Message boards : Number crunching : Running on a 4GB Raspberry Pi 4 - How to?
Previous · 1 · 2 · 3 · 4 · 5 · 6 · Next
Author | Message |
---|---|
Endgame124 Send message Joined: 19 Mar 20 Posts: 63 Credit: 19,876,234 RAC: 6,134 |
somewhat off topic, there is this ice-tower cpu cooler for Pi 4 that seem to have pretty good soc cooling performance, you can find them on aliexpress The FLIRC case I used (https://www.amazon.com/Flirc-Raspberry-Pi-Case-Silver/dp/B07WG4DW52?ref_=ast_bbp_dp) seems to be quite sufficient with no fan running. I let it run for 2 hours "stock" with the plastic top on and the highest temp I saw was 58c. Just popping off the removable plastic top and letting it run more 2 hours yielded a max temp of 55c. I then took an old heatsink I had on hand for a P3 (Thermaltake Volcano 6 cu) and just placed it on top of the FLIRC case, let it run 2 more hours, and the temp dropped to a high of 51c. All of this is just passively cooled - I'll probably eventually setup a 120mm fan to blow across a stack of Pi 4s if the performance of this one turns out to be what I'm looking for. |
Endgame124 Send message Joined: 19 Mar 20 Posts: 63 Credit: 19,876,234 RAC: 6,134 |
More Pi 4 in Flirc Case numbers: Ambient temp 68 degrees (my basement is pretty steady temp wise, but I didn't check between tests if the pi had increased the basement temp) Each configuration was run for 2 hours, and temps were recorded at 10 second intervals. Max temp recorded (unfortunately I discarded the dataset between runs, so I can't give average temp) HDMI is disabled for the testing. w/ top on 58c w/ top off 55c w/ top off + heatsink 51c w/ top off + heatsink + 60mm fan (side) 38c w/ top off + heatsink + 60mm fan (front) 35c Questions to be answered: Should I have added VGA heatsinks inside the closed case for other chips, such as the PMIC chip, or is that useless since there is no airflow inside the case? Would Ubuntu with Linux Kernel 5 (native raspberry pi 4 support) do more work at similar wattage / temp? |
akc42 Send message Joined: 24 Mar 20 Posts: 7 Credit: 709,575 RAC: 0 |
I have 3 x Raspberry Pi 4s running now. 1st one has a plastic case with slots in and a fan - its been running flat out 4 tasks at a time for a couple of weeks. Temp is 54degC. The fan is on much of the time, although no sign yet of any problem. 2nd one is only 2GB so runs one sometimes two tasks. It has a Flikr case and has a temp of 47deg C. I have been using it for other things the 3rd one has been running about 3 days now. I specifically bought it to add to my collection running rosetta@home and decided I would buy the case that is the heatsink, so only passive cooling and therefore no fan to wear out. Its running flat out 4 tasks at once and has been for a couple of days now. Its temp is 63degC. Even the hottest isn't excessively so. |
sgaboinc Send message Joined: 2 Apr 14 Posts: 282 Credit: 208,966 RAC: 0 |
hi some quick questions, the raspbian buster distribution is 32 bits i think. even for pi4 thanks, yeah, my Pi 4 is now running rosetta@home ;) one of the task looks like this https://boinc.bakerlab.org/rosetta/workunit.php?wuid=1043563907 i'm using my little low cost heat sink no fan https://www.raspberrypi.org/forums/viewtopic.php?f=29&t=271933 it idles just below 50 deg C 5 minutes into the run temperatures are now 67 deg C, now a little less at 65 deg C, now even lesser 62 deg C, back up at 70 C still running, a little fan does the magic 44 deg C, yes it is still running, full speed on r@h it is running 4 threads concurrently ! i think i'd upgrade my memory card to a bigger capacity one, sd utilization now runs at 98% used for a 16 GB card! |
sgaboinc Send message Joined: 2 Apr 14 Posts: 282 Credit: 208,966 RAC: 0 |
-... |
bkil Send message Joined: 11 Jan 20 Posts: 97 Credit: 4,433,288 RAC: 0 |
It's good to know that a heat sink can prove enough in milder climates. At first, Raspberry Pi 4 was known to only run acceptably with a fan. Did you apply all techniques for saving power and reducing temperatures mentioned in this thread? I think these were:
- Use WiFi instead of ethernet - Don't use an enclosure or use a metallic one - Disable unused peripherals - Undervolt - Underclock unused domains (TODO: underclock other things and measure the difference) - Use an efficient SD card - Operate the device perpendicular to the ground https://www.raspberrypi.org/blog/free-raspberry-pi-4-cooling-stand-with-the-magpi-90/ |
sgaboinc Send message Joined: 2 Apr 14 Posts: 282 Credit: 208,966 RAC: 0 |
