Established chip company Adapteva has launched a Kickstarter fundraising campaign to create a low-cost parallel chip board for supercomputing — a kind of Raspberry Pi for parallel programming.
Where Raspberry Pi's mission is to create a low-cost but capable computer for education, Adapteva's goal is broader: it aims to spark application development in the parallel computing industry by creating a low-cost board for developers to experiment with.
"We know that parallel computing is the future [but] it hasn't taken off yet," Andreas Olofsson, Adapteva's chief executive, tells me. "If we look at all the efforts in parallel processors and processors in general, there are very few winners."
By example, there are few applications outside the high-performance computing space that are capable of scaling their workloads over many cores, Olofsson said. Chipmaker Intel has also noticed this problem and produces a variety of FAQs and software packages to educate developers in the art of parallel programming.
However, there need to be more developers skilled in parallel programming for future applications to get the most out of the chips they run on. This is because in the mid-2000s, chip companies stopped being able to economically deliver significant clock rate increases on CPU cores and were forced to move to a multicore strategy.
But the number of applications that can effectively use the four, eight or 16 cores of modern chips is few. This means that instead of taking this approach, we run multiple applications at once and each core tends to handle a different application or set of applications. At some point, we will need more performance for each application and the only way to do this will be to run things across multiple cores.
"If you've got a hundred cores, you've got to have a hundred threads," Gabriel Loh, a principal researcher at AMD, says. "If you don't, you're not going to have as much out of it... that is a challenge."
To solve this problem, developers need to write applications that can use many cores effectively, Olofsson said. Unfortunately, as there are few applications there is less money floating around in the parallel-processing industry, so there are few developers and therefore less demand for the processors.
"It's truly chicken and egg," he said.
To deal with this problem, Adapteva has launched a Kickstarter appeal — 'Parallella: A supercomputer for everyone' — to fund the construction of low-cost parallel programming boards based around its 64-core Epiphany RISC chip.
Each board pairs a dual-core ARM A9 processor with a 16- or 64-core Epiphany Multicore Accelerator chip along with 1GB of RAM, a MicroSD card, two USB 2.0 ports, 10/100/1000 Ethernet and an HDMI connection. In addition to the hardware, each board will ship with a set of open-source development tools for the Epiphany chip and the Ubuntu operating system.
The low-power boards should consume around 5W and deliver 45GHz in equivalent compute performance, if all the chips are maxed out.
The $750,000 in Kickstarter funding Adapteva is looking for will go towards reducing the size and cost of the platform by retooling chip foundries and putting more money into chip testing and board assembly.
The project, if funded, will see the company produce boards costing $99. The fate of Adapteva's scheme will be decided when the funding deadline expires on 27 October. As of Tuesday, with 11 days to go, the project has raised $359,525 in pledges.
Besides Adapteva, Tilera and Intel both have chips in this area. However, Tilera's 100-core chips were launched over a year ago and their most recent processor scaled down to nine cores to focus on embedded applications. Meanwhile, Intel's Xeon Phi co-processor is designed for HPC, rather than general parallel computing.