I Just Wanna BOINC! -- Part 3: Alternative Single Board Computers

in #gridcoin7 years ago (edited)

part3-odroidxu4.jpg

Do you suffer from low magnitude disorder? Are you tired of only earning 6 Gridcoins per day? I too was once a victim of this hideous malady. But I finally said “ENOUGH IS ENOUGH!” and chose to unleash the hidden power contained within my hard earned stacks of fiat, perhaps altering the course of my entire life.

Welcome to part three of the multi-part series, I Just Wanna BOINC!, where you can follow along with my transformation from a low-mag nobody to undisputed Gridcoin royalty! (or at least to a slightly higher-mag nobody)

Part 1: The Pi Stack

Part 2: The Z620 Workstation, A Noob’s Perspective

Part 3: Alternative Single Board Computers

When I think of single board computers (SBCs) the Raspberry Pi is generally what comes to mind, but there are actually many other options. In this installment I will be sampling 14 of these alternative SBCs and determining which represents the best value in regards to Gridcoin profitability. With this knowledge in my arsenal, I can abandon the scattershot approach and instead base my future investments on these findings.

Here are the 14 boards I chose (basically the boards with 4+ cores that seemed to be reasonably priced):

part3-sbc list.png

And now, to the data! All of my current machines are included in this table, not just the new SBCs. If Steemit is squeezing this image and making it look blurry, right-click on it and select “open image in new tab”.

part3-datatable.png

The clear winner here is the OdroidXU4. To my surprise, it’s even more profitable than my second-hand Z620 workstation (which I got for a really good price, imo). I pay about $0.075 per kWh, but in areas with higher energy costs the low power consumption of SBCs would widen the lead over traditional PCs and Workstations even further.

I believe Dollars per MAG is a poor metric as it doesn’t consider energy consumption. I only included it here because I do see it being discussed quite often. I focus on Payback Time because it includes the operating costs.

Your profitability may vary depending on project selection, but this data remains useful because the relative profitability between the different SBCs will remain static (barring a particular SBC that is very overclockable, or you finding a sweet deal on some used machines).

The four boards with “no mag” are still not crunching work units. Three of them continue to return nothing but computation errors (“output file absent”) and one requires a serial port connection for the initial set up, which I haven’t quite figured out yet. Once these are fully functioning and attain their maximum RAC, I’ll post an updated data table showing how their value compares to the others. The Pine64 initially gave me a lot of trouble as well, but the computation errors were finally solved by using the Armbian OS (a big thanks to Moises for the advice).

I’d love to hear your opinions on BOINCing and earning Gridcoins with SBCs. I’m having trouble finding downsides to this approach. Even project selection isn’t as bleak as I once thought. These can run many more projects than just Universe@home and Yoyo@home while still avoiding direct GPU competition. Many CPU-only projects that have Linux on ARM (or Android!) applications can be crunched with these boards. For example, one morning I discovered that Universe@home had (temporarily) ran out of new work, so I switched every machine to TN-Grid, with great results and zero computation errors.

One of the best features of Gridcoin is that an equal amount of coins are allocated to each unique BOINC project, and each BOINC project has its own unique list of supported hardware. This means that we won’t be overrun by ASICs, or even GPUs, or even Xeon CPUs. My findings show that even low-power ARM chips that were initially developed for mobile phones are relevant to BOINC and Gridcoin and may indeed be the most profitable choice.

I’ll leave you with a low-quality photo of my ever-growing SBC collection. Perhaps one downside to this approach is the clutter.

part3-sbccollection.jpg

That's all for now, thanks for tuning in! You can follow along with my Gridcoin statistics here.

Sort:  

I'm poor as hell but I really want to start with gridcoin so this article is really helpful.

Upvoted and bookmarked.

Nice article @jimbo88 I have a question about performance Odroid XU4 can you tell me how average mag and GRC you getting from these? Probably answer is on screen: part3-datatable.png but I dont see it even in different web browser so I will be very grateful for answer :)

Yah, I have no clue why the datatable.png isn't showing up any more. I've also seem to have lost the ability to edit this post. The mag at the time of this post was about 25 per Odroid, now it's about 12.

