I am using an AMD Llano APU to calculate some reasonably sized Linear Algebra problems. I can get the low-cost memory copies working through using the Map and UnMap fine but ... the problem sizes must be small. So I am guessing this is for them to fit into the 128MB on the GPU?
Is this correct - if I load too much data into a buffer than the kernel fails at submission time due to an out of resources error.
It seems a real pain that we could not fill up system memory and then use the GPU to process the data.