Intel Larabee

[_tommo_]

Ok now we can read something about rasterizing on Larabee.

Reading now.

reading later. It is indeed very interesting...

but i wanted to see some screenshots

[jjp]

http://www.anandtech.com/weblog/showpost.aspx?i=659

How can I put it? I told you so Nevertheless, it will be interesting how their next shot will look like. Still no fixed function rasterization logic?

[Kojack]

Hmm, I wonder if Tom Forsyth is going to reveal any of the details on why the first Larrabee is cancelled.

[jjp]

Hmm, I wonder if Tom Forsyth is going to reveal any of the details on why the first Larrabee is cancelled.

I guess it is quite simple: 300W TDP and no competitive performance

[Klaim]

Well I'm not surprised, it's common to fail on the firs implementation of something "new". Personally I'll look wait 2 or 3 years again before thining it's a dead project.
Anyway, it's more the concept than the product that is interesting : if nvidia or AMD can do the same better, then it's all good

I just hope it will not be like the AMD k10(?) architecture failure (if my memory is right).

[jjp]

I am simply not surprised because the concept is at least in 3 respects seriously flawed:

- Trying to beat highly optimized fixed function rasterization logic with software running on general purpose processors.
- Not even having dedicated scheduling hardware.
- Keeping the overhead for the x86 instruction set, in a case where this is clearly not beneficial and proportionally also more expensive than it is for a normal CPU.

And I am slightly satisfied with their failure, because of their arrogance to generate all this hype without having a product.