Topton Jasper Lake Quad i225V Mini PC Report

[tusk9541]

Maybe what we need is one of those USB-PD+Buck converter combo deals.

TBH I'm a bit skeptical of how much power it would save, but I'll be testing one cable next week and compare it with the in-box 12V power supply. Adding another voltage conversion would probably be less efficient, but dunno if it would be significant as well.

If you're looking for a USB-PD charger, it's not hard to find ones that can output 12V either, it's just something to keep in mind. If it's true that the units can accept 15V as well, then you can buy a 15V or a multi-volt trigger.

 

[oneplane]

TBH I'm a bit skeptical of how much power it would save, but I'll be testing one cable next week and compare it with the in-box 12V power supply. Adding another voltage conversion would probably be less efficient, but dunno if it would be significant as well.

If you're looking for a USB-PD charger, it's not hard to find ones that can output 12V either, it's just something to keep in mind. If it's true that the units can accept 15V as well, then you can buy a 15V or a multi-volt trigger.

It's more in terms of having a common DC power bus and less just to swap one adapter for another. (for me at least) If the devices were all accepting USB-PD, you'd only need a USB-PD feed to power them all.

 

[spengbab]

Some perspective here. If you've seen my posts in the last few pages, bear in mind that my unit is the one with the smallest heatsink available. Most people are buying one with the heavier/bigger heatsinks.

And even when I got it and I tested it without modification, Prime95 on Windows was making the CPU throttle at about 90°C after about 30 minutes. You'll pretty much never encounter this scenario in real world usage, Prime95 is a torture test, it doesn't test real world conditions. People usually use it to test stability in overclocked systems. In all likelihood, for what I'm using it (bare metal OPNsense), just the way it came would have been just fine.

I've done a little modification and testing temps just out of curiosity and just to get a feel on DIY fanless heatsinks cause I'm gonna be modifying another system that has a fan to make it fanless, but what I've shared in those posts is by no means necessary for this unit.

That said, you may wanna test yours first, cause there may be some variation, while mine does have contact between the CPU die and the copper block, it's just barely so the cooling can be improved quite a bit with some pressure. I have not used a copper shim like others, but just a contraption I built in a pinch with a couple springs, a silicone foot and thick 3M double sided silicone adhesive strips. Just this bit of pressure made it so the same Prime95 test doesn't throttle at all, and temps stabilize at about 70°C.

Re: heatpipes and vapor chambers, I actually got a couple of those vapor chambers, and was going to test them, so I'll post the results hopefully soon. I got the 3x60x100mm ones. I still have to cut up a couple of copper plates, one .5mm one to fill in the cutout of the copper block that sinks a bit into the heatsink, and another about .8-1mm to act as the new heat spreader. I'm mostly interested to see how it affects the heat pattern that I got in the thermal image in the last page.

Thanks for posting this, not because it makes me think mine will be okay, but it balances things out a bit.

I placed an order for https://www.aliexpress.com/item/1005004822965821.html yesterday morning. Found this thread in the afternoon, read about 100 pages and had a bit of a meltdown and even cancelled the order XD. I resumed the order, pretty much on the understanding that it's a crapshoot and will be a nightmare if mine has anything other than thermal problems, so we'll wait and see what I get - I'll report back here with my findings, positive and/or negative.

Which box did you get? Given that you said "my unit is the one with the smallest heatsink available" I'm assuming the same one as I ordered (pic attached)? Just looked back and saw you got https://www.aliexpress.us/item/3256804761963374.html right? I guess the upside is, if I get any other box, the heatsink will probably be better than the one I ordered, haha.

 

[anterus]

thank you all for a good thread. Recap after 125 pages:
- Optimizations make sense if your internet speed is below 1 Gb. What I noticed, is that on 1.9G connection that I have - setting ASPM = Auto , disabling turbo, setting tunables ( dev.hwpstate_intel.0.epp =0 or hw.acpi.cpu.cx_lowest = C3 ) reduces the throughput to below 1G (usually around 700Mbps or lower), and so does enabling PowerD. All of this on bare metal OPNsense (latest).
- CWWK is of course the best, I got 4 port v5 N5105 unit that has black connector to the second NVME. No holes between CPU and NICs and the case. Current idle temps are 50C without BIOS tuning or tunables.
- For many use cases, the 1G speed is the practical max. For example, NordVPN when connected via my 1.9 line, gives at the most 1G speed or below.
- Enabling Suricata without virus filtering rules (most cpu-intensive) provide the throughput of around 400-600 Mbps on 1.9G line.
- Always update CPU microcode, it fixes bugs, and if you are using Proxmox - take a look at updating to the latest kernel.

