GPU industry-watcher Jon Peddie offered up some compelling stats in his latest blog. Peddie’s research group looked at the installed base of x86 CPUs and GPUs and calculated the total number of instruction cycles available on each architecture. If the study is even close to accurate, the results are stunning. According to the research, in 2009 there were almost 450 billion compute cycles available every second (38 quadrillion a day) , and the vast majority of those cycles — more than 95 percent — came from installed GPUs. Better yet, the installed processor base seems to be growing exponentially. Peddie writes:
A general figure of merit is to multiply the processor’s clock by its word size. The GPUs running at one fourth the speed of a CPU, and with just a 32-bit processor compared to the 64-bit CPU still deliver the most MIPS because of their overwhelming number of cores.