Dear all,

 

As said elsewhere, I'm interested in distributed scientific computing, namely BOINC, and more specifically the Einstein@Home project:

 

https://boinc.berkeley.edu/

https://einsteinathome.org/fr/home

 

ARM platforms seem to be a good choice for that purpose IMHO. Even though they lack the raw power of x86_64, at least they don't use a lot of energy and don't need active cooling most of the time (I can't stand the noise of a fan ).

 

At the moment I'm running BOINC on two ARM nodes: a Raspberry PI 3 (Raspbian) and a NanoPi M1 (Armbian). Both are quite dated now and don't allow me to contribute much to this project.

 

In your opinion, what would be the most powerful SBC running Armbian at the moment ? I've read some nice reviews about Rockchip CPU, are they worth it ? Since GPU jobs are optimized for Nvidia and AMD (OpenCL Full Profile), I'll only be able to run code on the CPU at first.

 

Best regards,

 

Magic Sam

I think the best possible SoC that is also well supported is any device featuring RK3399. Yes there are successors now, the RK35xx series but it will take years go get to the same mature level as the RK3399 is now.

Dear @Werner,

 

Thank you for your quick reply, really appreciated

 

Best regards,

 

Magic Sam

OK, I just went ahead and ordered a NanoPi R4S from Friendly Elec:

 

https://www.friendlyarm.com/index.php?route=product/product&path=69&product_id=284

 

https://www.armbian.com/nanopi-r4s/

 

I'll let you all know as soon as it arrives

 

Best regards,

 

Magic Sam

Great choice. I had one myself but somehow managed to kill it

Even made a custom fan hat for it. If you are interested reading my jabbering about this board take a look here: https://zuckerbude.org/nanopi-r4s-router-overkill/

 

Another board with RK3399 that is fairly cheap are for example the quite recently released OrangePi 4 LTS (where LTS stands for 'we had to change the network chip due to shortage and simply call it LTS for no reason like we always do').

Great review @Werner, thanks for sharing

 

I bought a cheap aluminium heat sink for my NanoPi M1, and it doesn't do a great job IMHO (temperature still at ~70°C, I even downclocked the CPU to 1200 MHz but to little avail). The board is resting on a hard surface, so air can't circulate freely around it, that may explain why.

 

I guess I'll have to invest in better heat sinks like the ones in your article to properly cool down the NanoPi R4S. I also ordered the metal case which Friendly Elec claim is good for thermal dissipation. We'll see whether that's true or not

 

Best regards,

 

Magic Sam

My strongest ARM-based BOINC crunchers are two Hardkernel Odroid-N2+ boards, one running Android 9 PIE, the other Armbian. While the Rockchip RK3399 has two ARM Cortex-A72's and four ARM Cortex-A53's, the Amlogic S922X of the Odroid-N2+ has four ARM Cortex-A73's (@2.4 GHz) and two ARM Cortex-A53's (@2.0 GHz). So more, faster, and better high-performance cores and less, but faster, efficiency cores. Both N2+ boards have a 80mm fan under the big standard heatsink and this helps to maintain a CPU temperature below 40 degrees centigrade. 

Without fan they run quite a bit hotter. Fan screws double as stand-offs.

Edited February 2, 20224 yr by Dirk_P_Broer
formatting

Hi all !

 

Just to let you know I received a text message today about the NanoPi R4S arriving in my country

 

I had to pay my national mail delivery service ~6€ to release the package from customs.

 

Cheers, Magic Sam

Hi all

 

I received the NanoPI-R4S yesterday !

 

It has been running Einstein@Home jobs since the moment it arrived:

 

https://einsteinathome.org/show_host_detail.php?hostid=12919930

 

Thermal dissipation seems to be good: the metal case is burning hot to the touch, yet the CPU temperature is "only" ~60° C, with 4 cores always running at 100% and 2 cores mostly idle.

 

I only wish there were more RAM. Boinc gets oom-killed some times.

 

Cheers, Magic Sam

56 minutes ago, magic_sam said:

I only wish there were more RAM.

The RK3399 can only handle memory up to 4GB.