Larrabee delayed to 2011 ?

You're hopping for a virus?

 

I'd like to take this moment to copyright the term VD before someone else takes the encentive... Yep, virtual disease.

This just in... Whaddaya mean VD is already taken?!?

 

Sorry for the OT, but I just spilled my coffee all over my desk while reading it

 

http://www.brightsideofnews.com/new...ient-truth-intel-larrabee-story-revealed.aspx

To the IMG folks reading in case you haven't noticed your designs are utterly inefficient but because of that Intel should buy you out

 

Let me guess what BSN really stands for.

 

It's actually BS^N..

 

You're going to design digital teapots with an intel inside sticker in the future

 

I am always struck how these guys are completely unable to do research on their own. For them, research is exclusively asking other people who they presume ought to know something. Actually looking at public information in addition to that is unthinkable; why would you ever do that? It's a serf's job! (this is much more applicable to Fudo and Theo than Charlie, although even for him I'm not convinced the ratio is optimal).

The claim that a "PowerVR core at 4 GHz could be powerful enough to start competing with low end and mainstream parts" is especially hilarious.

 

I won't go too much into the article. I don't have any insight into whatever drama may have existed with contractors and tax schemes or an airliner.

That there appear to have been troubles in development, I think most observers would agree.

There are some glaring technical pionts of the article that conflict with already disclosed data, which without elaboration on the author's part shows dubious understanding of the topic.
It's like seeing a baseball reporter blanking out on what an ERA is.

As an example, I've not seen any indication that Larrabee uses AVX, and every single time the article says Larrabee does, my suspension of disbelief is shattered.

Maybe I'm old-fashioned, but this really doesn't seem to be a good way to operate.

On another note:
http://www.anandtech.com/video/showdoc.aspx?i=3651&p=7

Anandtech's graphics coverage has appeared a little wobbly in some respects, but I am curious about this point.
There's been a halo of possibly contradictory quotes about Larrabee's target range, and what chip is supposed to come out. We've had Intel execs promise mainstream penetration, then balls-to-the-wall performance, then power-efficiency.

Now Anandtech gives this little tidbit of marketing positioning.
I have some difficulty imagining that huge 32 core chip we saw photos of potentially sliding into the same price bracket as Juniper is right now.
Just how wobbly should we consider Anandtech's grasp of the topic?

 

The BSN link, like the SirEric line I think.

 

My take on it?

http://www.semiaccurate.com/2009/07/27/plagiarism-rampant-it-journalism/

http://www.semiaccurate.com/2009/07/27/when-caught-dont-destroy-evidence/

All my checks say Q2/10 and possibly slipping a little on software, not hardware. If B0 turned out as they think, and if anyone can get simulation right it is Intel, then all should be fine for mid-10, drivers willing.

-Charlie

 

Well there have been quite a few changes regarding his stance for IMG; wonder who had his fingers in that one.

 

Yeah. But you were in the room when I asked that question. Why didn't you bring it up then?

Seriously though, I asked about the microarchitecture, and a power efficiency based on implementing the ISA, not a specific CPU.

-Charlie

 

My point is that the overhead is very different for something like Atom vs. Nehalem, vs. Sandy Bridge vs. something else.

David

 

I find this presentation some may find some interest to it. It's about High-Fidelity Volume Rendering and how they map to CPU, GPU and larrabee.

 

I wonder why their splatting implementation sucks so much, there is nothing inherent in what they are doing to explain why there should be a difference in the rendering between raycasting and splatting.

 

Indeed, there is nothing inherent in splatting to match the quality of volumetric ray-casting (VRC) as soon as the sampling-density is high enough. It is true that to match (VRC) splatting needs slightly lower sampling density but savings in number of samplings does not compensate the computational overhead. The statements above just represent the result of my experiments with splatting and VRC clearly it does not preclude someone from having more luck with splatting. Still I'm not aware about splatting implementation which matches the best VRC; if you know such example please provide the reference; I mean not just article with claims but verifiable reference I may run on computer.

 

Not saying it would be faster, just strange that it had such huge artifacting.

Although I wouldn't call the subvolume rendering used in that paper (and Voreen) pure raycasting anyway. It's hybrid image order/object order.