I've been watching the Adapteva story unfold with great interest for several years now. This all started when Andreas Olofsson left his job, founded Adapteva, and -- working in his basement, living off his pension fund, supplied with sandwiches by his wife -- single-handedly invented a new, ultra-low-power computer architecture.
Olofsson went on to take the first version of his system-on-chip (SoC) -- called the Epiphany -- all the way to working silicon and a packaged prototype. (See my: From RTL to GDSII in Just Six Weeks!)
The Epiphany consists of an array of 32-bit floating-point processors, each with its own local memory, and all connected by a high-speed switching matrix. The whole device is designed from the ground up to perform humongous amounts of floating-point processing extremely efficiently while consuming very little power. In fact, the result is said to be 50 to 100 times more efficient than anything else out there in terms of gigaflops per watt.
Sometime later I heard that Adapteva had launched a Kickstarter project to create a personal supercomputer for only $100! Known as the Parallella, this credit-card-sized board boasts an Epiphany from Adapteva and a Zynq All Programmable SoC from Xilinx. And, just a few weeks ago, we heard the news: Adapteva's $100 Parallella Supercomputer Platform Now Shipping.
The credit-card-sized $99 Parallella supercomputer.
The combination of the Zynq and the Epiphany means that the Parallella provides a unique platform for developing and deploying the most compute-intensive applications.
The Zynq includes a dual ARM Cortex-A9 hard core processor with its own floating-point units and on-chip cache. This is ideal for running the Linux OS and/or an RTOS. The Zynq also provides a high-performance DDR memory interface along with a bunch of communications interfaces (SPI, I2C, CAN, USB, UARTů). And, very importantly, the Zynq boasts a large amount of traditional programmable FPGA fabric, which can be used to implement things like custom hardware accelerators.
Meanwhile, the Epiphany provides an array of 32-bit floating-point processing cores -- either a 4x4=16 array or an 8x8=64 array, depending on the type of Epiphany being used.
So, what can you do with the Parallella? To be honest, you are limited only by your own imagination. For myself, all sorts of embedded vision and embedded speech applications come to mind. As another example, consider the following one-minute video showing the guys from Adapteva building what they describe as "The world's most efficient Beowulf Cluster" in only 30 minutes:
Olofsson and Adapteva are on a mission to spread the word about the advantages of parallel computing. To further this aim, they have has just launched the Parallella University Program, or PUP for short. This is great news from universities' point of view, because the tools are free and everything about the Parallella is open-source. Also, there's an ever-growing Parallella Community at Forums.Parallella.org.
Of course, one key consideration is the incredibly low $99 price tag for the Parallella supercomputer itself, which means that many universities can afford to leap into the fray with gusto. In fact, Olofsson tells me that PUP already has 7,500+ orders from both individuals and from more than 100 universities around the world.
Having said this, Olofsson notes that the majority of these orders have come from well-to-do countries like America, Canada, Europe, and Japan. Sad to relate, $99 is still too much for academic institutions in many parts of the globe. Thus a key element in PUP is the fact that Adapteva will be donating one Parallella-16 platform for each 100 units sold via the Adapteva online store (universities eligible for the PUP must be actively involved in parallel computing research and education).
University representatives can apply at www.parallella.org/pup to sign up for the PUP program. Olofsson says that one of the PUP's goals is to maximize diversity in terms of research areas and geographical representation, so he's hoping to see universities signing up from all around the world.
For myself, I pledged $99 in the original Kickstarter campaign, so I'm now eagerly awaiting the arrival of my very own supercomputer!