How to fold on CPU of fanless PCs and milti-core experiment

A forum for discussing FAH-related hardware choices and info on actual products (not speculation).

Moderator: Site Moderators

Forum rules
Please read the forum rules before posting.

How to fold on CPU of fanless PCs and milti-core experiment

Postby wezh » Fri Apr 03, 2020 9:34 pm

Date of experiment: March, 31, 2020.

How to fold on CPU of fanless (passive cooling) mini PCs and comparison of folding power (CPU performance) vs CPU temperature vs CPU power consumption.

I have fanless mini PC w/ no videocard. Videocore is inside of Intel Core i5-8250U CPU chip (release date: Q3'17; lithography: 14 nm). It is mini pc (Hystou P03B-i5-8250U) in full-alluminium case (as was selling).
Mini PC powered by Manjaro Linux (based on the Arch Linux OS) 19.0.2 KDE and currently up-to-date.
I have FAHClient v7.5.1-3 terminal (command line) package and FAHControl v7.5.1-1 package w/ GUI installed.

CPU hyper-threading is off in BIOS/UEFI. CPU has 4 physical cores, so max 4 program threads w/o switching activity from one process to another by cores.

Pre-history.

I have CPU with configurable TDP: nominal TDP (15W), TDP-up (25W) and TDP-down (10W).
As on TPD-nominal I saw folding temperature of 65-72C while 2-3 cores using and significant CPU clock drops (throttling), I set up CPU to TDP-down (10W) and made it in BIOS/UEFI.

Preparing experiment.

I turned off almost all user software. Some system health monitoring tools stays remaining such as CPU usage, temperature sensors panel. Network usually consumes CPU power so I turned off network connection in software and pull off network cord from network port to prevent BIOS/UEFI to use network and to load CPU to process that data.

I see that CPU idle is 1-2% of whole CPU (of all cores while average CPU clock speed is about 1400 MHz) load by system + staying running some PC health monitoring tools.

Fanless mini PC located on the table on the height of about 80cm under the floor of the living room.

During folding I never touch keyboard and mice to get so close to clear FAHClient data as I can.

Temperature of air is about 18-20 degrees by Celsius constantly during all experiment (window in another room is partly opened and outside on the street is -4..-6C). It is late evening (23 o'clock).

FAHClient already has WU to process. It is:
Project: 13850 (Run 0, Clone 23970, Gen 5)
Unit: 0x0000000b287234c95e788b551106bb3e

Starting experiment.

************************

1) Folding process with all 4 cores:

Data from FAH Client:
23:09:05:WU01:FS00:0xa7:Completed 330000 out of 500000 steps (66%)
23:12:08:WU01:FS00:0xa7:Completed 335000 out of 500000 steps (67%)
23:15:12:WU01:FS00:0xa7:Completed 340000 out of 500000 steps (68%)
23:18:15:WU01:FS00:0xa7:Completed 345000 out of 500000 steps (69%)

Observations:
CPU temp: 50-52C during all folding process
CPU clock: 1800-1900 MHz per core

Calculation:
Performance: 3 minutes 3.33 seconds per processing of 1% of the WU = 3.06 min/%

************************

2) Folding process with only 3 cores:

Data from FAH Client:
23:21:42:WU01:FS00:0xa7:Completed 350000 out of 500000 steps (70%)
23:25:08:WU01:FS00:0xa7:Completed 355000 out of 500000 steps (71%)
23:28:34:WU01:FS00:0xa7:Completed 360000 out of 500000 steps (72%)
23:32:00:WU01:FS00:0xa7:Completed 365000 out of 500000 steps (73%)

Observations:
CPU temp: 52-54C during all folding process
CPU clock: 2100-2200 MHz per core

Calculation:
Performance: 3 minutes 26 seconds per processing of 1% of the WU = 3.43 min/%

************************

3) Folding process with 2 cores:

Data from FAH Client:
23:36:08:WU01:FS00:0xa7:Completed 370000 out of 500000 steps (74%)
23:40:22:WU01:FS00:0xa7:Completed 375000 out of 500000 steps (75%)
23:44:36:WU01:FS00:0xa7:Completed 380000 out of 500000 steps (76%)
23:48:50:WU01:FS00:0xa7:Completed 385000 out of 500000 steps (77%)

Observations:
CPU temp: 54-57C during all folding process
CPU clock: 2400-2550 MHz per core

Calculation:
Performance: 4 minutes 14 seconds per processing of 1% of the WU = 4.23 min/%

************************

4) Folding process with 1/single core:

Data from FAH Client:
23:54:28:WU01:FS00:0xa7:Completed 390000 out of 500000 steps (78%)
00:00:53:WU01:FS00:0xa7:Completed 395000 out of 500000 steps (79%)
00:07:18:WU01:FS00:0xa7:Completed 400000 out of 500000 steps (80%)

Observations:
CPU temp: 57-60C during all folding process
CPU clock: 2900-3050 MHz

Calculation:
Performance: 6 minutes 25 seconds per processing of 1% of the WU = 6.42 min/%

************************

5) Directly after that I switched to 4 cores folding for 10 minutes.

