Leaked Presentation Reveals AMD's Fusion Strategy

[AtenRa]

http://www.ngohq.com/home.php?page=articles&go=read&arc_id=154

 

[Idontcare]

Pretty cool, can't wait for the Llano laptops to be available.

Hopefully once the hardware is available retail-wise then more OpenCL apps will be developed to take advantage of the APU.

It seems like the launch-offerings are rather slim, software wise, but this is always the case with new ISA extensions so it is not surprising.

 

[dpodblood]

Interesting. It would be nice to see the graphical abilities of mainstream laptops improve, along with battery life. Looks like Fusion aims to do both.

 

[AtenRa]

Llano mobile performance in 3D Mark Vantage ??

http://www.3dnews.ru/news/611684/

 

[Terzo]

Notice at the bottom of every slide, "Confidential - Under Embargo until June 14, 2011 at 12:01 am Eastern".

I'm hoping this means June 14th AMD will release a lot more information on Llano (Bulldozer too would be nice) and have product availability shortly thereafter.

The Llano part makes sense, right? Didn't AMD announce a few months back that they were shipping Llano to partners?

 

[Phynaz]

Always fun to see the marketing guys practice their craft.

Intel uses numbers 3,5,7...AMD uses 4,6,8.

AMD's chips must be better because they have bigger numbers

 

[drizek]

Always fun to see the marketing guys practice their craft.

Intel uses numbers 3,5,7...AMD uses 4,6,8.

AMD's chips must be better because they have bigger numbers

Ya, noticed that too.

Thing is though, even numbers don't sound as fast as odd numbers. Also, they basically just nmed them after Audis.

 

[piesquared]

Notice at the bottom of every slide, "Confidential - Under Embargo until June 14, 2011 at 12:01 am Eastern".

I'm hoping this means June 14th AMD will release a lot more information on Llano (Bulldozer too would be nice) and have product availability shortly thereafter.

The Llano part makes sense, right? Didn't AMD announce a few months back that they were shipping Llano to partners?

That will be the Fusion Developer Summit, so it makes sense.

Also, these will be the real threat to sandy britches. Bulldozer for enthusiast, workstation and server. Since there is no architectural difference, dropping it down to the very low volume enthusiast is vastly simplified giving the much larger, much higher margin server market good volume. Looks like quite a good strategy. Sort of similar with what intel is doing with sandy bridge and sandy bridge e, and so the convergence is beggining. The big question is, who has the more attractive platform. I think it's pretty safe to say that Fusion offers more appeal to consumers, but we'll see if intel successfully managed to hamstring them again.

BTW, there's also a reference to the Z-series tablet APU's, and an article here:

http://www.fudzilla.com/processors/item/22893-amd-fusion-z-series-promises-ultimate-hd-tablets

 

[Medu]

AMD are starting to flesh out their lineup. From netbooks to servers, assuming BD is competitive with SB, they have very compelling products. Hopefully they can make some serious money over the next few years and pump it into R&D.

 

[JimKiler]

Ya, noticed that too.

Thing is though, even numbers don't sound as fast as odd numbers. Also, they basically just nmed them after Audis.

Sent from my SPH-D700 using Tapatalk

But at least AMD's number represent the number of cores. Go ahead and resume the argument about whether they are "true" cores or not.

I participated in an AMD survey once about graphics cards and I said the number should be the only thing needed to know if a card if faster than other, none of the 5660 vs 5660 ultra crap. It should be 5660 and 5670 or something similar. Seems like AMD is making it easier to understand how a card will perform by its name.

 

[exar333]

That will be the Fusion Developer Summit, so it makes sense.

Also, these will be the real threat to sandy britches. Bulldozer for enthusiast, workstation and server. Since there is no architectural difference, dropping it down to the very low volume enthusiast is vastly simplified giving the much larger, much higher margin server market good volume. Looks like quite a good strategy. Sort of similar with what intel is doing with sandy bridge and sandy bridge e, and so the convergence is beggining. The big question is, who has the more attractive platform. I think it's pretty safe to say that Fusion offers more appeal to consumers, but we'll see if intel successfully managed to hamstring them again.

BTW, there's also a reference to the Z-series tablet APU's, and an article here:

http://www.fudzilla.com/processors/item/22893-amd-fusion-z-series-promises-ultimate-hd-tablets

The challenge for AMD will be threefold.
(1) Compete with performance
(2) Market successfully
(3) Longer to market vs. Intel

Intel already hos VERY solid marketing, a product that has been in the maket for 6 months, and very good performance already. AMD is late to the party and will have to play catch-up, regardless of how well it performs. This was a challenge with PhII as well. It was a solid product, but was too little too late, in many ways.

 

[piesquared]

The challenge for AMD will be threefold.
(1) Compete with performance
(2) Market successfully
(3) Longer to market vs. Intel

Intel already hos VERY solid marketing, a product that has been in the maket for 6 months, and very good performance already. AMD is late to the party and will have to play catch-up, regardless of how well it performs. This was a challenge with PhII as well. It was a solid product, but was too little too late, in many ways.

