AMD: R9xx Speculation

[Mianca]

Just some quick amateurish quetions from a long time lurker:

Given

(a) the current market situation (i.e. AMD still holding the most porfitable and competitive GPUs in most of the market segments) and

(b) the recent reports of GloFo's 28nm process being just a few more months away

Why would AMD not just want to skip their tweeked 40nm architecture in favor of concentrating their engineering resources on designing/building new chips on the 28nm node? Weren't there complex test-structures (early pipe-cleaners?) around for a few months now? What's the point in releasing a half-step 6xxx family in Q4 2010 when you could very well just spend the winter on (cheap-to-develop) 5*90 tweaks - and announce your entirely new 28nm family in Q1 2011 (with availability targeted for Q2 2011)?

I just don't get the point in doing a half-gen upgrade that late in the R8xx lifecycle (or, for that matter, that near to the alleged start of the new 28nm process)? Maybe AMD are just fooling around with us - and are shooting for their real-deal-2400-shaders-28nm-chip directly? What would be the probabilities of this?

As I said: I'm just a naive amateur in this field so I might not get a lot of important points - but all those rumours of AMDs new R9xx family still being produced on the 40nm process don't seem to add up very well in my eyes.

Hell, they could probably even start their new lineup with a not-that-complex 28nm 67xx chip to "get used" to the new process - and it would still be better than most stuff they could possbily come up with using the old 40nm node?

 

[no-X]

What's the point in releasing a half-step 6xxx family in Q4 2010

Xmas sales.

They already missed back-to-school, nVidia's GTX460 seems to be the winner of this period. It wouldn't be clever to give up the X-mas sales.

If the HD6xxx architecture increases efficiency (performance per square milimeter), it could solve the problem with insufficient manufacturing capacity.

 

[rpg.314]

Just some quick amateurish quetions from a long time lurker:

Given

(a) the current market situation (i.e. AMD still holding the most porfitable and competitive GPUs in most of the market segments) and

(b) the recent reports of GloFo's 28nm process being just a few more months away

Why would AMD not just want to skip their tweeked 40nm architecture in favor of concentrating their engineering resources on designing/building new chips on the 28nm node? Weren't there complex test-structures (early pipe-cleaners?) around for a few months now? What's the point in releasing a half-step 6xxx family in Q4 2010 when you could very well just spend the winter on (cheap-to-develop) 5*90 tweaks - and announce your entirely new 28nm family in Q1 2011 (with availability targeted for Q2 2011)?

I just don't get the point in doing a half-gen upgrade that late in the R8xx lifecycle (or, for that matter, that near to the alleged start of the new 28nm process)? Maybe AMD are just fooling around with us - and are shooting for their real-deal-2400-shaders-28nm-chip directly? What would be the probabilities of this?

As I said: I'm just a naive amateur in this field so I might not get a lot of important points - but all those rumours of AMDs new R9xx family still being produced on the 40nm process don't seem to add up very well in my eyes.

Hell, they could probably even start their new lineup with a not-that-complex 28nm 67xx chip to "get used" to the new process - and it would still be better than most stuff they could possbily come up with using the old 40nm node?

IMHO, if you have verified your new/semi-new architecture in production/silicon, it would make sense to recover it's costs by selling it. And if perf/$ is improved, it means higher margins all around.

 

[trinibwoy]

They already missed back-to-school, nVidia's GTX460 seems to be the winner of this period. It wouldn't be clever to give up the X-mas sales.

I don't see why the existing 5xxx series won't sell very well into the back-to-school period. Same goes for Xmas sales. Nvidia still doesn't have anything compelling against Juniper and Redwood or Cedar.

 

[Bouncing Zabaglione Bros.]

Why would AMD not just want to skip their tweeked 40nm architecture in favor of concentrating their engineering resources on designing/building new chips on the 28nm node?

And what if 28 nm is problematical (like the cancelled 32nm) or late and constrained (like 40nm)? AMD want to get new product out this autumn, they don't want to maybe be stuck waiting for 28nm in the middle of 2011. It's been very high risk these last few years to rely on foundries being able to get their new processes up to speed for mass manufacturing according to their promised schedules.

