Originally Posted by EaquitasAbsum
Yes you have a point, but not as many applications use OpenCL or HSA, once again AMD is on the right track but they need to build within the what software is using now. They will never see software companies "catch up" when they have such a small market share compared to Intel. Sad, but true.
That being because it is hard to write applications to utilize OpenCL in its current state. HSA makes writing applications a walk in the park, as all you do is use the OpenCL SDK functions in your application after you setup a device driver. As I said, right now you have to work with two completely separate memory buffers for OpenCL to be even possible. One memory space in the system memory for the CPU, and another memory space in vram for the GPU. That doesn't make things any easier, when you have to pipe resources from your CPU to the GPU and back again over and over. With HSA, your application uses one memory space. And both the CPU and iGPU can use that same memory space. So for an example if I created a single byte memory space with my application like you would traditionally (CPU), and stored a byte value of 25 in it. I could then switch over to OpenCL and use the GPU and pull that value right from the same memory space. So communicating back and forth between the CPU and the GPU is seamless. Right now OpenCL is like two separate buckets next to each other full of quarters, to move quarters from one to the other. You have to take them out, and move them to the other. This is how memory works for OpenCL right now. With HSA, there's only one bucket and it holds all your quarters. It's one of them things where you wont see any applications utilizing it, until the processors actually hit the market. Because right now even the developers don't have these APU's, and you simply cannot write an application and debug it without using the actual hardware that supports it. So we're not sure how widely HSA will be adopted once it releases. Tho you can bet that applications written for HSA will crunch numbers faster than using equivalent discrete hardware, because there is simply no PCIe bandwidth limitation. I mean I get your point that the amount of software that uses this technology isn't that great, but what i'm trying to tell you is that is the whole reason why AMD made HSA.