Parallella Community

Just today I found that I might 'just not' be in the first batch of 1600 or so boards.

I'm a fervent opponent of the idea of cheap supercomputing, and still have the intention of doing actually interesting things with it, but isn't it a bit low-tech do deal with New Chip Technology and advanced Xilinx programmable logic and have only news about parts and added cases, etc ?

I mean I'd be interested in using lots of programmable logic and maxing out the communication channels, and at least academically work on sort of the "modern times" Connection machine. that of course requires HiTech, and not primary a sort of accelerated Raspberry PI with previous generation Pad technology, no matter how I appreciate and respect such effort (which in fact I do).

That makes me not worried about Andreas' great designer talents and abilities at all, but, like, I got a $100 early Blackfin DSP board early on too, and that was full of electronics. Not that that's a good comparison, but if the Adapteva board is going to get serious followup and business, it appears rather in trouble to me, unless it would be say a eastern European import export company or something.

Anyone care to comment?

T.V.

I'm seriously looking forward to see your HiTech projects!

I fully expect that we will come to see the full spectrum of uses — from those who backed the project simply because they wanted to see it succeed, to those with applications that are pushing the boundaries of computer science (among other fields!)

Remember, it's still early days, and simpler projects also have a vital part to play in helping to raise awareness of the platform, finding bugs and contributing to documentation etc.

Cheers,

Andrew

PS. I'm very much looking forward to seeing what you build with Parallella also

Well, it's like this week I received my Xilinx non-spam mail about their latest development in the Kintex-7 and so on area:
few dozen 16Gb/s (gigabit per second) transceivers with hundreds or even thousands of (pretty capable) DSP slices might be more attractive, though possibly more expensive, too.
A well as couple of megabytes internal very fast SRAM, and direct connectivity to external DRAMS, comparison-wise just like 4 banks of XMP1.3 PC memory, are probably a greater attraction at the moment. I mean I actually wait for people to come forward with examples (I'd know a few good ones myself, as I'm sure Andreas as well) which do not require huge connectivity with memory and between-boards to make use of the MIPS power.

Of course working designs with software are a boon, too, and the low power and small size of the board I hope to receive at some point in the future (but seriously, O.O.M. year later is a problem) make it easily more than a mixer computer with serial connectivity and a display (of which there are more), but the experience with the "new" platform will take time and effort, so the long term continuation of this type of project may well depend on what the practical prospects are!

T.V.

FWIW : I'd grasped Parallella especially on account of the 'parallel programming for the masses' idea. In particular it has, and will further, enable entry for many like myself who could probably 'light up' intellectually & practically at the opportunity entertained by the lower cost. So it is 'higher tech' than I'm used to rather than bleeding edge. It's an interesting combination of ARM, FPGA and Epiphany ( with interfacing betwixt ) and for me at least seems plenty enough 'rocket science'. Naturally I'd love more than 32KB per core, but if wishes were fishes ....

In any case I'm somewhat more focused now upon what fun I might have with the gadget, instead of The Adapteva Odyssey.

Cheers, Mike.

( edit ) I failed to mention the ecosystem that has collected around this project too. That's a neat boon!

My concern is that the effective bandwidth, the subject space where the actual Parallella 16 or more core usability can practically be applied to with good speedup, the amount of interest of another processor learning curve with the current surroundings to the project and the appeal compared to the Raspb. P. for other people than in electronics simply isn't enough.

I mean DMA not working, on/off bandwidth that frankly sucks, the number of FPGA lanes compared to state of the art point to point connection bandwidth makes that a for a noticeable percentage of applications, the board without serious improvements to that hard data, sucks. And I don't like to have to write that. And honestly, the signals I can extract from the current environment which registers interest in the project don't make me believe in a project future along the lines of the multinational colors of beneton or something, but rather that some serious collaborators are replaced by more sucky components.

I still believe my (of course very modest! ) investment will very much pay off, but I'd have liked it if either academically or hard-core high-tech-wise the bar would have been raised a bit. I mean I like to learn about practical bus connectivity on the "bleeding edge" if possible, or interesting Open Source FPGA programs, electronic considerations for board design, and make contacts with interesting people able to produce cheap, small,but interesting add on boards, but thus far I'm not learning much.

T.

The "Adaptiva Odyssey" includes the fact that 2/3 of the original team has departed since the Kickstarter

New funding and, in effect, a very large partner that isn't going to disappear anytime soon, should allow the Epiphany architecture to increase capability and become compelling: in particular, more cores and more memory per core. When a 64 core chip arrives at a reasonable price the fun can really start. Meanwhile the Parallella, after it finally arrives, should prove to be quite good for learning (much) and prototyping.

At an academic level ? As it is, not really.

While the delays have been disapointing, I thought Parallella was an interesting technology mix worth gambling on, and still do. It'll be interesting to see what resources other than funding Ericsson brings to the table. They've apparently been working on a parallel debugger.