ARM CPUs (not yet)

[PantherX]

Thanks for all. I mean ARM64 beta is available: https://foldingathome.org/beta/ Not others announcements elsewhere
I have two SBCs, OdroidC2 and a Pine Rockpro64, I set client-type=beta on both and now I get a8 core and 16810 project WUs...

Welcome to the F@H Forum oreggin,

Please note that currently, there's no public or closed Beta for ARM client. The version you have linked to is still a test build and needs further issues to be resolved before any kind of beta is announced.

Since you're using the client-type=beta flag without being on the Beta team, it is violating our forum's T&C's (viewtopic.php?f=16&t=8) so unfortunately, we can't help you out

I would also like to point out that experimenting with ARM client might end up hurting the science instead of helping us (viewtopic.php?f=24&t=35998). We completely understand your passion to contribute toward the fight against the pandemic and rest assured, those that are working on it are doing their best but alas, they too are humans who have a life and need a break, juggling life, etc. Thus, just be mindful of the bigger picture and make an informed decision

[MeeLee]

It'd be very cheap to enable ARM testing.
For instance, you can find a $35 TV box running android 6 on Ebay, with an AMLOGIC S905X3 CPU.
It's a fairly modern quadcore running at between 1,7-1,9Ghz (depending on where it has been made).
You can even install a fully fledged Linux for ARM on there!

The next interesting ARM product would be the octacore ones.
They use only the 4 low performance cores in Android, but perhaps Linux has access to all 8 cores for folding?
They're like $80+.
The newest chipsets will feature dual core CPUs running at ~2,2Ghz and 6 low performance cores running at 1,7-2,1Ghz (depending on brand and model).
Those are also interesting because they have more cores to crunch data. 6 are available in Android, and perhaps more in a full Linux OS.
Not sure how the faster cores will work with the slower cores though..
These type of products are available on modern phones with Snapdragon 700/800 series, but will be available soon for the rest.

[bruce]

Not sure how the faster cores will work with the slower cores though...

If the ARM code works the same way the x86 code works GROMACS runs at the speed of the lowest core. That means you'll want to set affinity for just the fast cores and configure it as a fast slot and collect the affinity for the slow cores to create a second slot ... or something like that. ... or if heat is a problem, forgo the slow slot.

This has been demonstrated on CPUs with/without hyperthreading where affinity can be configured.

[bumbel123]

Tested that ARM FAHClient on one of my Raspberry 3 (64bit), it does work/run ... the heatsink I put on all of the RPi does get hot, damn hot, I tested with an opened case.

The runtime of a WU was roughly 5 days and the resulting output is ~500 PPD ... so, really nothing for short term folding ... ^^

[Joe_H]

Tested that ARM FAHClient on one of my Raspberry 3 (64bit), it does work/run ... the heatsink I put on all of the RPi does get hot, damn hot, I tested with an opened case

Please turn off the ARM client. There are still problems associated with the ARM client and folding. When they are ready, then there will be an official announcement from the Folding@home Consortium for a beta test. Until then they are still working through some problems, people jumping the gun are complicating sorting that out.

P.S. The minimum hardware will probably end up being a Rasberry Pi 4 or better.

[bumbel123]

Please turn off the ARM client. There are still problems associated with the ARM client and folding. When they are ready, then there will be an official announcement from the Folding@home Consortium for a beta test. Until then they are still working through some problems, people jumping the gun are complicating sorting that out.

P.S. The minimum hardware will probably end up being a Rasberry Pi 4 or better.

The FAHClient Beta for ARM is public available, I had a RPi3 here to give it a first try ... so what's your problem?

I'm aware that this is beta testing and I wrote clearly readable for everybody "Tested that ARM FAHClient ...". Where do you read it's still running?

I don't see any high power computing output coming up here even with some code fine tuning, also not on a RPi4. Maybe it can break the 1000 PPD wall ... but it'll never become a sprint capable folding hardware. But hey, some things don't needed to be rushed ...

For test purpose I had e.g. a Intel NUC8i3 running, Win10, just on one thread out of 4 for a couple of month, it has done 4000-6000 PPD ...

[Neil-B]

I think Joe_H has already pointed out what the problem is in that none invited testers utilising the ARM client at this time are complicating things and whilst you didn't write it was still running, neither did you explicitly write that it was no longer running

I really wish that beta page wasn't linked from the web site - or even published tbh .. As I understand it Clients are put there initially to allow invited testers to easily access ... At some point there is an announcement that opens the Public Beta phase (if required), but until then Clients published there aren't meant for general beta testing ... I know, anything published on the web not behind access controls is inherently public but ... Perhaps in future someone might look at possibly putting pre Public Beta clients behind access controls or utilise another approach (maybe Dropbox or some such) to avoid situations like this.

[bumbel123]

... whilst you didn't write it was still running, neither did you explicitly write that it was no longer running

Hmm, I'm not nativ english, but "Tested" is past ... ?

