Is Parellella dead? Is there something equivalent or close to it? Is anyone working on anything of this nature? Or know of any such project or product? I find precious little about it. A parallalla forum has no recent posts. There was a post about one guy starting Parallella then going on to something else.

Details: I discovered the Parellella board five years back or so and was interested. I now have the time to do something. My thoughts are to build a small cluster with, say, four Parallella boards, or equivalent, get a couple of small software projects working, then see if there is a market for that or for larger systems.

I think of this type of project as a Beowulf cluster, but far more efficient than using cheap/old computers in terms of space and power. If the hardware can be had, the hardware part will be simple. The software part, a tad bit more difficult.

I post here because I see Linux as the hands down best option. I think there is existing code in github and maybe other places.
Any thoughts?

From my quick scan of it, The design is over 10 years old. Fastest frequency is probably a few hundred Mhz, & Arm 9 is 32 bit. No cpu features to boast about, it appears to be a GPIO toggler in the main. Processor power is insignificant. The ram doesn't appear to go over 1G. USB-1.1, & 1GB nic seems like the best peripheral. You're stuck network booting it, it seems.

Boards like that have to be contract manufactured, returns have to be handled, etc if they are sold. Contract pricing varies. I imagine it was a good idea at the time, but faded away.There's a significant investment for a manufacturing moq required without the interest from the publicto make investing a wise choice. Also, manufacturers often drop chips and that could lead to significant expense, redesigning & reprogramming.

That's $99. For less than that you can buy a Raspberry Pi 4, with 8G of ram, with an option to boot from USB, and that's a serious player. It even has GPIO. The only outstanding question on the parallella is why someone keeps paying for the site.

1 members found this post helpful.

From a blog, I think by the developer, there was a picture with two stacks of these cards, each one 22 cards high. From what I can tell, the card has a two core cpu that is the interface to the 16 core device. Presumably they use shared memory. That frees the 16 core device from interfacing and I/O chores making them faster for the intended processing. There were follow on chips with 64 and 1024 cores. The goal is not a full fledged computer, just a support package for as many cores of CPU power as can be had in as small a package as possible. I suspect that most process intensive process do not need multiple Gigs of ram. See my thoughts below.

However, and concluding from the blog I read, they did not invest sufficiently in the software and did not garner enough interest for someone to create said software. So it died on the vine.

However, again, Digikey still has it listed. So if someone purchased a modest number of these, and developed some software, it might be possible to start something. Even if a new card must be designed, there must be some software that people can try.

I have two thoughts for that. Write some code that does 128 or 256 bit fixed point arithmetic and a utililty to calculate the Mandlebrot set. I did that on a PC back in the 80's that used 128 bit arithmetic and zoomed in to the resolution limits of those numbers. In this concept, each core would work on one horizontal scan line of the image. Each point is completely independent of its neighbors. Very CPU intensive but way less than 1 K of ram for the data, and probably about 1 K for the code.

The other thought is to create a solar system model. Each core is assigned one quadrant of space and calculates how each celestial body within its quadrant will move during each time period. This task is more difficult because each core would have to query its neighbors as to what is nearby. Celestial bodies will move amongst the cores That could be pretty cool for a demonstration. Again, not much memory and very CPU intensive.

Finally, the most important part of my question is: Does anyone know of any additional boards of this type that are available and in production? By that I mean a multi core CPU on a small board with minimal hardware. Nothing but a support package for a bunch of CPU power. My searching ability leaves much to be desired and I have not been able to find anything.

All said, Thank you for taking the time to read, think, and reply.

For those sort of features (No frills, cheap money), I'd be looking at servers

Depending on your price range, You should have a look at these two
http://www.solid-run.com
http://avantek.co.uk

Both are at the cheap end of arm servers. The solid-run server is particularly cheap. At least they boot.

Looked at solid-run and avantek. Avantek does not look close. solid_run does, but: Max core device is four and they all have multimedia stuff, wifi, bluetooth, and stuff not needed.
I just want the minimum to support multiple cores and a minimal interface, Ethernet. The parallella board was perfect. I wonder why essentially no one was interested.

Well, the one that tickled my fancy was the mini itx server for $750. 16 cores LX2160A there. I think few need the 64G they're offering. 16GB max, I'd feel.

https://www.solid-run.com/arm-server...b-workstation/

There's transport from Israel(?) also on top, I presume.

As for why nobody was interested: Back in History cpus were 8 bit.The biggest package was 40 pins. Motorola broke the mould with a 68 pin for the 68000. Cpus like the z80 could have 1,2,or even 3 bytes per instruction (not including operands) which slowed things down. The idea behind RISC was to have one byte per instruction, and limit the number of instructions. It wasn't a great improvement, because a MOV instruction needed 2 operands, each 2 bytes long if you had more than 256 bytes of ram.

Once you went to 16 bits, there were 65536 instructions to play with and RISC was passé. So nobody put energy into optimizing an OSS project. So risc will remain an interesting toy for academics and little more.

Yeah, compare that with the $99 price of the Parallella that has 16 working cores and a separate 2 core processor to handle all the interfacing. The more I think about it the more I think the second processor to manage the other chip was a great idea. Frees up the workers from the overhead. That is on top of the basic concept, 16 cores with Ethernet I/O and nothing else for $99.

They had a really good idea. But, Digikey has something like 300 in stock. Maybe as few as four boards could provide an example system. If someone could get some good software working, that might stir up enough interest for revive the project or start a new one.

I think the reality is, someone won't get some good software working, unless you do it. We haven't even discussed drivers. Risc has retreated into the theoretical, because it no longer belongs in the practical. Bearing in mind that your overclockable Arm in the RasPi is about 5x faster than your 16 cores, they just might come out even. The Rpi chip overclocks to 2.2Ghz without adjustment, I hear.

But it's your time, you money, your choice, and you do what you want to. I just had a quick look in the kernel source under the 'arch' directory. x86 = 117M; Arm = 36M, with 9.4M in arm64; riscv = 816k. You might hit imposing problems.

You might find stuff among academics or students who have done projects on riscv implementations who will be willing to share what they have.

Unfortunately, no-one else has jumped in with ideas or thoughts.

Re: Bearing in mind that your overclockable Arm in the RasPi is about 5x faster than your 16 cores, they just might come out even.

I am pretty sure not. The Parallella can run at up to about 25 GFLOPS and 19.2 GIPS (Gig instructions per second). I don't know how much of that 25 and 19.2 is for the Epiphany processor and how much for the CPU used to provide the interface with the host computer. There are 64 core and 1024 core boards.