Intel's CPU+GPU For Desktops And Notebooks To Arrive In Q2 2009

[jimmysmitty]

Santa Clara (CA) - AMD has been pushing its Fusion idea for quite some time. The company claims the Fusion concept was one of the main reasons why they acquired ATI in the first place. The timetable for Fusion has been set, but Intel is working on a similar CPU/GPU combination and we are hearing that Intel may have its product earlier in the market than AMD, TG Daily has learned.

AMD's Fusion is scheduled to arrive sometime in the second half of 2009, primarily targeting notebook applications. By the way, the dual-core CPU with an integrated ATI RV7xx core is codenamed "Black Swift", while the single-core with the ATI RV7xx core is known as "White Swift". There is no information on possible tri- or quad-core versions of Fusion at this time.

If you have been wondering what Intel is up to, we have reported previously that the company has been working on a similar CPU/GPU approach. Based on the Nehalem architecture, Intel will be rolling out two version of such a processor, Havendale and Auburndale.

Both Havendale and Auburndale should come to market in first half of 2009, were were told by industry sources. Both processors will combine two CPU cores as well as an integrated graphics subsystem based on the G45 successor. We can't call Intel's graphics solution really GPUs, since they do not support some key features of DX9 and DX10 APIs. Since these chips are based on Nehalem, they will also support Hyperthreading (Intel's original virtual dual-core approach that was introduced with the Pentium 4), which means that the processors will be able to handle four threads total.

What makes Intel's approach interesting is the fact that Havendale is a desktop CPU and Auburndale is a notebook CPU. Both feature an integrated dual-channel memory controller supporting DDR3-1333 memory, but neither of them has QuickPath, Intel's shiny new Front Side Bus: Auburndale will support QuickPath when connected to a NorthBridge chip from the Ibexpeak-m platform.

There is also a difference in processor sockets - Havendale will be placed on a new socket for Intel's desktop platform, LGA-1160; Auburndale will use the brand-new mPGA-989 socket. The thermal design power 75 watts for the desktop part, and 45/55 watts for the mobile part. Both processors will be manufactured in 45 nm. These power numbers may not sound that impressive, but consider that the first standalone dual-core Pentium was rated at up to 130 watts without graphics, and the progress becomes obvious.

Ibexpeak and Ibexpeak-m will be known as Intel's desktop and mobile platform for 2009, with Ibexpeak most likely being branded as "5 series" (P5x, X5x), while Ibexpeak-M will get a Centrino name.

The key difference difference between AMD Fusion and Intel's Nehalem + VGA approach, as it appears today, is the fact that AMD is expected to actually use a real and very capable GPU. So, if Intel wants to make anything out of their CPU+GPU project, we would hope that the company throws its current products in the toilet and starts working on a new and more capable 3D architecture and drivers. Something like this here.

Nehalem cometh.

Only thing here is that obvoiusly TGDaily has never heard of Larabee. And they are quick to critisize it compared to Fusion when AMD has a real GPU maker in its pocket. But I doubt Intel is dumb enough to not produce a decent GPU for it. We shall see.

 

[chookman]

Intel will definetly get a decent GPU in there... i believe the ones they are talking about here are the cheap "all-in-one" for office and buisness use, so therefore will not require the larger GPU processing ability.

 

[jimmysmitty]

Intel will definetly get a decent GPU in there... i believe the ones they are talking about here are the cheap "all-in-one" for office and buisness use, so therefore will not require the larger GPU processing ability.

I have a feeling it will be based on Larabee. Intel already has the G45 coming to support full 1080p HD. Maybe the business chips will have low end but us end users might get something decent.

 

[wolverinero79]

"We can't call Intel's graphics solution really GPUs, since they do not support some key features of DX9 and DX10 APIs"

Really? I thought GPU meant graphics processing unit. I didn't know GPU meant "must support all high end APIs".

"So, if Intel wants to make anything out of their CPU+GPU project, we would hope that the company throws its current products in the toilet and starts working on a new and more capable 3D architecture and drivers"

He clearly is missing the point of the product. Dual core instead of quad or 8 core chip with an IGP based chip. Hmm....I'm sure Intel is targeting the high end market.

I'm so glad THG hired the two inquirer writers, aren't you?

 

[jimmysmitty]

"We can't call Intel's graphics solution really GPUs, since they do not support some key features of DX9 and DX10 APIs"

Really? I thought GPU meant graphics processing unit. I didn't know GPU meant "must support all high end APIs".

"So, if Intel wants to make anything out of their CPU+GPU project, we would hope that the company throws its current products in the toilet and starts working on a new and more capable 3D architecture and drivers"

He clearly is missing the point of the product. Dual core instead of quad or 8 core chip with an IGP based chip. Hmm....I'm sure Intel is targeting the high end market.

I'm so glad THG hired the two inquirer writers, aren't you?

LOL. Yes I am.

I still don't think Intel is that stupid. Do you really think they will integrate a G33 into it? Especially when AMD plans on a full blown out GPU? You think Intel is going to sit there and let AMD grab all of that market to themselves? I think not. Intel has something in the works and you know it. With their R&D it might be a nice suprise. Too bad we have another year before we can see it huh?

 

[TeraMedia]

Q: Does the embedded GPU itself need to support all high-end APIs for the aggregate chip to be capable of doing so? I mean, if you can write a driver that implements up to DX7 in software (just saw this linked to recently - impressive work by the driver's authors!), can't you write a driver that implements DX9 or 10 by combining the CPU's GP capabilities with the embedded GPU's FP vector/parallel processing capabilities? Yes, it would steal CPU cycles, but not as many as doing the entire render in CPU. And perhaps the benefits of not hard-wiring to a specific API outweigh the costs of the increased CPU overhead.

Obviously, I haven't researched exactly what the proposed embedded Intel GPU's capabilities are, so my point might be moot-ified after such research.