oh wow, it certainly takes a long time (near 6 hours) but Pi4 did it, running one of the loads that i tend to see on PCs as well https://boinc.bakerlab.org/rosetta/result.php?resultid=1160487413 it seemed a Pi4 core is about 1/5 - 1/3 that of a haswell desktop 'hyperthread' core. but these days i'm seeing high core counts desktop processors doing much more threads and prowess than my modest haswell desktop i've updated some additional charts at the Rpi forum during the run. https://www.raspberrypi.org/forums/viewtopic.php?f=29&t=271933&p=1649414#p1649298 for now these are strapped together with loose parts, i'd need to work on making it kind of a 'case' so that i can conveniently simply run it. And as i'm able to get it down to 40 deg C running 4 concurrent threads. after i fixed up things i may even try overclocking it instead. i think right now it is doing like 1.5 Ghz as 'stock' speeds. it is pretty encouraging that Pi4 completed about 6 jobs thus far taking a longer stretch of time (i think like 1.5 times longer, not really measured, and perhaps returning less results) than on the desktop but nevertheless completing them |
sgaboinc Send message Joined: 2 Apr 14 Posts: 282 Credit: 208,966 RAC: 0 |
next thing tried, overclock to 1.750 ghz overvoltage 2 /boot/config.txt over_voltage=2 arm_freq=1750 temperature seemed modestly higher 45 - 47 deg C, heat sink + fan blowing, (have a good heat sink + fan blowing when meddling with overclocks, there seem to be a risk if the fan fail or heat sink dislodge, it may have some bad consequences. but otherwise i'm getting figures like 2451 floating point MIPS (Whetstone) per CPU 2.4 Gflops ~ 9.6 Gflops on all 4 cores it is probably about 5-10% more numbers crunched at best, this would probably benefit cases where data is in the onchip cache, if it needs to hit ram, the pipeline probably stall to fetch the data) 4 r@h threads all cores running load average 4.2 CPU frequency: 1750MHz Voltage: 0.9125V Scaling governor: ondemand |
sgaboinc Send message Joined: 2 Apr 14 Posts: 282 Credit: 208,966 RAC: 0 |
completed one round running overclocked to 1.75 Ghz, interestingly for some of the large / difficult models it managed only a single decoy / model https://boinc.bakerlab.org/rosetta/result.php?resultid=1160831838 i've seen quite many times work units are 'waiting for memory, despite setting a high % memory allowance, stalling between runs could have accounted for the poor performance the other wu seem to do just ok and may be return a little more models running overclocked https://boinc.bakerlab.org/rosetta/result.php?resultid=1160831374 temperatures stayed between 40 C - 45 C with heat sink + fan blowing https://boinc.bakerlab.org/rosetta/forum_thread.php?id=13732&postid=95331 |
sgaboinc Send message Joined: 2 Apr 14 Posts: 282 Credit: 208,966 RAC: 0 |
i saw something interesting https://boinc.bakerlab.org/rosetta/cpu_list.php Number of Avg. cores GFLOPS GFLOPs computers /computer /core /computer BCM2835 [Impl 0x41 Arch 8 Variant 0x0 Part 0xd08 Rev 3] 922 4.00 2.07 8.29 Intel(R) Core(TM) i3-2310M CPU @ 2.10GHz [Family 6 Model 42 Stepping 7] 70 3.96 2.07 8.18 that i3 seem to be a notebook computer part number, but i'm not too sure |
bkil Send message Joined: 11 Jan 20 Posts: 97 Credit: 4,433,288 RAC: 0 |
Could you please answer the questions posted above? |
sgaboinc Send message Joined: 2 Apr 14 Posts: 282 Credit: 208,966 RAC: 0 |
"VENETO boboviz" wrote: hi some quick questions, the raspbian buster distribution is 32 bits i think. even for pi4 follow this guide as quoted from VENETO boboviz, it works |
Endgame124 Send message Joined: 19 Mar 20 Posts: 63 Credit: 19,876,234 RAC: 6,134 |
I just got around to disabling Wifi and Bluetooth on my Pi 4. My APC 1500G UPS is my primary source of measuring power usage so its not the most accurate measurement, but its not bad. The pi 4 is fully updated with the newest firmware. With the Pi off, the UPS reads 28W (this is my wifi router, my Fios Modem, my Fios termination point box, etc). Pi on, hdmi off, idle. 29 W. Pi on, hdmi off and 4 Rosetta threads @ 100% CPU: 33W Pi on, hdmi off, Wifi off, Bluetooth off, led offand 4 Rosetta threads @ 100% CPU: UPS alternates between 31 and 32W likely the Pi 4 is running somewhere around 3.5 watts and the UPS can't quite decide if it should round down or round up. I've read that the ethernet is no longer attached to the USB bus (this is how they get real gigabit performance on the ethernet), so I may try turning off USB as well. With no USB devices, it may not make much of a difference. |
Boulybud Send message Joined: 3 Apr 20 Posts: 11 Credit: 351,250 RAC: 0 |
Hello Engame124, Thank you for your return, it's very disturbing. Have you done temperature readings in your different scenarios? I'm curious if disabling HDMI, BT, WIFI has an impact on the temperature CPU ? Thank you. |
Endgame124 Send message Joined: 19 Mar 20 Posts: 63 Credit: 19,876,234 RAC: 6,134 |
