[QUOTE="Bebi_vegeta"][QUOTE="Teufelhuhn"]
That's not what I meant when I said "bypass DirectX"...I mean not using a 3D graphics API at all.
On PC's OpenGL and DirectX do the same thing: they serve as a sort of translator that you use to talk to a video card. You see graphics hardware each has it's own "language" that it speaks. So when you make a PC game and you want to draw a triangle on the screen you don't program it to say that in all the different hardware "languages", you use DirectX or OpenGL. You tell that API to draw the triangle, then the API translates it to the "language" that the hardware understands. This is how things have been for a very long time...it's nice in that you don't need to worry about individual kinds of video cards, but it was limiting since you could only do the things that DirectX/OpenGL knows how to translate. You're also limited because video cards aren't very flexible to begin with: they are only designed to do things like drawing triangles or use textures.
Now with Larrabee things are a little different. It does support using DirectX or OpenGL if you want to do that and go through the whole translator bit...but if you want you can skip that. This is because Larrabee uses x86 cores: the same kind of cores that are in a Core 2 Duo or any other Intel/AMD CPU. So this means, if you wanted, you could talk to the hardware directly to do your graphics stuff. It would be harder to do and any code you wrote for this would only work on Larrabee and not Nvidia/ATI, but you would have a whole lot less restrictions. horrowhip
I still have my doubts on larrabee... GPU are that more powerfull in terms of calculation.
Intel will just scale the number of cores to match the competition...
Ultimately, if they need more than 32 cores, they can just go up to 48 cores.
Keep in mind that even the 32 core version still have about 1/3 the die size of an nVidia card, and is slightly smaller than ATi cards.
And that is in the 45nm process.
That is why Intel has such a huge advantage. Their design scales almost perfectly.
1:1 scaling is relatively simple with it.
The problem with massive amounts of cores is the power demands. It is likely, if they REALLY need to, they will launch with a 48 core version. That increase the theoretical power of the card by 50% in a single swipe...
And Intel has proven that they can do 32nm process REALLY easily and 22nm process with very well.
So, if they were to keep scaling down they would reduce power demands, increase number of cores and scale very well.
Intel will release with whatever they ultimately need to release with power wise.
Thing is, the 32 core version shows absolutely no signs of being less powerful than whatever nVidia or ATi can possibly cook up(unless nVidia has some rediculous secret GPU that increase performance compared to the GTX 280 by like 100%.)
What we can expect in the next year from nVidia and ATi is probably somewhere in the 50-60% range in terms of power incease and somewhere between 40-50% in performance increase.
Based upon what Intel has already proven, Larrabee should (realistically) be anywhere from 40-50% faster performance wise, than the GTX 280.
Hd4870x2 has 2tera flops of calculation power. No intel/AMD CPU comes even close to that power.... folding at home will show you perfectly what i'm talking about.
We won't see a 32 core intel very soon... 8core is supposed to be out by 2010.
Nvidia and Ati are also scalling down on to 45nm or 40nm very soon.
Seems to me that the new road for GPU is also going mutliple GPU on the same card.
Log in to comment