Today's ATI/AMD is all about the execution. They are ahead now, but Nvidia is starting to get it's act together with it's DX11 cards. I don't see that AMD are going to risk their schedule on an untried, and currently unavailable process that might end up painting themselves into a corner.

Instead they will refine what they have for their Fall refresh and make sure they get something newer and better for the market, and keep a lead over Nvidia. After all, it's not like AMD have to rush to 28nm in order to combat Nvidia, because Nvidia have not been able to execute on time and are still playing catch-up.

 

[no-X]

I don't see why the existing 5xxx series won't sell very well into the back-to-school period. Same goes for Xmas sales. Nvidia still doesn't have anything compelling against Juniper and Redwood or Cedar.

Since the launch of GTX460 gamers almost stopped to buy HD5850s. That's not a short-term problem, but a few months without response can make things worse.

People prefer new things. HD5850 will be more than one year old before Xmas, more expensive than in 2009 and it will appear to be morally obsolete. More and more people prefer GTX460 over HD5830, HD5850 and sometimes even over HD5870 (significanly lower price and comparable performance in many games after OC).

ATi could launch HD5890 around the launch od GTX480 - they didn't. They could launch HD5840 against GTX460, but they didn't. Excessive self-confidence isn't good. nVidia can launch full-fledged GF104 performing like HD5870 anytime. People will prefer newer solution despite comparable performance or price/performance.

 

[eastmen]

Since the launch of GTX460 gamers almost stopped to buy HD5850s. That's not a short-term problem, but a few months without response can make things worse.

People prefer new things. HD5850 will be more than one year old before Xmas, more expensive than in 2009 and it will appear to be morally obsolete. More and more people prefer GTX460 over HD5830, HD5850 and sometimes even over HD5870 (significanly lower price and comparable performance in many games after OC).

ATi could launch HD5890 around the launch od GTX480 - they didn't. They could launch HD5840 against GTX460, but they didn't. Excessive self-confidence isn't good. nVidia can launch full-fledged GF104 performing like HD5870 anytime. People will prefer newer solution despite comparable performance or price/performance.

I don't know about that. The gtx 460 even overclocked doesn't aproach the 5870. And don't forget ati's parts can be overclocked also. Then there is the fact that cypress is smaller than the gf104.

 

[Silent_Buddha]

Since the launch of GTX460 gamers almost stopped to buy HD5850s. That's not a short-term problem, but a few months without response can make things worse.

People prefer new things. HD5850 will be more than one year old before Xmas, more expensive than in 2009 and it will appear to be morally obsolete.

Those two statements are completely at odds with each other. If people have stopped buying the 5850, supply will outstrip demand and ATI will lower prices.

Since that hasn't really happened yet, the only conclusion you can draw is that 5850 is still selling as much as it did prior to GTX 460 coming out.

5830 on the other hand...

Regards,
SB

 

[no-X]

eastmen & Silent_Buddha: That's exactly the problem. RV870 is smaller than GF104 - that's good for ATi. But this advantage is killed by limited supplies. Despite (probably) very good margins, ATi made only $33M last quarter. Really not stunning results if you take into account their near-monopolistic position on DX11 market.

If the demand for RV870 stops to be higher than supplies (due to the GF104) and ATi will cut prices, it won't help them in any way. They will have the same limited supplies, which will be sold with lower margins.

Without new products, which would offer better efficiency or/and higher performance their situation cannot be improved.

ATi has good products, ATi has better margins, ATi has (or had) DX11 monopol, but how they used this advantage? I see almost wasted occasion...

 

[Silent_Buddha]

If the demand for RV870 stops to be higher than supplies (due to the GF104) and ATi will cut prices, it won't help them in any way. They will have the same limited supplies, which will be sold with lower margins.

I don't quite follow your logic. If demand drops below supply and there is no change in prices, then ATI will start building up inventory of 5850 which isn't exactly a good thing as that product will be depreciating in value while it sits in a warehouse.

At that point, they have to drop the price or take an overall hit to their revenue and profits. Price remaining higher than launch price remains a strong indicator that demand is still far higher than available supply, in other words, they still sell just as many cards now as they did prior to GTX 460. Demand may or may not have gone down, but it's still higher than supply.

As to the rest, I agree. It's incredibly unfortunate for AMD, but fortunate for Nvidia, that TSMC has had problems producing 40 nm wafers in high enough volumes.