Naw, the problem with PhII wasn't that it was too little too late, it just never had any extra value proposition to consumers. There hasn't been anything new on the market for years, just the same thing faster and that's getting old real fast (as witnessed by the emergence of tablets, which consumers seem to enjoy while being an order of magnitude slower). So that's not the purpose of Fusion, it's about tapping into the changing paradigm. Absolute performance is Bulldozer's category, and usability and hitting the trends is where Fusion shines.

So,

1) not necessary, usability is far more important than performance in benchmarks.
2) they've been doing a good job at marketing despite intel's attempt to divert the dialogue.
3) there will always be something that is coming next and later to market. ie, ivy bridge with DX11 and OpenCL on GPU will be at least 6 months behind Fusion. Shortly after, as the 32nm process will be fully ramped with a planned %100 production, Trinity should hit the market.

Again, absolute performance isn't what the target market for Fusion is looking for, it's usability and enjoyment. I don't see %90 of the market enjoying running super pi %10 faster than whatever else is on the shelf. There is no appeal in that to the bulk of the market.

 

[Phynaz]

But at least AMD's number represent the number of cores.

The slides don't state this. As a matter of fact the "A" series is all Llano based, and I'm not aware of an 8 core Llano part.

On the second page they also talk about A8 notebook battery life, and I highly doubt that AMD is about to drop an 8 core notebook CPU.

 

[Khato]

The most interesting takeaway in my opinion would be found on the second slide of page 2. The Llano die-shot in the background would confirm that the nice looking 'square' die shot released back in November was indeed mere a cropped portion of the entire die. Going with the 226mm^2 die size claimed on the first slide of that page would result in the graphics taking up around 88mm^2, which is a fair deal larger than I would have expected given that the 400 shader Redwood core is supposedly only 104mm^2 on a 40nm process.

 

[Topweasel]

The slides don't state this. As a matter of fact the "A" series is all Llano based, and I'm not aware of an 8 core Llano part.

On the second page they also talk about A8 notebook battery life, and I highly doubt that AMD is about to drop an 8 core notebook CPU.

Might be Units. 4 100SP units, 4 CPU units.

 

[busydude]

Might be Units. 4 100SP units, 4 CPU units.

That cannot be true.. it is multiples of 80. 80SP's = 1 SIMD.

 

[Topweasel]

That cannot be true.. it is multiples of 80. 80SP's = 1 SIMD.

Okay. Maybe not. I don't know then. Maybe the highest SP is 320 and not 400. I could be just be that AMD wanted to use the same numbering scheme that the FX uses so that they know that the A8 is the highest combo of hardware (though not always of speed) like it is on the FX line.

 

[OCGuy]

But at least AMD's number represent the number of cores. Go ahead and resume the argument about whether they are "true" cores or not.

I participated in an AMD survey once about graphics cards and I said the number should be the only thing needed to know if a card if faster than other, none of the 5660 vs 5660 ultra crap. It should be 5660 and 5670 or something similar. Seems like AMD is making it easier to understand how a card will perform by its name.

Huh? They just rebadged the 5770 to 6770, and the 5870 is faster than the 6870. They have gotten worse.

 

[jimbo75]

Huh? They just rebadged the 5770 to 6770, and the 5870 is faster than the 6870. They have gotten worse.

You cannot blame AMD for TSMC's incompetence.

 

[Tuna-Fish]

You cannot blame AMD for TSMC's incompetence.

You can blame them for not naming 68xx => 67xx, and rebadging 57xx. Barts wasn't at any point supposed to be faster than Cypress. Why exactly did it get a higher model number?

 

[Topweasel]

You can blame them for not naming 68xx => 67xx, and rebadging 57xx. Barts wasn't at any point supposed to be faster than Cypress. Why exactly did it get a higher model number?

Easy market slot. In a perfect world AMD would have had had a die shrunk 5700, for a 6500 line or something like that, Cypress or something close would have fit in the 6700 slot and something bigger would have gone in the 6800 slot. TMSC screws up and AMD is stuck with the same process and needs to eak out performance they rework Cypress for more performance, but they can't spend a fortune reworking every core. So they designed a core that would fit in between the the two and at a size that they had good success with (the 4870).

It had more to do with slotting Barts in somewhere but not pushing everything down directly because the die size of them start really cutting into profits below the x7xx series.

 

[AtenRa]

Llano die picture from the AMD Presentation

 

[Khato]

Llano die picture from the AMD Presentation

Yup, which matches up quite well with an earlier comparison of the nice looking cropped die shot that AMD released to a manipulation of a Llano wafer shot. Well, except for the fact that this is nowhere near so blurry =D

Either way, going with the 226mm^2 total die area, the total GPU logic ends up being something around 88mm^2. Which is important to keep in mind when comparing performance to Sandy Bridge at its ~41mm^2. (Looking at AMD/NVIDIA products, they typically get 1.65-1.8x the performance from doubling the number of execution resources, eg 5770 -> 5870.)

 

[AtenRa]

Computex Taipei 2011 Coverage Exclusive Performance preview of AMD LIano vs Intel Sandy Bridge

http://en.ocworkbench.com/tech/comp...preview-of-amd-liano-vs-intel-sandy-bridge/2/

 

[drizek]

Epic domination in graphics. Amds bee optimizing for 3dmark for a decade now, intel doesn't stand a chance.

I could never recommend the i3 to anybody now, just based on those two benches. Even the i5 is a tough sell to someone who doesn't want a dedicated graphics chip.