Few questions still remain:
- How do you find the quality of CWWK thermal paste, if there are no gaps, did you replace it with good thermal paste (e.g., ArcticSilver, or Noctua), and did you see lower temps?
- What are your Suricata throughput speeds on N5105 on >= 1G connections, and which rules have you enabled? Is there anything that can be done to improve IDS/IPS throughput speed?
- There were few recommendations regarding how to improve PPPoE throughput, does anyone have recommendations about how to improve throughput further? May consider upgrading speed to 3Gb line.

 

[Immortal]

Suricata isn't really very well optimized for MT, the newest version has some improvments and will be shipped with OPNsense update 23.7 so like a month or two from now.

That being said, there is no way you can run it on N5105 with decent amount of rules and get like > 1 Gb/s connection speed. The only optimization that can be done here is pick and chose rules based on your needs. Even then this will be a hard ask of N5105. If thats your goal, you should upgrade to something like i5/i7-1260p or Ryzen 5600U+.

 

[rabman]

So I've been having a super intermittent and fussy problem with my new Topton N5105 i226 (x4 NICs).

For hardware I'm running a HP EX900 Pro 265GB NVMe SSD, 2x8GB of Crucial DDR4-3000 (running at 2933); SSD passes SMART extended, and RAM passes memtest86+. I'm running Proxmox on an EXT4 partition with two virtual machines; OPNsense (with three of the four i226 NICs passed through directly to the VM, the last NIC is for the Proxmox on bare metal) and Ubuntu (virtual networking).

Now for the problem: Both virtual machines seem to randomly hang, with no rhyme or reason that I can gather. The proxmox host, itself, has been rock solid. The VMs don't crash at the same time and they can be up and running just fine anywhere between several hours to 4-5 days. I have changed all sorts of settings for the VMs from machine type, host type, and most other guest settings that you can tweak.

Has anyone seen an issue like this before? Any suggestions would be more than welcome. I'm not quite to the pulling my hair out stage but the high intermittency of the problem is making this a real pain to troubleshoot.

I've had similar problems with VM's hanging or rebooting on my N5105 (I have OPNSense and DietPI VMs) and stability is much better with these three changes:

1. Updating the Proxmox kernel to 6.1
   • apt update
   • apt install pve-kernel-6.1
2. Updating microcode after appending "non-free" to each deb line of /etc/apt/sources.list
   • apt update
   • apt install intel-microcode
3. Preventing the kernel from entering a cstate lower than C1. My config uses grub (/etc/default/grub, iommu settings are for i226 passthrough)
   • GRUB_CMDLINE_LINUX_DEFAULT="quiet intel_iommu=on iommu=pt intel_idle.max_cstate=1"
   • then run update-grub

I've run almost two weeks without issue but had to reboot to update to OPNSense 23.1 so don't have more time than that. I'm not sure which modification makes the difference -- I haven't tried rolling them back individually to root cause.

 

[Stovar]

Few questions still remain:
- How do you find the quality of CWWK thermal paste, if there are no gaps, did you replace it with good thermal paste (e.g., ArcticSilver, or Noctua), and did you see lower temps?

It felt fine to myself, just a normal silver type blob of paste. I took it apart and while its very tricky shining a torch between the copper heatsink and cpu contact I don't believe there was a gap but its always possible there was one. Removing the mobo though I could see my cwwk model the cpu left a imprint on the thermal paste (both sections). Running Prime95 all cores were 100% and hit around 52c max for 1 hour.

This was with the chucked in fan on the underside of unit and entire unit felt hot but warm enough to hold in hands for a minute or 2, rebooting and checking bios health cpu was at 47c.

I hope that means there is decent enough cpu contact there, if not its still working fine under load for an hour with good cpu temps.

