Southern Islands, 2010?

[Bateluer]

http://www.xbitlabs.com/news/video/...ation_DirectX_11_Family_of_Chips_in_2010.html

Advanced Micro Devices said that its ATI graphics business unit has shipped over sixteen million of DirectX 11-compatible graphics processing units and reiterated plans to introduce second-generation DirectX 11 graphics chips later in 2010. Besides, despite rumours, AMD expects supply constraints to ease in the second half of the year.

"Demand for our GPU offerings in the quarter was very strong, and shipments were temped only by supply constraints. This was the second straight quarter of record GPU shipments, led by an 18% sequential increase in notebook discrete units. More than half of our GPU shipments in the quarter were DirectX 11-capable and [so far] we have shipped nearly 16 million DX11 GPUs to date," said Dirk Meyer, chief executive officer of Advanced Micro Devices, during the quarterly conference call with financial analysts.

AMD announced revenue for the second quarter of 2010 of $1.65 billion, a net loss of $43 million, or $0.06 per share, and operating income of $125 million. The company reported non-GAAP net income of $83 million, or $0.11 per share, and non-GAAP operating income of $138 million.

ATI's revenue increased 8% sequentially and 87% year-over-year to approximately $470 million, which is higher than GPU revenues received by ATI when it was independent back in 2006. The sequential increase was driven by record GPU unit shipments, partially offset by a decrease in average selling price. The year-over-year increase was driven by an increase in GPU shipments and ASP. Operating income was $33 million, compared with $47 million in Q1 2010 and a loss of $17 million in Q2 2009.

AMD also reaffirmed plans to introduce the next-generation ATI Radeon HD 6000-series graphics processors in 2010. However, this time AMD did not say that it planned to refresh the whole lineup, but only promised to debut some of the novelties.

"In the second half of the year, we expect GPU demand to remain healthy and supply constraints to ease. We remain on track to bolster our GPU leadership with the introduction of our second-generation DirectX 11 graphics products later this year. [...] We will start introducing the second-generation of our DX 11 products before the end of the year," said Mr. Meyer.

Nice. Hopefully, we'll see 2GB cards launch along side 1GB models. Definitely need more video RAM than 512MB, but I don't want to wait 6 months longer for more VRAM.

 

[OCGuy]

I fully expect SI to launch right before Xmas. It is a 1/2 gen, but to be honest, that is rather exciting considering the crap we have been spoon-fed in this latest generation.

5XXX was basically a 4XXX on steriods.

4XX is basically an arch that is ahead of it's time, and has much maturing to do before it can reach it's potential. nV tried to cram too much on too large a chip.

 

[tincart]

Fudzors says October for the low-end SI parts and maybe the high-end: http://www.fudzilla.com/graphics/graphics/atis-next-generation-to-come-in-october

Not exactly the most reputable source in the world, but the info matches everything else pretty well.

 

[happy medium]

Fudzors says October for the low-end SI parts and maybe the high-end: http://www.fudzilla.com/graphics/graphics/atis-next-generation-to-come-in-october

Not exactly the most reputable source in the world, but the info matches everything else pretty well.

Funny, I stated this in another S.I. thread because I read the same thing a month ago. I read they were doing the same 4770 thing, mid-range first ,high end later. Who knows?

 

[Dark4ng3l]

I have seen speculation all over the place. The one thing that strikes me though is that i'm not sure what they would have to gain by refreshing the 5400 and 5500 card at all. I mean those cards don't have real gaming performance anyways and considering that this is not a shrink it seems like a waste of engineering to me. I would probably have kept the same chips and just clocked them a bit higher or something.

The only reason that I can see for the refreshing them like this is that they have implemented some new features that would matter even at the low end. Something like 3 monitors without the need to use DP maybe? Besides that I don't see why they would waste their time making something that is not built for speed slightly faster.

