It delivers up to three times more performance per watt than the Fermi streaming multiprocessor, making it possible to build a supercomputer that delivers one petaflop of computing performance in just 10 server racks.
The Brahma15 is a 21, 000 DMIP Quad-core ARMv7-A instruction set compatible multiprocessor. 32KB instruction and 32KB data caches per processor are backed by a shared 2MB L2 cache and feed the multi-issue, out-of-order superscalar 15-stage plus write back pipeline of the Brahma15.