Observation:
CPU temp: 52-54C.

Conclusion:
Probably full-metal mini PC case heated up but that CPU temp shows that 4 cores folding has less CPU temp than using less CPU cores.

************************

Additional measured data:

Measured consumed CPU power in s-tui v1.0.0-beta3 monitoring tool of all foldings (single core, 2 cores, 3 cores and all 4 cores) are equal: 10W +/- 1 W. Can't say exactly cause scale has that discrete values.

Power consumption of mini PC and its power adaptor with idle CPU: 9.5-9.7W (by Brennenstuhl PM 231 E physical watt-meter device for power outlet), while CPU package consumes 2.1 +/-0.15W (by s-tui monitoring tool).

Whole mini PC including its power adaptor power consumption from mains (power outlet, by physical watt-meter device):
1 core: 18.7-19.3W
2 cores: 19.2-19.8W
3 cores: 19.3-20.1W
4 cores: 19.7-20.1W

Power adaptor power consumption while no load (mini pc is not connected): 0.0W (by physical watt-meter).
Power adaptor power consumption while mini pc is connected, but turned off: 1.3-1.4W (by physical watt-meter).

Conclusions:

1. The less CPU core using, the more CPU clock you get.
2. The more CPU clock you get, the bigger CPU core temperature you get.
3. No matter how many cores you use: is will be always the same CPU power consumed. Max consumed power by CPU based on TDP profile you select.
4. No matter how many cores you use: is will be always near the same power consumed by whole mini PC device.
5. The more cores in use, the more performance you get w/ almost ideally the same consumed power.

Want to fold with lower CPU temperature? Increase CPU cores number in use.
Want to fold faster? Increase CPU cores number in use.
Want to fold with less power consumed? It is near the same: no matter a single core you use or all 4 cores in use.



************************

Folding points per day (PPD) score

Folding PPD for Intel Core i5-8250U w/ configured TDP-down in BIOS/UEFI (to 10W) and in my fanless mini pc is about 4.7k (4700 points per day) if to load all 4 cores with hyper-threading CPU feature turned off (1 physical core = 1 program thread).
And this is under almost ideal circumstances: no user processes, no keyboard and mice activity, even no network, all 24 hours unstoppable.

Calculation base:
Processing of 1% of the WU having 1000 points of base credit takes 3 min 3 sec (3.06 min).
If to imagine that all folding points of base credit of all possible WUs has the same processing performance speed - i.e. the same processing complexity per point - then extrapolation will be like this:
1% of WU has credit of: 1000 points / 100 = 10 points.
How many intervals of 3.06 min is in 24 hours: 24 * 60 / 3.06 = 470.6
So, 470.6 intervals by 10 points each gives in total: 470.6 * 10 = 4706 points

Note: All that calculations made on the WU, now a WU. Moreover: for base credit of the WU. There are more factors for actual estimated credit you see in your score and it is just not less than base credit. I saw that for some WU with about 250 base credit I finally got 1139 points after result was uploaded. See more.

************************

Summary of TDP-nominal (4 cores) vs TDP-down (4 cores) comparison data.

Later on April, 5, 2020, on another WU of another project.
Project: 14372 (Run 102, Clone 3, Gen 3)
Unit: 0x000000049bf7a4d65e84ace36877a6de

4 cores, TDP-nominal (15W):

Observation:
FAHClient data:
17:14:58:WU00:FS00:0xa7:Completed 500000 out of 2500000 steps (20%)
17:16:22:WU00:FS00:0xa7:Completed 525000 out of 2500000 steps (21%)
17:17:43:WU00:FS00:0xa7:Completed 550000 out of 2500000 steps (22%)
17:19:06:WU00:FS00:0xa7:Completed 575000 out of 2500000 steps (23%)
17:20:30:WU00:FS00:0xa7:Completed 600000 out of 2500000 steps (24%)
17:21:52:WU00:FS00:0xa7:Completed 625000 out of 2500000 steps (25%)
17:23:15:WU00:FS00:0xa7:Completed 650000 out of 2500000 steps (26%)
17:24:38:WU00:FS00:0xa7:Completed 675000 out of 2500000 steps (27%)

CPU clock: 2600-2700 MHz.
CPU temp was 60-61C at the start and 63-64C at the end of experiment (window to the street was open, temp outside is 2-3C, it is sunset, 17 o'clock).
Power consumption of CPU package shown in s-tui tool: 20 +/- 2.1W.
Whole mini PC including its power adaptor power consumption from mains (power outlet, by physical watt-meter device): 31.5-32.1W.

Calculation:
Performance: (84 + 79 + 83 + 84 + 82 + 83 + 83) / 7 = 82.57142857142857 = 83 seconds per 1% of the WU.