Here's an updated table. Disregard the Z620 and the laptops and the Nano Pi Neo Air. They were shut off for the last 5 weeks because I went on a camping trip, but I did leave the rest powered on (right click, then open in new tab to see this clearly):

Untitled1.png

The payback periods are being calculated on a mag unit of 0.417. I'm not sure if this is correct, but to create 48k POR coins per day with only 115k total magnitude, I'm assuming this will be the long-term average. Though since March, this has NOT been the case, which I still don't understand. It seems to be between 0.20 and 0.25. So, to summarize, take my payback period numbers with a grain of salt. They could be twice as long as shown, or even worse, these could be completely unprofitable at current GRC prices and current mag units.

Awesome! Thanks I just wonder why MAG dropped so much. More people in project?

It's not just the project, but GRC in general. There's a lot more people using it now and there's only so much magnitude to spread around. :-)

If you get a chance to try it out I'd recommend the Udoo X86 from https://www.udoo.org/udoo-x86/ it runs windows/linux/android and has built in arduino as well. Supports USB C and upto 3 X 4k displays, it's the most powerful sbc out there...

Yah these sound pretty cool. The x86 is a big selling point (from a BOINC perspective). The price (from $125-$259) is a little high, imo.

I have only just briefly looked into Gridcoin, but I do have the X86 Ultra edition. I'll try and get more familiar with setting up BIONIC/Gridcoin on it and if I can get it up and running I'll post some info/benchmarks on it. Would there be one client that works better than the other...? i.e. window or android or linux, are there benefits using one over another...?

I doubt the OS choice would affect the benchmarks very much. If you install BOINC and add a project, it will automatically run the BOINC benchmarks. You can see the benchmarks by using the advanced view (view -- advanced view), then the figures will be listed in the event log (tools -- event log). The benchmark data will also be listed under your host on the BOINC project's website. For example, here's what the host data looks like for my Odroid XU4. It would be cool to see what the Udoo shows under floating point speed and integer speed.

Excellent work! I have a lowly Pi2 and Pi3, so I grovel at your feet oh lord of SBC BOINCing

Thanks! I've been seeing a lot of people hate on mobile chips, but they're pretty damn profitable. I don't get it. The power consumption is incredibly tiny and the chips are impressive considering you're spending less than $50 in most cases.

The battery in two of mines (Elephone P9000 and Sony Z3) started to bulge after a couple of months. Of course their batteries are welded on, glued in place and guarded by the military so there's no way to use them without.

Try turning the temperature down on the BOINC UI, its by default at 40C but I now set mine at 30C, charging Li-Ion while its warm can damage the battery

Oh didn't think of that. Too late now but I'll do it on the others. Thanks.

Glad I could help the mighty @ravonn, the slayer of superblocks ;)

Indeed, they are very impressive

Thanks for doing this research and providing a great summary of your results! Which of these is the easiest and least technical to get BOINCing? Can someone who just knows Windows PCs set this up without too much trouble or is this approach best left to people with a bit more technical savvy?

I need a new computer soon and may stick with a laptop and then try to up my BOINC/Gridcoin game with these!

I dove in head-first with zero knowledge. You basically download an operating system and flash it onto an SD card. Then you put the sd card into the board, plug in your monitor/keyboard/mouse and give it some juice.

With just a few Linux commands, you should be good to go.

Here's what I did for each of these:

Update the OS first: sudo apt-get update
Install BOINC: sudo apt-get install boinc-client boinc-manager
If you want to view/control remotely: sudo apt-get install x11vnc (or pick the vnc of your choice, or you can just ssh, if you're comfortable with command line boinc).

For BOINC crunching purposes, I would stay away from the boards that I listed as having "no mag". These have been a pain.

Dollars per mag will rule once again when Universe@home finally releases its OpenCL FP64 application! I am waiting and preparing... preparing and waiting... to unleash my 7970s there.

Of course, low-power & many devices approach will always work with CPU-only projects and I am sure you will do just fine when you switch to yoyo or TN Grid completely. Not to mention that new BOINC projects are appearing all the time :)

Yes, project selection may be an issue at some point and if the GPUs come flooding into Universe@Home I will be running away as fast as possible :-)

My point about dollars per mag was this: when computing the profitability of something, you have to include the operating costs (energy consumption in our case). Dollars per mag misses a big part of the picture.

Yes, I admit I've been promoting dollars-per-mag mantra recently to create some extra attention (GPU mining always being the hottest topic in crypto, pun intended) and to point out that older GPUs are not necessarily weaker when running BOINC. But there are many scientific algorithms which will never be ported to GPU and this low power approach of yours is the future really, although it may not win many upvotes in "show your powerful rigs" threads :)

Quite a geeky stuff for me although if I had a know-how I could also build that for a little fun with rewards.

