Parallella: building for a parallel computing future
This post takes a look at a development board which is inspired by Arduino and Raspberry Pi but with a very different goal: to make parallel computing accessible to everyone.
The Arduino and Raspberry Pi are incredibly flexible and may be put to many different uses, and there is no shortage of similar boards — many of which inspired by them — being announced with slightly different features. But when was the last time you heard of a low cost board which aims to make parallel computing accessible?
Dual-core ARM 9 with a multicore accelerator
The Parallella board will be based around the Epiphany multicore accelerator from Adepteva, and will provide 16 or 64 floating point RISC cores that are clocked at 1GHz and that are connected together by a low-latency on-chip mesh network. The board will also include a dual-core ARM A9 CPU and this will run Ubuntu Linux and act as host to the Epiphany chip. The system will have 1GB RAM, microSD storage and ports will include USB 2.0, Ethernet and HDMI.
From prototype to product
Epiphany-based hardware has already been prototyped but it remains for this to be turned into a finished design, and chip product retooling or “full mask tapeout” is also required in order to bring the cost of the Epiphany processor down to the point where a $99 board is possible.
A Kickstarter campaign is being used to raise the funds for this work and to enable an initial bulk production run, with a $99 pledge securing a 16 core Parallella board in return. Other rewards include a 64 core Parallella board for a $199 pledge, and four 64 core boards pre-configured as a cluster for a $975 pledge (the 64 core Parallella is subject to the stretch funding target being met).
A commitment to open
Adapteva promised that they would release the documentation for the Epiphany architecture and its SDK if the campaign reached $500,000 by 7th October. This target was not met but Adapteva decided to take a leap of faith and have released both of these.
The Epiphany SDK is based on upon standard open source development tools including the Eclipse IDE and the GCC C/C++ compiler.
Parallel computing matters
Many suggest that the future of computing is parallel, as scaling performance through increasing processor clock speeds and symmetric multiprocessing will only take us so far. But parallel computing requires that programmers think differently and introduces new problems of its own. In providing low cost hardware, open access to detailed technical documentation and an open source SDK, Parallella will enable a much wider community to attempt to address these enduring challenges.
Top image: The EpipanyTM Multicore Solution (© Adapteva)