Hello Engame124, Flirc Case Max Temps over 2 hours per test, hdmi off (sudo tvservice --off) for all tests: 1) w/ top on 58c 2) w/ top off 55c 3) w/ top off + heatsink 51c (old thermaltake volcano 6cu heatsink) 4) w/ top off + heatsink + 60mm fan (side) 38c 5) w/ top off + heatsink + 60mm fan (front) 35c 6) w/ top off + 60mm fan (front) 38c 7) w/ top off + 60mm fan (side) 40c 8) w/ top on + 60mm fan (front) 42c 9) w/ top on + 60mm fan (front), wifi off, bluetooth off, led off 42c 10) w/ top on + 60mm fan (front), wifi off, bluetooth off, led off + Stacked (Pi 3B+ in flirc case on top) 43c I'll turn off the fan and unstack the pi 3 and 4 and see where the temp ends up. *edit* In case you're wondering, I'm getting the max temp like this: while true; do temp=`/opt/vc/bin/vcgencmd measure_temp | awk -F "=" '{print $2}' | awk -F "'" '{print int ( $1 )}'`; if [ $temp -gt $maxtemp ]; then maxtemp=$temp; fi; echo $maxtemp; sleep 10; done |
sgaboinc Send message Joined: 2 Apr 14 Posts: 282 Credit: 208,966 RAC: 0 |
another way is install rpi-monitor https://rpi-experiences.blogspot.com/p/rpi-monitor.html https://github.com/XavierBerger/RPi-Monitor you get nice temperature graphs and lots of other statistics on the web. another thing is in /etc/boinc/cc_config.xml, i entered <options> <allow_remote_gui_rpc>1</allow_remote_gui_rpc> </options> then you need in addition the password found in /var/lib/boinc/gui_rpc_auth.cfg this is a sym-link to /etc/boinc/gui_rpc_auth.cfg if the password can't be found in it, you could try deleting that sym-link and restarting boinc-client. i found it re-generates the password. then you can connect to it from remote in boinc-gui there is some kind of risk if this is after all on a public network, but on a private lan that you are in control, it is perhaps safe |
Endgame124 Send message Joined: 19 Mar 20 Posts: 63 Credit: 19,876,234 RAC: 6,134 |
Additional temp data 11) Passive w/ top on, hdmi off, wifi off, bluetooth off, led off. 58c Looks like turning off bluetooth / wifi / led makes no temp difference. I have now acquired another PI 4. This one I'm overclocking with these settings: over_voltage=6 arm_freq=2125 UPS reads a solid 5 watts for this configuration, so a 41.7% increase in clock speed requires a 42.9% increase in power. 2147 is the maximum frequency, so I have a little room if I want to add more clock. I'm not sure over_voltage 6 is required for this clock speed, I'll probably try backing voltage down if it looks like its stable at 2125. Apparently the GPU clock is attached to things like the bus speed, so there may be additional gains to be had increasing the GPU speed, even though I have hdmi off. First things first, though, and the first is just verifying stability. |
Endgame124 Send message Joined: 19 Mar 20 Posts: 63 Credit: 19,876,234 RAC: 6,134 |
My pi 4 4GB was NOT stable at over_voltage 6 and clockspeed of 2125 Mhz with 4 Rosetta threads, though 3 threads worked fine. With 4 threads, it would run for 3-4 minutes, then spontaneously reboot as soon as the 4th rosetta process finished loading. I also tried over_voltage 4 and 2000Mhz with the same behavior, so I'm now trying at over_voltage 4 and clockspeed of 1900, and it seems stable over the last 30 minutes. I'll try pushing 2000Mhz at over_voltage 4 later |
bkil Send message Joined: 11 Jan 20 Posts: 97 Credit: 4,433,288 RAC: 0 |
That sounds nice, especially the default consumption of 3.5W. Could you perhaps plot a diagram with each attempted frequency and the lowest over_voltage at the given frequency? It can also be set to a negative number. I'm interested in how much undervolting counts in temperatures. |
Endgame124 Send message Joined: 19 Mar 20 Posts: 63 Credit: 19,876,234 RAC: 6,134 |
That sounds nice, especially the default consumption of 3.5W. Could you perhaps plot a diagram with each attempted frequency and the lowest over_voltage at the given frequency? It can also be set to a negative number. I'm interested in how much undervolting counts in temperatures. Would a table like this work? Freq Over_Volt Result on 4 Units 2125 6 Fail 2100 6 Fail 1900 4 Pass Of note, the rebooting caused from the testing at 2125 and 2100 caused my work units to fail. The testing I did at over_volt 4 and 2000 freq seemed to be affected by this - the work units were canceled by Boinc itself at some point, and the 4 completed at 1900 were different units. I'm now trying 2015 freq on over_volt 4 and it seems to be working fine (2025 caused the random hangs - the OS wouldn't crash, but it would get "stuck" for 20 seconds, then recover, then get stuck again). |
Message boards :
Number crunching :
Running on a 4GB Raspberry Pi 4 - How to?
©2024 University of Washington
https://www.bakerlab.org