Its interesting. I have one RPi3B+ with 1Gigs of RAM and it crunched a lot of WUs but meanwhile R@H increased minimum RAM requirements from 850MB to ~1700MB and now it hasn't got WU-s.

My RPI 3B+ is still happily crunching.
I often get the message that 1,7GB is required, but in the end I always get WUs. But it already happened that new WUs did not come for 24h, so I have a 1 day buffer.

I switched it on back, we will see.

I have also an OdroidC2 with 2Gis of RAM and it crunching fine with 2 WUs in parallel around 50°C.

Is this due to RAM limitations?

Yeah, sometimes one of WU fails because initiating OOM killer :)

BTW my RPI4B+ (4Gig of RAM) crunching only one WU at a time because of stock heatshink and it runs around 55-60°C.

Could it make more sense to downclock it and run more WUs in parallel? CPU cores usually become more efficient at lower frequencies.

Hmmm, I don't known this. I don't know if I can downclock my RPI4b CPU because I use mainline kernel from Manjaro.

If I'm right there are several types of WUs. Some of them needs less memory (500-600M) and others needs 1.5Gigs. If rosetta sends only WUs that fits our machines then the whole chrunching would be more optimistic. It knows the settings and allowed MEM usage. If it knows statistics about WUs runtime parameters tt is a simple computation...

there is something about the android phones running on arm cores
the high end ones often have pretty decent ram and some have as many as 8 super scalar cores
specs for an 'old' Samsung Galaxy S7
https://www.gsmarena.com/samsung_galaxy_s7-7821.php
the thing though is that we'd need to run a 'flavor' of android os that doesn't interfere and throttle the tasks.
it makes a lot of sense to throttle that on a regular mobile phone and in regular use.
but running boinc is anything but 'regular' i.e. the phone would need to run off a power supply
and it may even need a heat sink + fan to dissipate all that heat

for now 4GB Pi4 is probably an ideal platform to realize that kind of performance
the a72 superscalar arm cores really made a difference
but they run hot, so a sink + fan is needed
more of that over in the Pi4 thread
https://boinc.bakerlab.org/rosetta/forum_thread.php?id=13732
down clock would sacrifice performance and may result in longer time per model and less models returned and in turn less credits per task. overclock is the most interesting thing happening now but u'd need a good sink fan combo to do all those, i think overclock provides the returns in credits per watt . hour. at the expense of more sink-fan hardware.

Due to non-linear frequency dependence on voltage, but linear relation with performance, it could very well happen that it performs better performance/watt.

Is it possible to run Rosetta (or Boinc projects in general) on Windows 10 on ARM?

---

Is it possible to run Rosetta (or Boinc projects in general) on Windows 10 on ARM?

I don't think Windows 10 on Arm is a supported platform (or ever will be). The hardware itself is compatible with Rosetta@home, but only if you somehow run it on Linux in a virtual machine.

I don't think Windows 10 on Arm is a supported platform (or ever will be). The hardware itself is compatible with Rosetta@home, but only if you somehow run it on Linux in a virtual machine.

I dont own one, I was just curious whether it could run Boinc projects like it does some other x86 software.

---

I don't think Windows 10 on Arm is a supported platform (or ever will be). The hardware itself is compatible with Rosetta@home, but only if you somehow run it on Linux in a virtual machine.

I dont own one, I was just curious whether it could run Boinc projects like it does some other x86 software.

It might be able to emulate the 32-bit Rosetta for Windows, but the performance would be complete trash and not worth the effort.

It might be able to emulate the 32-bit Rosetta for Windows, but the performance would be complete trash and not worth the effort.

Some synthetic tests running in the emulation make it look on par with older ULV chip like the Core M line (which may be trash, but there are people out there running Rosetta on this trash).

---

I've got a Win10 laptop, switched from LHC for now, running 8hr tasks but just for curiosity I've sparked up an old Tesco Hudl that was lying in a drawer. (It's not clever enough to run WCG tasks) It's only got 1Gb of RAM so sometimes refuses to get work with the "...needs 1716.61MB but only 716.81MB available..." message but all the tasks it has returned so far have been valid. I'm letting it use all of 1 core and as much of the available ram as it needs on 2hr tasks. I think letting it use more cores would be pushing it and longer tasks might need more memory, although I might experiment on both counts.
"Every little helps". It might just be one of these short tasks that finds the key. Who knows.
Stay safe. Stay well.

I think letting it use more cores would be pushing it and longer tasks might need more memory, although I might experiment on both counts.

The time a Task runs for doesn't affect memory usage (that much- during a run it will use more or less memory at various points).
Some Tasks only need 250MB of RAM, others will need 3GB of RAM- regardless of how long the Task is set to run for.

---

Grant
Darwin NT

Thanks, Grant,
It didn't seem too happy running 2 cores using 250-400MB each so I've put it back to 1 but upped the time to 6 hours. If it goes well overnight, I'll deselect time limit and just let it run whatever it gets. The scheduler appears to know that it doesn't have much memory and has only sent WUs it is capable of running . The laptop got a task that needed 1.3GB yesterday but it has the capacity, and spare, to handle that. I'm letting it run WCG and Ibercivis as well and the other one is doing Folding, just to cover all bases. I've even got the phone doing Dreamlab.

I'll deselect time limit and just let it run whatever it gets.

The project's default Target CPU Runtime is 8 hours. Some tasks will finish before then, some may take longer (up to 10 hours longer at which time the Watchdog timer will end the Task). But generally the vast majority of them will run for the Target 8 hours, give or a take a few minutes.

---

Grant
Darwin NT

Maybe someone already posted on this, but here it goes:
I have 4 nvidia jetson tx2 boards running boinc right now.
I noticed that tx2 is pre-configured to run with 4 CPU cores: 4 cortex a57 after OS flash
However, jetson tx2 has 6 cores, and after "unlocking" the extra 2 Denver2 cores, I found that they are not being allocated towards rosetta's WUs.
So, my question is: how do I inform to my Ubuntu's boinc that I have two extra CPU cores?
I already check the .xml files, but this is not trivial for me, so any help would be much appreciated.
Thanks.

As a heads up to the (more adventurous) Raspberry Pi folks on this thread, there is a 64-bit RPiOS beta. It came out in late May and has been updated numerous times since the initial release:
https://www.raspberrypi.org/forums/viewtopic.php?p=1668160

I have the 64-bit RPiOS running on three Pi boards. All of them are v4 boards (Model B), two with 4GB of RAM and one with 8GB of RAM. I've had this setup running since early Sept and R@H has been working fine that whole time, no issues at all. Not sure what the estimate is on when the GA version of the 64-bit OS will be ready.

Turing Pi have announced they are working on the Turing Pi2. The original is an mITX board that holds 7 Rpi CM, CM3 or CM3+ modules.

The Turing Pi2 is still going to be mITX and will hold 4 CM4 modules. It also provides a layer 2 managed switch, 2 x SATA III and 2 x mini PCI express on the board. See their web site for details. They are targeted as Kubernetes clusters but there should be no reason why you couldn't run them as a BOINC cluster.

---

MarksRpiCluster