 

[Lonyo]

I would have thought they would go for the mid-range first.
There's the HD5830 at the moment, but a new chip to fit between the HD57xx and HD58xx cards would make more sense, something smaller than the HD58xx die but more powerful than the HD57xx cards. That's where they're really missing something and where would make most sense for a new chip.

Seems odd to aim elsewhere first.

 

[Dark4ng3l]

I would have thought they would go for the mid-range first.
There's the HD5830 at the moment, but a new chip to fit between the HD57xx and HD58xx cards would make more sense, something smaller than the HD58xx die but more powerful than the HD57xx cards. That's where they're really missing something and where would make most sense for a new chip.

Seems odd to aim elsewhere first.

Well they don't always have the choice of when the products become ready. They time these things so that they try and complete the projects based on the timeline that they want. Usually they want to launch the highest end part first so they organize their engineering for that to happen. Now building 4 different chips like this does have some overlap in terms of architecture but they are developed into cards from there independently.

When you are trying to launch 4 chips in a say 3 month window all it takes is one minor delay to mess up the order of completion of the different projects. Also when you are talking about filling the gap in the lineup you need to understand that they started design on these things probably a good while before Evergreen even launched so it's not like they are able to perceive the need to target the weakness in the lineup that probably would not exist had this generation been like previous ones. You would probably have 200$ 5850s by now and the gap would be non existent.

Anyways we have to wait and see what will happen with this. I have heard so many different things that you have to think people are making stuff up in the middle of completely lack on solid information. Either AMD's new "secrecy strategy" is really good or there might be some significant delays with the higher end SI parts.

 

[nitromullet]

Perhaps they are focusing on a higher mid-range chip that performs a little better than the current 5870 chip, but can be used on a dual gpu card that doesn't have the be scaled back to meet PCIe spec.

IIRC, ATI's strategy was to build smaller, mid range chips and use multi-gpu for the high end. What they ended with the 5-series is a pretty large chip that performed nicely on its own, but didn't pan out so well for their multi-gpu strategy. While a kick ass card, the 5970 had to be trimmed down to meet spec and the cherry picked chips required to produce the card made it pretty scarce and expensive. If they keep going in the direction they've gone since the 3-series with increasing chip size, eventuality they will be back where they were with R600 and where NVIDIA is with Fermi.

 

[tincart]

I have seen speculation all over the place. The one thing that strikes me though is that i'm not sure what they would have to gain by refreshing the 5400 and 5500 card at all. I mean those cards don't have real gaming performance anyways and considering that this is not a shrink it seems like a waste of engineering to me. I would probably have kept the same chips and just clocked them a bit higher or something.

Those are likely the cards going into OEM computers. Makes sense to get those out the door first and try to get some design wins for Christmas system buyers.

 

[sandorski]

Those are likely the cards going into OEM computers. Makes sense to get those out the door first and try to get some design wins for Christmas system buyers.

Also, the SI equivalents might be Cheaper to produce.

 

[at80eighty]

I would have thought they would go for the mid-range first.
There's the HD5830 at the moment, but a new chip to fit between the HD57xx and HD58xx cards would make more sense, something smaller than the HD58xx die but more powerful than the HD57xx cards. That's where they're really missing something and where would make most sense for a new chip.

Seems odd to aim elsewhere first.

pretty much what I think too - atleast from a sales perspective it would make more sense to have a card that fits a sweet spot rather than overhauling a slot currently taken by one of their existing products. that said id totally be eager to pick up something that could take on the 5970

 

[Wag]

pretty much what I think too - atleast from a sales perspective it would make more sense to have a card that fits a sweet spot rather than overhauling a slot currently taken by one of their existing products. that said id totally be eager to pick up something that could take on the 5970

But why would they have any motivation to produce a card that is faster than the 5970 when it is still in short supply (and selling for >$600)? I imagine if they did have such a product on deck it wouldn't see the light of day until Nvidia comes out with a legit challenger like a dual-GPU GF104. Otherwise they can just rest on their laurels for some time to come.