Regards,
SB

 

[eastmen]

eastmen & Silent_Buddha: That's exactly the problem. RV870 is smaller than GF104 - that's good for ATi. But this advantage is killed by limited supplies. Despite (probably) very good margins, ATi made only $33M last quarter. Really not stunning results if you take into account their near-monopolistic position on DX11 market.

If the demand for RV870 stops to be higher than supplies (due to the GF104) and ATi will cut prices, it won't help them in any way. They will have the same limited supplies, which will be sold with lower margins.

Without new products, which would offer better efficiency or/and higher performance their situation cannot be improved.

ATi has good products, ATi has better margins, ATi has (or had) DX11 monopol, but how they used this advantage? I see almost wasted occasion...

The real question is , are they still supply constrained ? and if so , to what degree.

If the gf104 is bigger than the cypress it means that ati could price the 5870 at gtx 460 prices and still make a profit. It will be a big negative for nvidia as the 5850 and 5870 start to decrease in price. IF there is a 40nm refresh part coming out , it may simply be meant to slot in above the 5870 with the rest of the line being moved downard.

 

[no-X]

It will be a big negative for nvidia as the 5850 and 5870 start to decrease in price.

nVidia had 3-times higher profit than ATi even before the GF104. nVidia doesn't need to sell GPUs to gamers (nV has tegras, ions and teslas), but ATi does. I think lower price can harm ATi more than nVidia.

 

[Sontin]

If the gf104 is bigger than the cypress it means that ati could price the 5870 at gtx 460 prices and still make a profit.

Without yield numbers this is a wrong assumption. AMD must sell 5830 and 5850 for good yields and lower cost per chip.

 

[Kaotik]

nVidia had 3-times higher profit than ATi even before the GF104. nVidia doesn't need to sell GPUs to gamers (nV has tegras, ions and teslas), but ATi does. I think lower price can harm ATi more than nVidia.

nVidia made almost 42 million loss with Tegras, and around 50% more profit with GeForces and Ions than Teslas / Quadros.

 

[eastmen]

nVidia had 3-times higher profit than ATi even before the GF104. nVidia doesn't need to sell GPUs to gamers (nV has tegras, ions and teslas), but ATi does. I think lower price can harm ATi more than nVidia.

ati's fire gl line up is pretty good.

Without yield numbers this is a wrong assumption. AMD must sell 5830 and 5850 for good yields and lower cost per chip.

cypress could have much better yields . After all ati is able to sell even bad chips. Right now there are now salvage gf104 parts ot there.

 

[cal_guy]

nVidia had 3-times higher profit than ATi even before the GF104. nVidia doesn't need to sell GPUs to gamers (nV has tegras, ions and teslas), but ATi does. I think lower price can harm ATi more than nVidia.

Tegra has been a consistent money loser for nV and I 'm not convinced that Tesla is much of a moneymaker with it's high R&D costs and low volume.

 

[Sontin]

cypress could have much better yields . After all ati is able to sell even bad chips. Right now there are now salvage gf104 parts ot there.

If nVidia selling all chips as GTX460 they will have equal yield numbers. So the cost per chip should nearly the same for nVidia and AMD. But if AMD can't sell 5830 to their customers anymore then their cost per chip increase which will lower their margin. What will happen when they lower the price for the 5850 and 5870, too?

 

[eastmen]

If nVidia selling all chips as GTX460 they will have equal yield numbers. So the cost per chip should nearly the same for nVidia and AMD. But if AMD can't sell 5830 to their customers anymore then their cost per chip increase which will lower their margin. What will happen when they lower the price for the 5850 and 5870, too?

of course no chip yields 100% so its silly to even put that forth. There will be broken chips on both ends. Ati has two salavage parts and nvidia has none .

 

[Sontin]

of course no chip yields 100% so its silly to even put that forth. There will be broken chips on both ends. Ati has two salavage parts and nvidia has none .

GTX460 has only salvage part.

 

[no-X]

Well...

tegra is money looser, tesla isn't money maker, GF100 was debacle...
vs.
cypress has higher margins, cypress has better yields, fire gl line sells well...

Can anybody explain, why nVidia earned $137M and $131M during last two quarters, while ATi earned only $33M and $47M?