Intel's Sandy Bridge architecture to feature up to 2 GPUs on 1 monolithic die

[AuDioFreaK39]

Sandy Bridge CPU to have up to 2 GPUs



Monolithic on-die

We were quite shocked to learn that Intel might be the first company to ever launch two GPUs on the same die. Not only does it look like Intel might be the first with monolithically integrated with the CPU and have Fusion before AMD, bit it might be the first “graphics” company to ever launch two GPUs on the same die.

According to Intel’s plans that were revealed to partners and industry insiders Sandy Bridge, brand new 32nm architecture will have one to two graphics cores on a monolithic die.

Intel will also let you connect discrete graphics via PCIe 16X and even SLI / Crossfire should be possible with 2x 8X PCIe slots. This is likely to happen on desktop boards rather than notebook.

At this time we are not aware that Nvidia and ATI will launch their dual GPU solutions on a single chip, but if you look a high-end card, for example a Fermi, it card has 512 shader cores and if you cut it in half you get a mainstream card. With this in mind, ATI and Nvidia practically make high-end chips out of four low end GPUs.

Source

[Oliverda]

Dual GMA 4500.

[Behemot]

Oh god you have some faster wire Gonna to del my thread...

[zalbard]

FUD? Why not just use one single bigger GPU core? Or is it going to be some sort of 2 CUDA cores, just not CUDA? The die's monolithic... Doesn't make much sense, does it?

[Meaker]

FUD? Why not just use one single bigger GPU core? Or is it going to be some sort of 2 CUDA cores, just not CUDA? The die's monolithic... Doesn't make much sense, does it?

Yeah the reason people have not done this before was because it was better to just join the two, they must each serve some sort of different purpose.

[G.Foyle]

Yeah, right, but why would anyone want to put two GPUs on the same die?

[Hornet331]

this makes no sense at all... espeical since there is no multigpu approach from intel till now...

[zerazax]

Dual GMA 4500.

just what i always wanted!

[god_43]

i think that it is probably 2 reasons for this:

1. to better compete with llano, i do not think that anyone here can argue that intel is better at graphics than ati.

2. these two gpus have some sort of special function.

[Serra]

I'm guessing dual GMA for better dual display performance?

Definitely did not see this coming from Intel, but good for them if they can pull it off I suppose.

[Tenknics]

Maybe one dedicated for HD video decoding/h.264/blu-ray etc and the other for gaming/graphics? They've stressed modularity over the past few years with nehalem and westmere.

On the flip side, maybe since LRB will no longer be making it on die anytime soon, they realized that their alternative graphics core (GMA whatever) wouldnt be enough so they are putting 2. That seems more logical to me.

[Larvitar]

Dual GMA 4500.

Yeah... GMA X5200 should be a "revolutionary" IGP, but it was just a common IGP, no "revolutionary" performance. Intel does not know how to GPU's high performance.

(word of who had GMA 845, 865, 900, 950, 3100, X3100, and X5200 about to have)

[RaZz!]

At this time we are not aware that Nvidia and ATI will launch their dual GPU solutions on a single chip, but if you look a high-end card, for example a Fermi, it card has 512 shader cores and if you cut it in half you get a mainstream card. With this in mind, ATI and Nvidia practically make high-end chips out of four low end GPUs.

how... what, errr, he? with what in mind? i don't get this part :F

[trn]

Dual GPU cores on one die onboard the CPU? Why? By the time they are done with this thing its gonna look like an old Pentium 2 Slot 1 cartridge!

[Katanai]

Well imo, doubling up the crap only gives you a bigger crap...

[Manabu]

Maybe one dedicated for HD video decoding/h.264/blu-ray etc and the other for gaming/graphics? They've stressed modularity over the past few years with nehalem and westmere.

Video decoding is not done in a GPU, but in dedicated logic/DSPs. Like UVD on ATI GPUs, or PureVideo on Nvidia's, or Broadcom chip at some netbooks, etc. It is totaly different logic, even inside the same silicon, in the same way as a memory controller is different from an CPU core.

I see no reason for an "dual core gpu". I call FUD in this too.

[BababooeyHTJ]

How often is Fudzilla ever right when it comes to this kind of stuff? My local weather man is more accurate while trying to forecast a snowstorm two weeks in advance. There are two types of links that I won't click on and one is toms and the other is fudzilla.

[MirageSys]

Redundancy because the have extra allocatable space on die? I think not.

Multiple monitor support? I think not.

Hack-eyed attempt to run them together with Lucid technology? Maybe >_<

One dumpy GPU and one Dumpier GPU for energy conservancy? Might work.

Fud being full of it? Most likely.

[Nedjo]

why there's no [FUD] at the beginning of this thread title?

BTW even Intel knows that idea of monolithic dual-die GPU is pure nonsense

"At this time we are not aware that Nvidia and ATI will launch their dual GPU solutions on a single chip, but if you look a high-end card, for example a Fermi, it card has 512 shader cores and if you cut it in half you get a mainstream card. With this in mind, ATI and Nvidia practically make high-end chips out of four low end GPUs."

and chocolate manufacturers make 300g bars out of lots of small dices

eh Fudo should know better than this... but then again ridership age/knowledge should always come first

[rintamarotta]

just what i always wanted!

in Slifire?

[Larvitar]

in Slifire?

In QPIFire

[initialised]

In QPIFire

Yeah, there so rubbish we have to plumb two of them directly into the QPI to get anywhere near 5450 performance!

[FischOderAal]

FUD? Why not just use one single bigger GPU core? Or is it going to be some sort of 2 CUDA cores, just not CUDA? The die's monolithic... Doesn't make much sense, does it?

This! I totally agree! I mean, you can barely use Intel's GPUs for anything else than office related work, wth would you need two GPUs for?! Unless they are "Larrabee"-cores which can be used for OpenCL-computation as well.

Doesn't AMD follow that path with their "APU"? Use the GPU for some parts of the calculations?

(Sorry if what I'm writing is a bit mixed up, had a few Mojitos and a little bit of that: http://img10.abload.de/img/img_3144fe2a.jpg )

[Iconyu]

If those two GPU's are Larrabee based, then thats a shed load of x86. They'll make nvidia nervous as it'll sit in the area where they are trying to sell Fermi at 2k a pop. If that's the plan Intel don't have to beat Fermi, they just have to not use 300watts.

[~CS~]

If true, likely a marketing gimmick from Intel.Advertising dual core GPU's after success of dual core CPU , could be a good strategy.I already see an advert explaining what GPU is , and how doubly awesome this one is.