Perhaps in future someone might look at possibly putting pre Public Beta clients behind access controls or utilise another approach (maybe Dropbox or some such) to avoid situations like this.

What situation?

Did I kill somebody? Did I burn that forum here? Did I destroy F@H? Did I push the moon out of its orbit? And the most important thing, did I ask in any way or kind for support or help with that FAHClient Bete for ARM?

[JimboPalmer]

https://www.dictionary.com/browse/burn- ... behind-you

https://www.urbandictionary.com/define. ... %20to%20it.

I am thinking you need a meme about burning bridges before you even arrive.

In other news, it may be obvious why the Beta team is a closed group, why non disclosure agreements exist, and why some people never leave their parent's basement; but I may be wrong.

“There are three kinds of men: the one who learns by reading, the few who learn by observation, and the rest of them who have to pee on an electric fence for themselves.” (Will Rogers)

[Neil-B]

To answer your questions:

Tested is past but doesn't mean you have stopped testing ... Situations where people inadvertently use a beta client that isn't intended for public use yet ... No ... No ... No ... No ... (and whilst it is not the most important thing) No.

Tbh the most important thing is that by using the client you may have inadvertently have made beta testing of the client harder than it might otherwise be.

It isn't the end of the world - but it simply makes the devs lives harder (actually having seen latest from Joe_H - it isn't just making lives harder it is actually interfering with new core development - the value of that new core is very significant in terms of throughput and performance across much of the existing CPU folding resource and so this is a big deal imho) ... Hence why I wish the beta clients weren't posted in a publicly accessible location until formal Public Beta calls are announced.

[Joe_H]

They had to take a project offline temporarily due to these unasked for tests of the ARM client. So yes, in total the number of people jumping the gun have caused problems, and the project is being delayed.

Look up Project 16810, that is COVID-19 related and is currently also a Beta test of the new A8 folding core for CPUs. It has been suspended since Tuesday or Wednesday. This is a direct result of people running the ARM client. Before this it was found that the limited internal testing on ARM resulted in a very high error rate, and those tests were suspended until fixes could be put in place and tested.

Your argument is the page is "Public". So it can be reached, but as pointed out there are no links on the various F@h site pages pointing to that page. There has been no official announcement asking you or others to download and test any of the Beta versions of the client posted there. When a version is ready for public beta testing, there will be an announcement.

So I repeat to you and anyone else running the ARM client without being asked to, Turn Off your ARM client if you haven't already done so. Leave it Off until there is an official announcement. Until then it is not welcome, and is contributing to problems.

[bumbel123]

Well, if you fear that you shouldn't bring that client easy available public for download. It was even in the press, that F@H is going to support ARM hardware ...

I just let one WU finish to fold on the RPi3, as I don't see much sense in folding 500 PPD. Maybe there are ARM CPUs out there, able to do serious folding speed. But it's for sure not those little DIY PCBs, called whatever PI.

So, come down, take my note as serious feedback, as this was my intention. I hadn't issues btw. to make and let it run, it did look pretty stable, what I've seen.

[Joe_H]

Which press? And something mentioned in a press article other than an official announcement being released does not constitute an invitation to start using the software. "Going to support ARM hardware ..." does not equal "A client is out and ready for everyon e to use right now."

As for not having any issues, the first persons running the first internal tests on ARM thought everything was fine, that is until someone else noticed that a very high percentage of WUs returned by them were marked as "Faulty" in the database. Some fixes were needed in the client software, and possibly other related software, I do not know if all fixes have been made yet.

[MeeLee]

The RPI ARM processor's heat issue is because it's burned at 40nm, which is quite large.
Even the 3B+ needs active cooling on all core load.
Without it gets thermally throttled.
In my opinion, a Pi isn't competitive with x86 architecture because of that.

AMLOGIC S905x3 is 12nm. They could almost run without heat sink.
Their newer versions are 10nm, and are very competitive with even Ryzen 9 3000 series CPUs (that is, as long as you put enough units in parallel to compete with performance).
For a 3900x, to get the same performance, you'll need between 15 to 20 6 core ARM processors though.
At <5W per unit under full load, that's 75-100W, vs 200-250W for AMD Ryzen CPUs.


Most of the time, the slower cores on a big-little setup (especially if they're 6 of them) run parallel threads more economically than faster octa cores using only the little cores.
However, I think it should be possible to set a 2 core slot, for the fast cores, and a 6 core slot for the slower ones, just as long as the Linux driver isn't dynamically adjusting core load.

[oreggin]

Since you're using the client-type=beta flag without being on the Beta team, it is violating our forum's T&C's (viewtopic.php?f=16&t=8) so unfortunately, we can't help you out

I would also like to point out that experimenting with ARM client might end up hurting the science instead of helping us (viewtopic.php?f=24&t=35998).

Thanks for your info, I removed beta client-type and waiting for released cores and WUs. After third read of rules it seems to me it is too strict to being beta tester