As someone who has used Larrabee I can confirm it already is a QPI device, and uses a QPI to PCI-e bridge like the new i5/i7. This is Intel's silicone building block concept.

The battle between CPU and GPU vendors is endless, as each wants to maximize its share of user spending. In the end, any efforts to harmonize are kind of like multifunction printers - i.e. not very good at anything. On-die integration is even a mixed bag - less choice in CPU vs. GPU performance, competition for memory bandwidth, and extremely high power consumption.

More articles like this please.
Not that other junk to make the various fan boy factions run amok.
I'm slowly losing my desire to come to this website.

in BSN zine, intresting article, although 32 nm roadmap is ok, some question as to when DX11.1 will be ready, Over year, maybe 1 &1/2.
drashek

http://www.brightsideofnews.com/news/2009/11/2/amds-next-gen-gpu-manhattan-and-northern-islands-use-32nm-process.aspx

Why is anyone taking Larrabee for granted? It is still pretty much vaporware and considering Intel's "superior abilities" in GPU engineering, it's first generation is going to be a failure compared to existing offers from 'alternative vendors'. There are just too many ways to FCUK it and I bet Intel is not going to get it straight the first time Larrabee is around.

Taking this vaporware as competitive to even current GPUs is probably not very wise and that's why this INQs article is not right. At least the 2nd part. Should have used a better judgement.

a nice Quad core surrounded by 128 mini me Pentiums, all with HT, i have a 6600 just now and its fine, im going to break tridition and not upgrade till that thing arrives.

I wonder if Windows will recognise the Pentiums as CPU cores, and if so can windows display 264 cores in task manager?

In the current adolescent fear-of-incompatibility, what -will-people-say IT world this money-driven phobia is halting the previous almost childish groundbreaking concepts.

Where are the good few men that dare to go double or nothing, boom or bust ? But no, it's Industry Standards (tm/r/c) that 's behind the wheel.

With people like Hector at the rudder worrying more for making bucks on insider trading then technology its half a miracle Istanbul actually left the gates of Santa Clara.

Who cares if he's left the building; the damage has already been done; and that's why there are NV cards on PCIe on my dualsocketed Opteron board. With an Ageia PhysX card to boot; that was the only relevant development the last 36 months.

Blerch. Shame on you IT sector.

Much More coherrent than article 1. Scan Line Interface may Be Missing word.

Scalable Line Interface should be increased from 2 to 3 sections per screen, as 24 screens isn't practical, push that excess into one screen.

With each card running less pixels better, SLI might break thru ALL dumpy Intiiala & Really Change performance.

By Giving U of Minnesota to Sir Hector Ruiz Be same thing. Call it: Ruizopolis or San Ruiez.

drashek

They were thinking about this in the initial fusion tech discussion. Well, all kinds of accelerators, graphics / GPGPU must have slipped their minds. They later dumped this for a reason known probably only to them.
So I don't really expect AMD to do this even if it worked like you said. And that it would.

Now that is some fine Inquirer journalism, not the usual ironical BS about nothing important. Keep on the good work.