I continue to be unable to recommend ATI, not because of their hardware, but because of software support. Why has such a fundamental feature as asynchronous PCIe transfers been put off for so long? Many algorithms spend similar amounts of time in computation as they do in transfer, and DMA support would double the speed of these algorithms. As we are using GPUs to do RAID (a fairly simple calculation, so it falls in the class of almost as much transfer time as computation time), the lack of asynchronous transfer support is a huge blow. Is there any estimate at all of when this feature may be added? ATI has some good things going for it, particularly EDC memory and unhampered double precision floating point performance, but I'm afraid the lack of DMA support is a deal-breaker for our application.