Folding Forum

manalog wrote:Yes, started folding yesterday and both the device were able to stay inside the "timeout" mark. The Motorola already finished a WU.

I don't know if soon or later they will get wu that stay inside "expire" but not "timeout". This, curiously, happened to me not on arm devices rather on a powerful Ryzen 5 laptop. In this case it is better to continue crunching the WU or abort it?

By the way starting from yesterday I've seen 3WUs in my "ARM Cluster", 1 completed, 2 in progress, 100% completed or predicted to be completed widely under the "Timeout" deadline.

I am impressed and tbh surprised ... I am aware of laptops that are struggling to meet the expirations !!

There are projects that are small enough in WU size and processing time to be assigned to ARM clients, they try to keep the larger ones off being assigned to ARM. However for the Intel/AMD chips they mostly can only assign based on the number of CPU threads in the request. So a laptop or desktop may get a WU that may not get completed within the timeout.

Every so often I get one of those small WUs on my laptop. Take about 3 hours of a 2 day timeout period.

A couple of things I've noticed in my second day of running FAH on two arm devices:

1) CPU usage it's stuck at 72% for both; I thought it was only the Motorola due to Android limitations, and I was a bit surprised when I saw the same on the monitor of the Orange Pi 4, running Armbian. While the Motorola phone never reached 100% even with Boinc, the Orange Pi 4 ran Boinc at 100% usage for an entire year.
2) On Android devices, be careful with the TPF: it changes A LOT during the execution of the workunit.

By the way, no error to report nor wus not being completed in time. The two devices are still crunching without any problem; the phone is not terribly hot, I still haven't put a fan on it.

From my experience with Boinc (very similar to FAH), is that most modern Android phones having an 8core ARM CPU, they can only run up to 4 cores on the A50-series (low performance) cores.
Setting affinity to 5 or above usually results in overheating.
Phones with 8nm or smaller lithography can run 4-6 cores easily, without exceeding 35C (5C below the max of most batteries; the °C values can be read from CPUZ).
The Qualcomm Snapdragon 730 is one of the few mid-range CPUs that has 2 fast cores and 6 slow cores; and can run 6 cores without overheating.
More common 10nm models usually can run fine at 4 cores, but if your model has 14nm, you'd have to reduce the amount to 2 or 3 cores (not to exceed 40C) in a phone sized, and enclosed device.

The Cortex-A70 cores (found in the Pi4) are much faster, the A50 series are much slower; the Neonverse (not many have this) are even faster than the A70 cores.

Setting your model to beyond the slow cores usually leads to overheating, as one or more of the cores need to constantly read more L-cache data from RAM; and/or if the used L-cache per core is small enough to load 2 WUs per core, it'll run very optimally, using ~ every clock cycle, and therefore overheating much easier.

One thing you can do with old phones or tablets, is:
1- Open the device, and improve the CPU cooling, by manually installing a heat sink on the CPU at least.
2- Separating the battery from the CPU
3- Run from a sufficiently powerful power adapter (usually 5V 2A is more than enough for most phones).

Existential question are starting to arise

Orange PI,
PRCG 18436
ETA 2.00 days
TPF 31min 46 sec
Assigned 2021-11-28T14:37:31Z
Timeout 2021-11-30T14:37:31Z
Expiration 2021-12-02T02:37:31Z
Now: 2021-11-28T20:02:36

So it seems it will miss the timeout for just 6 hours, while it will stay safely inside the expiration. I am not interested in point, only science. Should I keep folding it or abort it? I read some forum post but still I don't know what is the consensus about Timeline vs Expiration. Will my WU be assigned immediately and so be a waste of energy and computational power or, considered that the expiration is after 2 more days, this chance tend to zero?

A WU is reassigned to another machine after the timeout delay ... you can still complete it, but results will be duplicated.

When expiration delay is reached, the client deletes the WU from your system.

The slowest CPU that can complete WUs on time is the RPi 4 (although p16969 won't complete before timeout, and I think it's the only exception).