 

[Stovar]

Just in case anyone is wishing to open up their unit (mine is a CWWK 6 nic unit build dated 1/2023), its very easy takes few minutes just keep the screws together in sections ie you remove 4 black ones you put them all in one area. I just put them in blue tack ball and away from my working area.

Some pics in case

4 black screws on top


4 black screws on other side


Inside 8 screws (mixed silver types)



You can take pictures of the insides of the units, I forgot to do the underside pic but that is the easiest one just remember there are screws hidden under the ram ddr slot and also there might be one under the nvme so its best to remove all ram and storage to get to those hidden screws before removing the entire mainboard from unit.




That was the cpu contact, I shined a torch but couldn't see a gap but you never know so ran prime95 for an hour, with all 4 cpu cores maxed it hit
around 48-52c max fine. Unit is hot but warm enough to hold, bios temp reported 47c so it feels stable and good this is with an included small fan
running on 7volt mod adaptor for silent cooling. Probably don't need the fan but ill leave it there in case + it cools the nvme and ram.

 

[Stovar]

The WooYi N5100 i225 box arrived today. Ordered it the 22nd of Oct, it was marked as shipped on the 24th, so 2 weeks total.

It came with a knockoff "Delta" 12V 4A power supply. Looking up the model number shows a 48V PSU, but the build quality, sticker, weight and connectors are also a giveaway. I guess I shouldn't expect a name brand PSU with a $113 PC though.

I have the 1338NP-12 (ver. 1.4) board, and after playing around in the BIOS, I managed to get the idle power down to 5W (measured at the wall), sitting at the Win10 desktop with a 4k monitor plugged in. Without the monitor it goes down to 4W. This is using 2x4GB Samsung 3200 and 256GB Teamgroup MP33 SSD.

The only settings I changed in BIOS was under --- Chipset > PCH-IO Config > PCI-E Root Port (1-8) > ASPM set to Auto.
I did this for each PCI-E root port. EDIT : as well as DMI Link ASPM set to Auto.

I also noticed the seemingly high 1.7v Vcore reading in HWinfo, but this appears to be normal from what others have told me, because these chips use a FIVR (fully integrated voltage regulator) to regulate voltage at the die.

I did a teardown before even testing the machine, and it had a ~0.25mm gap between the aluminum riser block and CPU die. I sanded the scuffs out of the block and repasted it to the chassis, using a vise (with soft jaws ofc) to squeeze out excess paste. Then pasted and pressed the 0.3mm copper shims onto the block. Added kapton tape over the CPU substrate to avoid any risk of shorts.
A ~10 min prime95 small FFT test got the CPU to ~60c. Idle is ~35c.

The default PL is 15W/20W, and all-core turbo is set on override to 2.8GHz. (normally the max single core turbo speed).
Once it drops to the 15W limit, it's running about 2.6GHz all-core.

Thanks for the summary, jak4! About "Fiddling with varios Bios-Settings will most likely be necessary. ", I was going through all the pages and found that this will be mainly those settings here:

• Enable PL1 and set at 10000
• Enable PL2 and set at 12000
• Enable C-states
• Set PECI to off
• Enable turbo (it’s fine once you’ve got proper settings)
• Enable Speedstep
• Set ACPI as auto
• Enable acoustic noise reduction.
• If you use pfsense or opnsense, you can tweak the following tunables

Alternatively, you can also try to just set to ASPM to "Auto" as described here.

Is there anything else to mention?

thanks good info, those figures are decent still and similar to others had here give or take a few watts, many were hitting around 8-9 watts at idle and 15-19W peaking, I think I would be happy with those numbers just out the box alone.

There were some threads here, here, here, here, here that discussed C-states, powerD and stability concerns, still interesting read over on openwrt regarding the iGPU. If that power save work does something use it really, every bit helps you might find some bios tweaks in those links could help just might be good to double check stability too.

