AMD talks up its first Fusion chip

[god_43]

CHIP DESIGNER AMD has been talking up some of the x86 design features for its first Fusion processor at the International Solid State Circuits Conference (ISSCC) in San Francisco this week.

The upcoming 'Llano' Accelerated Processing Unit (APU) will see the joining of a 32nm silicon-on-insulator (SoI) Phenom II quad core CPU with a DirectX 11 capable GPU on the same die. This is a more sophisticated approach than Intel's at present, in which it simply adds a GPU chip to the processor package and calls that 'integrated'.

http://www.theinquirer.net/inquirer/...ks-fusion-chip

http://www.eetimes.com/news/latest/s...2700324&pgno=2

[Hornet331]

Oh dear, and again we see the monolithic bs card in game....

[freeloader]

Oh dear, and again we see the monolithic bs card in game....

Who cares about monolithic versus 2 separate dies? It's performance per watt that matters. I don't care how any company gets there.

[Kai Robinson]

It seems that whenever Intel has no competition, it becomes complacent. AMD seems to be far more innovative than Intel lately, even if its performance isn't as high end.

[informal]

The Llano's RV8xx DX11 part with 400+ stream processors will simply be in its own league when it comes to performance... Add in the hybrid CF capability with discrete cards and I can't see what kind of redesign the current Intel GMA series need to receive to be even closely competitive with Llano's GPU when Intel integrates them into SB cores.

[zalbard]

The Llano's RV8xx DX11 part with 400+ stream processors will simply be in its own league when it comes to performance... Add in the hybrid CF capability with discrete cards and I can't see what kind of redesign the current Intel GMA series need to receive to be even closely competitive with Llano's GPU when Intel integrates them into SB cores.

This will rock if the driver support is decent...
And hope OCing will be just as good as for Intel chips, too!

[god_43]

Oh dear, and again we see the monolithic bs card in game....

? (i love these emoticons)

if being monolithic is bad, then why did intel not stay with a glued dual core = quad core thing? only time will tell if SB will have a better approach than llano. in the end; intel went the monolithic route soo....idk.

[Hornet331]

Who cares about monolithic versus 2 separate dies? It's performance per watt that matters. I don't care how any company gets there.

exactly this, it doesn't matter if its one die or not.

? (i love these emoticons)

if being monolithic is bad, then why did intel not stay with a glued dual core = quad core thing? only time will tell if SB will have a better approach than llano. in the end; intel went the monolithic route soo....idk.

see above.

[Hans de Vries]

Presentation sheets and more info:


http://www.anandtech.com/cpuchipsets...oc.aspx?i=3736
http://www.eetimes.com/news/semi/sho...leID=222700375
http://www.eetimes.com/news/semi/sho...2700375&pgno=2


Regards, Hans

[Macadamia]

Core = 9.69mm^2
Core + L2 ~ 1.5x of that? = ~15mm^2

So basically quadcore logic under 90mm^2 (+IMC,HTPad...)?


IMO the total die size should be 140-180mm^2 or so, but no longer do they have to sell it at $99- fetching double the price should be easy, not that it would be a good purchase for desktop.

For laptops though I think this should be their way in, especially with the emphasis on power management- seems to me that this is a laptop oriented design for the most part, being able to target $300+ ASPs that AMD could previously only dream about.

[madcho]

i don't think this is for laptops, this will not include BD cores. The number of transistors show us that.

maybe i'm wrong, but laptops will come with BD cores. My point of view this is for smaller than laptops, maybe single core for iPad of Micro$oft. With a GPU this could be an Ipad killer.

good night ^^

[FischOderAal]

It seems that whenever Intel has no competition, it becomes complacent. AMD seems to be far more innovative than Intel lately, even if its performance isn't as high end.

That's normal for an underdog imho. If you can't beat your big competitor in the field ppl care the most try to be innovative.

Big companies (especially when they nearly have a monopoly) tend to become slow and uninnovative (does this word even exist?).

