Royi (Who@Cares.com) on 9/28/10 wrote:
---------------------------
>Jack (jumpingjack4@hotmail.com) on 9/28/10 wrote:
>---------------------------
>>Royi (Who@Cares.com) on 9/28/10 wrote:
>>---------------------------
>>>JS (JS@NOSPAM.COM) on 9/28/10 wrote:
>>>---------------------------
>>>>Where from do you get the 35% figure?
>>>
>>>
>>>What's ATI and nVidia Graphics Card market share combined?
>>>I said 35% to the least.
>>>Better think of it, what's Intel's share?
>>>
>>>I think Intel should have kept it the way it was with Nehalem.
>>>i5 series will have built in GPU. i7 won't.
>>>In this case not just disable it but either reduce price or put something else in which improve performance.
>>Essentially, they are.... I suspect.
>>
>>However, I do not really see this as a crack for AMD to exploit since they are
>>pedal dawn, racing as fast as they can to the exact same thing.
>
>The way I see it, AMD take the integration in a different approach.
>Their dream is that the APU or whatever they call it will take care of all the
>Vector / Parallel Operations and the CPU will take care of the Serial Operations.
>In their dream, one day all the "Future" SSE / AVX or whatever operations will be computed on the VPU.
>If you buy a discrete Graphic Cards still the VPU will be the "Parallel Co-Processor".
>
>Intel on the other hand just create a "Classic" GPU.
>If I want a discrete GPU I don't want to pay for a piece of silicon I don't need
>or use. I'd rather pay less for the same (Or similar performances) or have something
>else to improve performances instead of the GPU on the CPU.
>
>In my opinion, it's a crack.
>Only for a third of the market, yet it's still a market share bigger than AMD holds now.

Ohhh, I misunderstood you... I thought your 'crack' was that Intel was offering this GPU-CPU product with something extra you didn't want to pay for. My point was AMD is rushing to do the same thing... and they are.

AMD is good at inventing new terms, but the fact is that Fusion or the APU is essentially the same thing, a classic GPU melding into the same die. Where I see a crack is that AMD is DX11 compliant and OpenCL compliant, where as this revision of Intel's version is only DX10.1 compliant.

That does not change things terribly much though, the direct compute engine in DX11 runs on DX10 compliant hardware, Intel's EUs are still programmable just like any shaders in the typical discrete VPU as you put it.

If I were to catagorize this as a crack, it is in that regard. However, Intel seems willing to try and fill much of those cracks --- how successful they are is really a big question mark. There are two killer applications for GPUs in the market that this version of Sandy Bridge (and AMD's future Llano) service -- 3D gaming and video transcoding/encoding. On the 3D gaming, Intel has made substatial progress but they will most certainly still lose to Llano when it finally launches. The transcoding, however, is a different beast. Intel is providing a fixed function dedicated block -- agian, how successful this may be remains a huge question, but ultimately it closes one of those cracks -- especially since it is likely the most energy efficient means for covering this usage model.

Ultimately, the market will segment or fracture along those lines, as you suggest. Intel will be launching a gpu-less sandy bridge into the high end, because as good as an integrated GPU via Intel SB or AMD Llano, they will not supplant the market that craves the extreme performance of a dedicated discrete graphics solution.

Now, onto your 'dream'.... to say that an x86 SIMD or newer AVX is going to translate over and execute on that VPU seems like a daunting task, to say the least... perhaps David can comment on the feasibility.