One way is the commoditization of computing: actual computer power is now a commodity, with very low prices and standardized interfaces. Ethernet is the new pallet, and servers are trucks.
On the other extreme, the embedded landscape demands catering for each application, demanding a lot of GPIOs, specific hardware and hand work. It gets expensive.
Of course, the Parallella just shows this in such a nice way because it evens the (massive) computing power in all offerings, that differ only in interface. Actual consumer products will try to balance this, but it still tells us a history.
The fact that price goes down as the application goes
from server to desktop to embedded is very telling, in
several ways.
Basic marketing trick. Always name your products as "good, better, and best" in some form. Unfortunately, it feels slimy to me, and I don't like it.
That said, I do like the idea of the Epiphany chip they're offering. But based on cost and performance alone, its clear that a $120 AMD R7 260x graphics card will be superior to what they offer here. (With its 14 compute units at 1.1 GHz, can do 896 SIMD integer or single-precision floating point operations per clock). R7 260x also has Windows / Linux drivers and OpenCL support included...
Epiphany is doing a disservice to themselves if they are trying to compete against "desktop" and "server" computers. Their niche is in their performance / watt. GPUs, with their 5GHz+ GDDR5 RAM, ~GHz clock speed, and super-parallel architectures will continue to dominate supercomputing at the ~100W to ~500W levels.
Epiphany IV is a supercomputer design at ~2W. Anything more is settled by the current laptop market. (AMD Kabini hits 150 GFLOPs at ~25W for only ~$60 CPU on a $30 motherboard)
The name "server", "desktop", and "embedded" are marketing terms. They're all the same CPU.
The server version is cheaper because it leaves off the HDMI port and associated circuitry.
They certainly have some margin, and it's probably bigger at the embbebed version, but it's not only a marketing trick.
I backed the original Kickstarter, have a bunch of Parallellas, and due to a cash crunch have to liquidate a bunch of them. Above is the link to my Craigslist ad.
These have equivalent specs to what Adapteva is calling the "embedded" version, but they come with the older, smaller heat sink and thus will still need an external fan.
Note that I'm both selling below Adapteva's list price and that the version I'm selling is marked "out of stock" on Adapteva's site. :-)
How do you program the FPGA? I' asking this because I have a heavy OO background and even when C is not entirely strange to me, I'm not sure if it's as easy as to make a weekend project with that or not so much.
Your thoughts on this? Did you chain together more than one?
Thanks for your comments!
The crew at Erlang Solutions is working on running Erlang on the Epiphany chip, but I haven't seen any visible progress from them.
I can't tell you anything about programming the FPGA. You don't need to directly mess with the FPGA to use the Epiphany. From what I've seen, the jump from C to VHDL is even bigger than the jump from OO to C, so probably not a weekend project. Also, there's no FOSS toolchain for programming the FPGA--you have to use Xilinx's tools, about which I've heard mixed reviews.
I bought the cluster with the intention of, well, building a cluster, but to date haven't actually tried that--Real Life got in the way of doing that. I do have all the hardware sitting here to connect them over plain old Ethernet.
Still need to get the bugs worked out of my Xilinx on Ubuntu setup though. Annoys me to have to run Windows in a VM to use the toolchain.