The 64-core Parallella is alive
adapteva.com
adapteva.com
Most of the folks like me who would have bought something today while reading about it will just forget about it later. These guys are missing out on a huge opportunity.
They're not likely making money on the Parallella - in fact a lot of the delay for the Kickstarter campaign was due to issues related to cost (e.g. the design is cut to the bone, and they managed to eventually get very good pricing from Xilinx for the Zynq etc.). It looked like they were in trouble for a while until they got a cash injection from Ericsson and Carmel Ventures early this year.
As such, while they'd certainly benefit from more exposure, and getting it in the hands of more people, they also have every reason to manage the process so building boards doesn't get in the way of actually evolving their chip designs etc. too.
Since they are relatively low volume, it seems to be pretty hard for them to get the necessary parts from suppliers reliably. I think they need a huge customer for just the chip (and hence plenty of working capital), or a large investment infusion to be able to deliver more boards in high volumes. They seem to be focused on finishing out what they have sold so far before committing to any new sales, which is probably a good thing since it's taken them so long to just fill the kickstarter orders.
I would expect to see them on HN again in the future if anything positive happens, it's how I found their kickstarter, after all.
I'm still waiting on my Coin.
It's possible that we can thank the FTC for this:
The Federal Trade Commission’s (FTC’s) Mail or Telephone Order Rule covers all merchandise ordered by mail, phone, over the internet, or via the fax machine. It stipulates that, if a merchant does not promise a specific delivery time, the merchandise ordered must be delivered within 30 days of the merchant’s receipt of the order (or the date merchandise is charged to your credit card). If the company is unable to ship within the promised time, the company must give the buyer the choice of agreeing to the delay or canceling the order and receiving a prompt refund. However, if you are applying for credit to pay for your purchase and a company doesn't promise a shipping time, the company has 50 days to ship after receiving your order.
http://www.hcs.harvard.edu/~scas/wp/wordpress/?page_id=24
In other words, in the United States, it is not legal to take pre-orders, and incur a delay, without offering customers their money back. So 100,000 pre-orders could come in, and in case of a delay -- since they are forced to offer a refund -- they could lose, potentially, all of their funding.
Imagine ordering parts for 100,000 boards and having 50% of your customers take their money and run, in case of a delay. That's a fairly unmanageable risk.
Well assuming they're planning to produce another batch, they're already incurring some risk anyway. Allocating units out of that batch as pre-orders doesn't cost them anything, and probably improves overall sales.
Thanks for pointing this out.
Edit: Ok, here's my quick summary. Please correct me if I'm wrong:
This looks like a small PCB (raspberrypi-alike) that sits the main attraction: a 16- or 64-core Epiphany coprocessor, as well as an ARM cpu to run the OS. Not sure how these relate in performance to other coprocessors (GPUs with OpenCL?). Power draw seems low (5W). Would love to read more about the architecture, why Epiphany processors are special, etc.
The main CPU is substantially faster than the Pi, but it doesn't have HW accelerated graphics, so it's not a speed daemon for desktop/workstation type use.
As for the Epiphany, assume that it'll be slower than most GPU's for tasks that GPU's are good for. That is, if you can make do with few instruction streams, the Epiphany is not well suited for it, as most GPUs will blow the current chips out of the water in terms of performance.
If, on the other hand, your problem is poorly suited for GPUs due to lots of independent instruction streams, it may be better suited.
One of the most interesting aspects of the Epiphany is that is can also be connected into a grid - each chip has four high speed links that can be connected to other Epiphany chips, or be used to interface with the main CPU or off-chip memory.
The cores can all access each-others memory without any special instructions, including that of the cores on other Epiphany chips that are hooked up via the external links - the only difference between in-core and out-of-core memory access is the speed.
You can address core-local memory, memory in another core, or main memory the same way - the difference is speed.
There's no cache, and you're responsible for avoiding race conditions in memory access yourself.
EDIT: And if you want to read/write the same memory areas from multiple cores, it's your own responsibility to either get the timing right, or use other means (e.g. you can trigger interrupts in another core) to signal when it is safe for another core to access data.
In other words, could I write an epiphany webserver?
I get that GPUs are a bad fit for many problems, but how fast can this thing be if it has no local cache? Will this really be faster than a high-powered Xeon, or will it just consume less energy per operation? Is it a teaching tool? A tech demo? Something that makes Erlang/Haskell magically run much faster?
https://www.kickstarter.com/projects/adapteva/parallella-a-s...
I have a great amount of respect for the Parallella team: to be able to kickstart a custom chip, that promises very interesting applications, and deliver it within several months of the estimated delivery date with the setbacks they have had is absolutely astonishing for me. While I can't comment on the quality of the final product yet, I would say that they know how to run an excellent campaign.
I'd be curious if it beats GreenArrays (http://www.greenarraychips.com/) numbers of picojoules per operation. I wonder if those numbers are published for Parallela?
From reading the parallella docs, it looks like that chips runs 5 W on a "typical workload" while the GA144 runs .25 W at an absolute theoretical maximum, for a 20x difference in energy consumption.
You can program in floating point (I'm doing just that, with 6 cores performing Karatsuba-3 multiplication of 54-bit elements) but that is quite a bit slower than hardware DP multing (which Parallella boards lack too). FP multing will likewise be slower than hardware FP multing, which Parallella does have.
They're more similar to a GPU than to the Epiphany. Each SPE is more powerful in terms of Gflops, but the Epiphany CPU's offer more independent instruction streams. If your problem is basically well suited for a GPU (easy to vectorize) chances are it will probably do better on a Cell than the current Epiphany's. If your problem has lots of independent branching, the Epiphany stands a better chance.
I see this thing does OpenCL on a completely different architecture. Recent tesseract ocr versions supports opencl. Will I be able to run tesseract on this thing? Would I even want to?
also, OpenCL apparently isn't the preferred programming model for the epiphany, so performance wont be as good as bare-metal (but that is practically a tautology for any OpenCL device..)
See: http://en.wikipedia.org/wiki/Reversible_computing
Note that this is the computing equivalent of a straight line.
Related question: is Erlang running on Parallella yet?
I remain very interested - 64 or even 16 cores on a small form factor would be incredible.
I don't have the video cable yet and still have to attach the heat sink. They'd really like you to have a fan, so I have some work to assemble one.
I can recommend this kit for a case/fan: http://shop.abopen.com/ - that fan is enough to keep everything feeling cool to the touch.
The design is on Github too, if you want to do your own (and if nothing else the instructions shows you how to hook up a 5v fan directly to the board)
It takes some assembly and very light soldering, but not more than that you can get away with any crappy soldering iron and ideally a pair of wire cutters (but scissors will do)
[0]: https://cdn.shopify.com/s/files/1/0230/6005/products/Case4_g...
[1]: https://cdn.shopify.com/s/files/1/0230/6005/products/case3_3...
Edit: I just noticed that the two cases I posted actually differ. Perhaps the design was not yet settled and they have added provisions for a fan. Does anyone know what the cases that are actually shipping look like?