Posts by jives11
|
1)
(Message 8561)
Posted 2 Sep 2024 by jives11 Post: Thank you |
|
2)
(Message 8559)
Posted 2 Sep 2024 by jives11 Post: Thank you for your reply, which clarifies a number of things. The heat sink I have is a combined heatsink and case and has quite a substantial aluminium vanes across the top and bottom on 80% of the card. It doesn't have fans as I find the noise intrusive and try and achieve passive cooling, so a big heatsink and sat on a large flat piece of granite .I monitor the temperature using vcgencmd measure_temp The thing is most of the time the RPi5 can run 4 cores of asteroids 24x7 and not exceed 70 degrees. I've left it like that for days and it's quite comfortable. But sometimes it seems to go 10 degrees higher., and I can't figure out why . perhaps it slowly drifts into an upper temperature range and thermal cpu throttling means that we end up not clearing the WU and never lowering the temperature ? Are the WU all similar in demands on the processor ? Would all WU's exercise the NEO SIMD optimization that you mention ? That certainly accounts for why the RPi never gets above 62 C Thank you |
|
3)
(Message 8557)
Posted 2 Sep 2024 by jives11 Post: Hi, I have an RPi5 running only boinc, a mix of projects - asteroids , einstein, with a bit of TN-grid when it has work. It's mostly runs fine but once in a while asteroids will cause the machine to get extremely hot i.e 85 degrees. The RPi has a large heatsink above and below the board and under no load runs at around 34 degrees , and for all einstein,TN-grid and *most* asteroids, stays below 75 degrees. However every few weeks it seems I get asteroid workloads that push the heat up significantly i.e a further 10 degrees. I have tried to use preferences to restrict the amount of CPU on the RPi5, but changing this and issuing a boinccmd update appears to be ignored, as does restarting the service. So for the time being I have paused the machine running asteroids until I find a solution. the processor is ARM [Impl 0x41 Arch 8 Variant 0x4 Part 0xd0b Rev 1] the OS is Linux Debian Debian GNU/Linux 12 (bookworm) [6.6.31+rpt-rpi-2712|libc 2.36] An example of one of the workunits with this characteristic ,had the name ps_240826_input_49170_72_1 The machine is dedicated to boinc, is headless and has no gui running. a companion Rpi3 runs asteroids fine all the time and never gets above 65 degrees even with overclocking and tiny stick on heatsinks Why would some work units raise the CPU temperature so much ? One apparently unrelated thing I noticed which might shed some light is when I got the RPi 5 I mistakenly installed a 32bit version of the OS, and this had terrible problems with overheating with boinc, until I realised and went 64bit, after which everything run a lot cooler. Could some work units be using 32bit code ? |