 

[SlowSpyder]

Funny, I stated this in another S.I. thread because I read the same thing a month ago. I read they were doing the same 4770 thing, mid-range first ,high end later. Who knows?

The only reason I think that they'll come out with a high end single GPU (something like the 6850 or 6870, assuming those are the names) is because everything I've read says that they'll still be using 40nm. They are familiar with this process. With the 4770, it was a feeler of sorts for the new at the time 40nm process. If this was going to be a 32 or 28nm part than I would expect something a little more mid to performance range, or even low end to show up first. But, like anything else right now, that's just speculation on my part.

But why would they have any motivation to produce a card that is faster than the 5970 when it is still in short supply (and selling for >$600)? I imagine if they did have such a product on deck it wouldn't see the light of day until Nvidia comes out with a legit challenger like a dual-GPU GF104. Otherwise they can just rest on their laurels for some time to come.

Not that I'm saying whatever they come out with first will take on the 5970, but it could make sense I would think. What if they had a single 400mm2 40nm chip that gave 5970 performance, yielded well, and the card still sold in the $600 range? That vs. two 330mm2 chips in a card that costs around the same. I imagine their partners would love that. Again, I'm not suggesting that's what the case will be, but I do think there could be some motivation for them to create the part.

 

[Grooveriding]

The only reason I think that they'll come out with a high end single GPU (something like the 6850 or 6870, assuming those are the names) is because everything I've read says that they'll still be using 40nm. They are familiar with this process. With the 4770, it was a feeler of sorts for the new at the time 40nm process. If this was going to be a 32 or 28nm part than I would expect something a little more mid to performance range, or even low end to show up first. But, like anything else right now, that's just speculation on my part.



Not that I'm saying whatever they come out with first will take on the 5970, but it could make sense I would think. What if they had a single 400mm2 40nm chip that gave 5970 performance, yielded well, and the card still sold in the $600 range? That vs. two 330mm2 chips in a card that costs around the same. I imagine their partners would love that. Again, I'm not suggesting that's what the case will be, but I do think there could be some motivation for them to create the part.

Going on ATI having the lead over nvidia currently with their production cycles, I'm thinking they will wait to see what nvidia refreshes perform like and give us a single GPU S.I. card that outperforms nvidias refresh of the 480 later this year.

It will likely fall somewhere between 480 and 5970 performance. I don't think we will see a single gpu performer that matches 5970 performance until next year. That is a fairly monstrous GPU and not one I think we will see initially from S.I.

 

[lifeblood]

I bet they come out with something targeted at the $200 spot, specifically where the 5830 sits. The 5830 is made from "failed" chips. Enough time has passed the yields should be reaching a point where they have far fewer "failed" chips. But the $200 price point is still to important to not have a product there. Right now nVidia has the advantage there with it's excellent 460 product.

Just my $.02 worth.

 

[at80eighty]

But why would they have any motivation to produce a card that is faster than the 5970 when it is still in short supply (and selling for >$600)? I imagine if they did have such a product on deck it wouldn't see the light of day until Nvidia comes out with a legit challenger like a dual-GPU GF104. Otherwise they can just rest on their laurels for some time to come.

Oh I just said I would if they did - not that I actually expect them to. figured the smiley kind of gave it away already

 

[RussianSensation]

Since ATI is not moving to 28nm, and they are still constrained by 40nm supply, it could make sense to release a high end SI and drop the prices of 5850 and 5870 to compete with NV. If they price SI at $399 and it's faster than the GTX480, they'll enjoy nice profit margins until next year when 28nm is online. At the same time, 5850 and 5870 will be able to easily compete with NV at $199 and $299. I am not sure if they will have enough time to replace both 5850 and 5870 with SI derivatives given that NV 'reserved' lot of manufacturing space from TSMC with 460s.