Thats great, but anecdotal evidence doesn't really prove anything. CUDA lets you write multi-core code for GPUs, but imagine writing C code for hundreds of cores over a datacenter cluster. C isn't the right tool for that job.