Dubbed Many-core Integrated Accelerator of Wisconsin (MIAOW) the GPGPU is still in its kitten stage and needs a few developers and a couple of months on a newspaper before it is properly toilet trained.
Karu Sankaralingam, an associate professor of computer science at the University of Wisconsin-Madison led the team that designed MIAOW. He can see a day when big companies will be built using open source hardware, just as multi-billion-dollar Web giants owe their existence to open source software.
Sankaralingam added: “We really need more people contributing to open source hardware to improve the platform layer so there’s enough for entrepreneurs to build from it,” he said.
It all started when one of Sankaralingam’s students was working on an out-of-order processor design when AMD released as the instruction set architecture of one its graphics chips. That sparked the idea to build an open-source GPU.
It took 12-people 36 months to develop the MIAOW core. The idea was to create a functional GPGPU without setting any specific area, frequency, power or performance goals.
MIAOW uses just 95 instructions and 32 compute units in its current design. It only supports single-precision operations. Students are now adding a graphics pipeline to the design.
An in depth paper on MIAOW is available online but requires special access.
MIAOW compares favourably on several benchmarks to AMD’s latest high-end chip, Tahiti although obviously not all of them. Apparently AMD did look at the design and said it was not completely crazy.
However there is a small matter of potential for patent infringement law suits for anyone using MIAOW and it will be up to any open source community to sort that out.