the current place of Rv280
and
Rv360 take the current place of Rv350?

 

[kemosabe]

Deleted and reposted hereunder...

 

[kemosabe]

I think people are confusing the path RV350 will take with the path RV360 will take.

Rv350 take the current place of Rv280
and
Rv360 take the current place of Rv350?

That sure sounds like the reasoning Dave is trying to impart.

RV280 goes to the integrated segment.

RV350 design fabbed on 0.15u (at UMC?) to reduce costs and offer a compelling yet cheap alternative to the 5200.

RV360/Pro on 0.13u at TSMC with higher clocks (and a few tweaks) to offer 9500 Pro+ level performance in the lucrative $200-$250 segment and lessen the gap with the upcoming 9900 Pro. Will the 9800 non-Pro survive this shakeup?

 

[Fred da Roza]

I think people are confusing the path RV350 will take with the path RV360 will take.

Rv350 take the current place of Rv280
and
Rv360 take the current place of Rv350?

That sure sounds like the reasoning Dave is trying to impart.

RV280 goes to the integrated segment.

RV350 design fabbed on 0.15u (at UMC?) to reduce costs and offer a compelling yet cheap alternative to the 5200.

RV360/Pro on 0.13u at TSMC with higher clocks (and a few tweaks) to offer 9500 Pro+ level performance in the lucrative $200-$250 segment and lessen the gap with the upcoming 9900 Pro. Will the 9800 non-Pro survive this shakeup?

I agree. .15 at UMC seems to make a lot of sense from a cost perspective but it's completely a guess.

A simple respin of the 9600 Pro at .13 (RV360) seems a bit odd to me. Why just tweak something that can overclock the way it does already. Is something being added back in (like Hierarchical Z)? That would make more sense to me since ATI says they are getting good yields and apparently can push RV350 clock rates higher without respinning the chip. That’s complete speculation of course.

 

[Entropy]

I agree. .15 at UMC seems to make a lot of sense from a cost perspective but it's completely a guess.

A simple respin of the 9600 Pro at .13 (RV360) seems a bit odd to me. Why just tweak something that can overclock the way it does already. Is something being added back in (like Hierarchical Z)? That would make more sense to me since ATI says they are getting good yields and apparently can push RV350 clock rates higher without respinning the chip. That’s complete speculation of course.

Making a respin for nothing but increases in clock would seem pretty pointless. Binning would serve the same purpose without mucking about with a production line that appears to work just fine.

So as you say, a new chip would seem to imply new functionality. HierZ, other memory related improvements, improvements to the performance of stencil operations, AA performance improvements ... the list of possibilities is long indeed. A pity of sorts that rumour mongering and wild speculation is so sadly lacking in the midrange. "Somewhat faster GPU clock" doesn't really excite, and doesn't seem to make all that much sense either.

Entropy

 

[Fred da Roza]

Making a respin for nothing but increases in clock would seem pretty pointless. Binning would serve the same purpose without mucking about with a production line that appears to work just fine.

So as you say, a new chip would seem to imply new functionality. HierZ, other memory related improvements, improvements to the performance of stencil operations, AA performance improvements ... the list of possibilities is long indeed. A pity of sorts that rumour mongering and wild speculation is so sadly lacking in the midrange. "Somewhat faster GPU clock" doesn't really excite, and doesn't seem to make all that much sense either.

Entropy

Adding new functionality would truly distinguish it from the 9600 non-pro (instead of just artificially lowering the clock). Coupled with some memory related improvement (faster memory, a 256-bit memory bus…) and you have a reason to buy a 9600 Pro at a premium over a 9600. Ultimately ATI could have 3 variants of the "9600" (RV360, .13 RV350 and .15 RV350) covering the $100 to $250 segment of the market which would compete against the 5600 Ultra, 5600 and 5200 Ultra, respectively.

 

[kemosabe]

Making a respin for nothing but increases in clock would seem pretty pointless. Binning would serve the same purpose without mucking about with a production line that appears to work just fine.

So as you say, a new chip would seem to imply new functionality. HierZ, other memory related improvements, improvements to the performance of stencil operations, AA performance improvements ... the list of possibilities is long indeed. A pity of sorts that rumour mongering and wild speculation is so sadly lacking in the midrange. "Somewhat faster GPU clock" doesn't really excite, and doesn't seem to make all that much sense either.

Entropy

Adding new functionality would truly distinguish it from the 9600 non-pro (instead of just artificially lowering the clock). Coupled with some memory related improvement (faster memory, a 256-bit memory bus…) and you have a reason to buy a 9600 Pro at a premium over a 9600. Ultimately ATI could have 3 variants of the "9600" (RV360, .13 RV350 and .15 RV350) covering the $100 to $250 segment of the market which would compete against the 5600 Ultra, 5600 and 5200 Ultra, respectively.

I would speculate that there would be a 0.15u RV350 (64 and 128 Mb) under $100 for low end/OEM and that the 0.13u RV360/Pro will replace the 0.13u RV350/Pro in the mainstream/performance segments ($100-$200). If the 9800 non-Pro survives, it would cover the $250-$300 range and the 9900/Pro would sit at the top.

0.15u RV350 = Radeon 9400? (if slower than the current 9600 non-Pro - otherwise?)
0.13u RV360 = ??? (no clue what they're gonna call this one)

 

[digitalwanderer]

Oh man, if they can get some of that R(v)3xx down to the sub-$100 bracket it'll be game over for nVidia for market dominance in ALL categories!

 

[Fred da Roza]

I would speculate that there would be a 0.15u RV350 (64 and 128 Mb) under $100 for low end/OEM and that the 0.13u RV360/Pro will replace the 0.13u RV350/Pro in the mainstream/performance segments ($100-$200). If the 9800 non-Pro survives, it would cover the $250-$300 range and the 9900/Pro would sit at the top.

0.15u RV350 = Radeon 9400? (if slower than the current 9600 non-Pro - otherwise?)
0.13u RV360 = ??? (no clue what they're gonna call this one)

Was thinking thats how it would plays out myself. How cheap is .15 at UMC compared to .13 at TSMC. Would ATI still have decent margins offering a .15 RV350 in the $100 - $150 range. If it carried a slight ($10-$20) premium over a 5200 Ultra I would expect it could still sell, provided it was clearly better. As Dave said I guess we will find out soon.

Oh man, if they can get some of that R(v)3xx down to the sub-$100 bracket it'll be game over for nVidia for market dominance in ALL categories!

Do you think they have to? Can the 9000/9200 cover the sub $100. Isn't nVidia still pushing the GeForce 4 in that space. Just because nVidia is offering ultra low performing 5200 doesn't mean ATI has to follow. How do the 5200 non-ultras compare against the 9000/9200 variants?