Rather than throttling them, if temps do get a bit high I would reduce the number of cores in use for CPU work. That means tasks still get finished at full speed. I don't really do GPU work as my main project doesn't give out any. If at some point I get a second graphics card I might use one for crunching and one for my monitor.

Not an Intel NUC but I've been running these two for a couple of months with no issue. The are small form-factor PCs with AMD chips:
https://www.amazon.com/Beelink-GTR6-Windows-Computers-Desktop/dp/B0BTHP8TNW
https://www.amazon.com/Beelink-Windows-Computer-Display-2-5Gbps/dp/B09SYR619R

I need to keep my windows open because of the heat generated during the day, but otherwise no issue.

FWIW I ran a cost analysis earlier this year and decided Mini PCs were a better bang for the buck as compared to desktops: https://docs.google.com/spreadsheets/d/1oTH_pctZzEnrgjiLow6jFyLJsl9jcBdemTAzX8eoNbg/edit?usp=sharing

Lots of reviews of mini-PC's. Seems to be popular choices for low-cost, low energy consumption for basically full power computing equivalent to the best laptops minus the screen.
https://www.youtube.com/results?search_query=mini+pc

Yes, I do run Boinc on all cores. Usage typically varies between 60% and 100% with a "Waiting for Memory" status on some tasks when the CPU usage is lower. I have 12GB of ram which isn't always 100% used, but perhaps Windows is saving some for itself. I also think the "Waiting for Memory" tag is sometimes associated with LHC@Home jobs, which use a virtual machine.

Checking on the Ryzen 7 mini PC, I see that the motherboard temperature is 87 degrees C after running continuously all day.

Lots of good mini PCs out there as Keith says. One suggestion is to look up the CPU in passmark to compare. https://www.cpubenchmark.net/cpu_list.php

Let us know what you end up choosing!

Hi,

I was wondering if anyone has had any experience running the small-form-factor NUC computers with BOINC projects.

The reason I am considering NUCs is that I'm trying to reduce my energy usage. Therefore, my primary concern is not so much the performance. It is the heat. With my past power-hungry desktop machines with GPUs I was always able to run them at 100%, 24/7 because I installed cooling systems as necessary to keep temps within limits.

With off-the-shelf NUCs I'm not so sure that their stock cooling systems and designs would handle BOINC 100%/24/7. I'd hate to have to throttle them to keeps temps in line.

I'm looking at keeping the cost at below $400 per NUC and I will be needing at least two.

Any guidance? Thank you very much!

For somewhere around £400 / desktop I put together 2x24 thread machines that run flat out on all cores drawing less than 140 watts at the plug (OK, when the GPU is running a task the power draw can go up to 142 watts but I’ve never seen it above that).

I suspect that when you factor in the cooling in the small form factor and the efficiency cores in the intel chips, you would struggle to match that with an NUC.

You should stay away from Intel hybrid cpus for BOINC computing. No scheduler so far can figure out how to keep tasks off the efficiency cores reliably.

Consider AMD Mini PCs for BOINC tasks as they offer a good balance of performance and efficiency. Watch temperatures, and if needed, reduce core usage rather than throttling. Explore cooling options if your NUC gets too hot.

For me, the right answer will depend a lot on the projects you are most interested in. While I crunch other projects during the big gaps in work availability from CPDN that is my main project.They want work units crunched as quickly as possible. (Whatever you might think from the deadlines.) That means for me, running a tower with good cooling. (I use liquid cooling but there are more than adequate air cooling options.) For those tasks, the think that I can do to speed up my current system the most is to get faster RAM and more of it if I want to use all cores at once. With CPDN on my box maximum throughput of tasks is using N-1 real cores where N is the number of real cores present. With MCM1 tasks from WCG I get significant throughput gains using up to 2N-1 cores so all except one virtual core. Most of my crunching will be done from solar once I get a battery for about six months of the year. During the darkest months I will reduce my hours of crunching.