It's a lot of fun to hop in and learn as you go. You could start with a Raspberry Pi. They're probably the most user-friendly and easy to set up.

OMG that's really a lot!!! You have beaten me XD

Excellent post! I've already got some experience with the Odroid XU4, and they're great boards that have gone completely under most people's radar! They even recently released a fan less version that looks very promising in regards to stock speeds.

Pretty interesting that you were able to switch to TNgrid without errors since they don't officially support Linux on ARM (AFAIK).

Yah the Odroid is very impressive. Here's the TN-Grid applications list, which shows the Linux/ARM support. I've been confused in the past because official BOINC documents like this are very out of date.

Only negative thing i have about it, is that the build quality can be a little dodgy. I've been through a lot of them at work, mostly for drone beta/prototype applications, and some of the boards were pretty badly made (bad soldering loose connections on the I/O stuff like that). Most are just fine though.

I have found that the XU4 throttles down to 1.9ghz (of a max of 2.0ghz) almost all the time. Even with the cooler on the board, plus another fan, plus 22C ambient room temperatures.

Did some research on the throttling down to 1.9GHz, and it seems the stock cooler simply isn't good enough to handle all cores on full load. I'm experimenting with some water-cooling at the moment, I'll post my results when I'm done testing. Might even try to OC the bastards if temps look good :P

Cool, I look forward to seeing what you discover!

I haven't noticed that, but we also don't monitor the clock speed that closely so I could easily have overlooked it. It's not too bad, but a little annoying if it's consistent though.

I used to BOINC. I enjoyed it. I wish I had gotten some bitcoin for it. This was back like 4 years ago, so you can imagine the appreciation. Then I found out that the earth is flat and the universe is small. Stars are only a few thousand miles away and may only be optical lens error. Sorry to disappoint.

Thanks alot for this article, it actually saved me some time :) i was just about to do almost the same thing when i found the artice about the pi stack :)
I got myself a NanoPi m3, made manual cooling solution and this thing works like a charm( well the Wifi is sadly extremely unstable).
Iam currently running on all cores but with 80% cpu time ( overheating is still
an issue) but iam looking forward to see how this small beast builds up mag.
Especially in areas like Germany, where the electricity is really expensive this can be a way of at least trying to get halfway even :)

This is a fantastic article! I am very grateful to you for doing all the research in advance. I am quite interested to know if you have any updated figures since the posting of this article two months ago. I am trying to figure out what the best crunch for my watt would be (not necessarily which one would make more GRC, although I guess that might be one and the same thing) Thoughts from anyone would be greatly appreciated!

Cheers!

Hi! Sorry I missed your post. I've been camping for a few weeks, but I've now returned to civilization. The data table has disappeared from this post for some reason. Here's a new one. Disregard the Z620 and the laptops and the Nano Pi Neo Air. They were shut off for the last 5 weeks (right click, then open in new tab to see this clearly):

Untitled1.png

Magnitude on the Odroids has dropped from ~25 to ~12 since this was posted a few months ago.

The payback periods are being calculated on a mag unit of 0.417. I'm not sure if this is correct, but to create 48k POR coins per day with only 115k total magnitude, I'm assuming this will be the long-term average. Though since March, this has NOT been the case, which I still don't understand. It seems to be between 0.20 and 0.25. So, to summarize, take my payback period numbers with a grain of salt. They could be twice as long as shown, or even worse, these could be completely unprofitable at current GRC prices and current mag units.

Great article. Seems like this image is missing from the main text at the moment?

Though since March, this has NOT been the case, which I still don't understand.

I've started just on the last day of August, but since then, this September, payouts are much less than 48k. No idea how it was before. Probably this explains differences? Have a look at my post 'RAC for Dummies'.

Wow your research is super appreciated. I have a couple old computers CPU mining GRC and I was thinking about buying an old server (your first project) but after reading this artical I am now like to see if i can build a CPU farm using these small one board computers. Is there anyway you can attach your data ie excel sheet as the photo is hard to read and hard to print as I want to go and do some digging in the local Bangkok markets for parts.

Hey there, sorry I missed your comment. I blame the (lack of) Steemit notifications. Maybe there's an option I can toggle somewhere. ..

Anyway, I just made a shared google sheet so you can dive in and see all the formulas and the cost breakdown for each machine: https://drive.google.com/open?id=1FI9cVr-djpslVIj0fFKHQDfdYmEfaBssvr6hviFqTWo