[>HyperlogiK<]

? (i love these emoticons)

if being monolithic is bad, then why did intel not stay with a glued dual core = quad core thing? only time will tell if SB will have a better approach than llano. in the end; intel went the monolithic route soo....idk.

It's not bad, it's just irrelevant unless it brings a performance increase or power cut.

[haylui]

AOD will allow users to unlock more SP????

[Hans de Vries]

Core = 9.69mm^2
Core + L2 ~ 1.5x of that? = ~15mm^2

So basically quadcore logic under 90mm^2 (+IMC,HTPad...)?


IMO the total die size should be 140-180mm^2 or so, but no longer do they have to sell it at $99- fetching double the price should be easy, not that it would be a good purchase for desktop.

For laptops though I think this should be their way in, especially with the emphasis on power management- seems to me that this is a laptop oriented design for the most part, being able to target $300+ ASPs that AMD could previously only dream about.

Later die pictures showed more GPU area so the die for so far is ~205mm^2



Regards, Hans

[qcmadness]

Later die pictures showed more GPU area so the die for so far is ~205mm^2



Regards, Hans

why there are 4 hypertransport links?

[freeloader]

why there are 4 hypertransport links?

Just a guess...Hans will obviously know more.

1 HT link to the GPU.
1 or 2 HT link(s) between the CPU cores.
1 HT link to the rest of the system. (PCI-E bus, southbridge, add in card, etc...)

[Hans de Vries]

Just a guess...Hans will obviously know more.

1 HT link to the GPU.
1 or 2 HT link(s) between the CPU cores.
1 HT link to the rest of the system. (PCI-E bus, southbridge, add in card, etc...)

Yes, interesting all these HT links... They are half width but probably
almost as fast as the full width HT's used up to now (>=6.4GHz)

Maybe for 2 dies per package for higher performance desktops?

or also for the HPC market (CRAY) to build supercomputers getting
a whole lot more TFlops from the GPU's?

or maybe one is used to transport monitor output data to the
South bridge?

Regards, Hans

[Otis11]

So any idea how these will perform?

Close to a Phenom II and ... ?

[qcmadness]

So any idea how these will perform?
Close to a Phenom II and ... ?

Porpus-based and double L2 cache, should be similar with PII performance.

[vietthanhpro]

Yes, interesting all these HT links... They are half width but probably
almost as fast as the full width HT's used up to now (>=6.4GHz)

Maybe for 2 dies per package for higher performance desktops?

or also for the HPC market (CRAY) to build supercomputers getting
a whole lot more TFlops from the GPU's?

or maybe one is used to transport monitor output data to the
South bridge?

Regards, Hans

Hypertranport 3.1: 3.2Ghz
-> HyperTranport 4.0 : 6.4Ghz.

[saaya]

Presentation sheets and more info:


http://www.anandtech.com/cpuchipsets...oc.aspx?i=3736
http://www.eetimes.com/news/semi/sho...leID=222700375
http://www.eetimes.com/news/semi/sho...2700375&pgno=2


Regards, Hans

cheers

Yes, interesting all these HT links... They are half width but probably
almost as fast as the full width HT's used up to now (>=6.4GHz)

Maybe for 2 dies per package for higher performance desktops?

or also for the HPC market (CRAY) to build supercomputers getting
a whole lot more TFlops from the GPU's?

or maybe one is used to transport monitor output data to the
South bridge?

Regards, Hans

hmmmm the latter makes a lot of sense... but do they need a seperate HT link for that? thats what confused me about intel as well, why dont they use that massive qpi bus to move all the display data

[LordEC911]

Have we seen stacked HT links before???

[god_43]

Porpus-based and double L2 cache, should be similar with PII performance.

how do you know its porpus based? besides many different improvements, it was my understanding that it would be based on the latest phenom core? i am sure there are plenty of details left out.

[Firestrider]

is dual channel ddr3 going to be enough bandwidth for both the gpu and cpu?