4 cores, TDP-down (10W):

Observation:
FAHClient data: <not preserved, there was only 3 complete steps with minimum deviation - it's too cold air in the living room>.
CPU clock: <not measured, suppose to be as was in earlier experiment dated as was on March, 31, 2020>.
CPU temp was 49-50C at the start and 52-53C at the end of experiment (window to the street was open, temp outside is 2-3C, it is sunset).
Power consumption of CPU package shown in s-tui tool: 9.9 +/- 1.1W.
Whole mini PC including its power adaptor power consumption from mains (power outlet, by physical watt-meter device): 18.4-19.2W.

Calculation:
Performance: (116 + 115 + 115) / 7 = 115.(3) = 115 seconds per 1% of the same WU.

Note:
Different wattage of the whole PC on 31 March, 2020 comparing to April, 5, 2020 can be caused by difference of sine wave in voltage, shape, harmonics distortion (switching power supplies adds distortion up to 200th harmonic) from power outlet (work day and free day, late and early evening). All other factors seems to be equal.

Conclusions if to switch from TDP-down to TDP-nominal, 4 cores folding:
Performance increase in 115 / 83 = 1.39 = 1.4 times.
CPU package power consumption increase in 20 / 9.9 = 2 times.
Whole mini PC power consumption increase is [ ( 31.5 + 32.1 ) / 2 ] / [ ( 18.4 + 19.2 ) / 2 ] = 1.69 = 1.7 times.
CPU temperature increased by 64 - 53 = 10C. I think it will be greater difference in CPU temperature values if to fold not 20 minutes only but for several hours a day (for example then pc is in idle state during day and nighttime) or 24/7.

Warning note:
Please make sure that temperature of power adaptor is also good while highly loaded such as folding process. Be aware: high temperatures can cause power adaptors/power supply units to become flammable.

Included no-name power adaptor (12V, 5A) of that Chinese fanless mini pc is fully silent but has temperature of 41-46C under previous typical usage (web browser, text editor, w/o folding process) on TPD-nominal. I think that 46C is not good for cheap power adaptor w/ only user load, so not heavy load. And that temperature has a external plastic surface of full-plastic case of power adaptor, so electrical components inside of it are greatly more hotter (as plastic has low thermal conductivity).
Temperature measured by cheapest "Vastar Digital GM320" physical pyrometer w/ scale of -50...380 Celsius.
I bought another power adaptor (12V, 6.67A) MeanWell GST90A12-P1M: it has max temperature of 35C of it's plastic surface (TDP-down, while folding for hours whole night on all 4 cores) but I can hear it from distance of 5 meters away from it. Not always: it produces certain sounds under certain power load: it is very quiet sounds like basketball players snickers makes on a floor of basketball game court.

15 May 2020 UPDATE: consolidation and results info:

Chinese fanless Mini PC Hystou P03B-i5-8250U in full-aluminium case.
Folding on Intel Core-i5-8250U CPU only.
CPU TDP-normal is 15W; using TDP-down 10W (to prevent overheating).
Maximum CPU temperature is +37C addition above a room's air temp and it is max I ever saw in TDP-down mode.
Hyper-threading feature is off (cause need high single-thread performance in several apps).
According to server stats (i.e. de facto) I have PPD of +15107 per day in average day of the first 2 weeks of May, 2020 (FAHClient v7.6.9, v7.6.13; Manjaro Linux OS) and I work on PC in parallel if to fold 24/7 with 3 CPU cores of 4.
The CPU Efficiency: 1.5 kPPD/W (in the mentioned config and work mode above).
Whole Mini PC and it's power adaptor consumes about 18-25W from power outlet (230V@50Hz).
---
https://stats.foldingathome.org/donor/90790185

7 Jun 2020 UPDATE: for last 41 days de facto (by server stats) got 607K points. So average PPD for that period is 14805 for my usage profile (Intel i5-8250U CPU only, TDP down 10W (instead of TDP nominal 15 W), hyper-threading is turned off, folding on 3 cores of 4 total, 24/7, everyday working on this fanless mini pc in parallel).
Last edited by wezh on Sun Jun 07, 2020 4:48 am, edited 17 times in total.
wezh
 
Posts: 12
Joined: Mon Mar 30, 2020 12:34 pm

Re: How to fold on CPU of fanless PCs and milti-core experim

Postby v00d00 » Wed Apr 08, 2020 9:20 pm

While not an ITX system, my 8320e is passively cooled and running 6 cpu cores dedicated to the cpu client and 1 core dedicated to gpu (the 8th core is used for running programs and X). The CPU has a Zalman-FX70 for cooling.

Image

CPU temps are around 40C with 7 cores maxed out. 2x 120mm fans running at about 500rpm on the top of the case draw the heat off. But even with the fans off the heat never seems to get past 46C

Unfortunately it doesnt seem to fit any sockets after AM3+ other wise i'd have put one in my gaming system.
Image
User avatar
v00d00
 
Posts: 363
Joined: Sun Dec 02, 2007 5:53 am
Location: UK


Return to FAH Hardware

Who is online

Users browsing this forum: sirdakz03 and 3 guests

cron