I am guessing a lot of us around here are from Europe and effected sharply by the energy crisis or just want that router type low watts feeling, my electricity and gas prices are going up by 40% this April so lots of fun ahead.

thank you all for a good thread. Recap after 125 pages:
- Optimizations make sense if your internet speed is below 1 Gb. What I noticed, is that on 1.9G connection that I have - setting ASPM = Auto , disabling turbo, setting tunables ( dev.hwpstate_intel.0.epp =0 or hw.acpi.cpu.cx_lowest = C3 ) reduces the throughput to below 1G (usually around 700Mbps or lower), and so does enabling PowerD. All of this on bare metal OPNsense (latest).
- CWWK is of course the best, I got 4 port v5 N5105 unit that has black connector to the second NVME. No holes between CPU and NICs and the case. Current idle temps are 50C without BIOS tuning or tunables.
- For many use cases, the 1G speed is the practical max. For example, NordVPN when connected via my 1.9 line, gives at the most 1G speed or below.
- Enabling Suricata without virus filtering rules (most cpu-intensive) provide the throughput of around 400-600 Mbps on 1.9G line.
- Always update CPU microcode, it fixes bugs, and if you are using Proxmox - take a look at updating to the latest kernel.

Few questions still remain:
- How do you find the quality of CWWK thermal paste, if there are no gaps, did you replace it with good thermal paste (e.g., ArcticSilver, or Noctua), and did you see lower temps?
- What are your Suricata throughput speeds on N5105 on >= 1G connections, and which rules have you enabled? Is there anything that can be done to improve IDS/IPS throughput speed?
- There were few recommendations regarding how to improve PPPoE throughput, does anyone have recommendations about how to improve throughput further? May consider upgrading speed to 3Gb line.

I am just quoting some very important post above regarding Power saving bios/tweaks so others can bookmark or refer back to them it saves people reading 120+ pages (but its a good read regardless!)

I also updated some of the tweaks with where to find them exactly in the bios of my CWWK 6 intel nic unit Jan 2023 unit (it may vary for other units and brands)

Advanced → ACPI Settings

Key	Value	Default
Enable ACPI Auto Configuration	Auto	Disabled

Advanced → CPU Configuration

Key	Value	Default
PECI	Disabled	Enabled

Advanced → Power & performance → CPU - Power Management Control

Key	Value	Default
Boot performance mode	Turbo Performance	Turbo Performance
Intel(R) SpeedStep(tm)	Enabled	Enabled
Platform PL1 Enable	Enabled	Enabled
Platform PL1 Power	10000	50000
Platform PL2 Enable	Enabled	Enabled
Platform PL2 Power	12000	65000
C states	Enabled	Enabled

Advanced → Power & performance → CPU - Power Management Control → CPU VR Settings → Acoustic Noise Settings → Acoustic Noise Mitigation

Key	Value	Default
Acoustic noise mitigation	Enabled	Disabled

Chipset → PCH-IO Configuration → PCI Express Configuration

Key	Value	Default
DMI Link ASPM Control	Auto	Auto

Chipset → PCH-IO Configuration → PCI Express Configuration → PCI express root port <1-8>

Key	Value	Default
ASPM	Auto	Disabled

Props to olavrb original post here and also to STH Community since they did all the above.

my older settings guide here but the same as above just in text format:

Spoiler

Enable PL1 and set at 10000(advanced>power and performance>CPU PMC>platform PL1 enable>Platform PL1 power=10000)
Enable PL2 and set at 12000(advanced>power and performance>CPU PMC>platform PL1 enable>Platform PL1 power=12000)
Enable C-states (advanced>power and performance>CPU PMC>C states>enabled)
Set PECI to off (Advanced>CPU configuration>PECI>disabeled)
Enable turbo (it’s fine once you’ve got proper settings) (advanced>power and performance>CPU PMC>Boot performance mode>Turbo Performance)
Enable Speedstep (advanced>power and performance>CPU PMC>Intel speedstep>enabled)
Set ACPI as auto (Advanced>ACPI settings>Enable ACPI auto config>enabled)
Enable acoustic noise reduction ???
I found "acoustic noise reduction" as "acoustic noise mitigation" here:

• Advanced -> Power & performance -> CPU - Power Management Control -> CPU VR Settings -> Acoustic Noise Settings -> Acoustic Noise Mitigation

ASPM (Chipset>PCH-IO Config>PCI Express configuration>DMI Link ASPM Control>Auto
(Chipset>PCH-IO Config>PCI Express configuration>PCI express root port 1-8 make sure each
1-8 has ASPM set to AUTO.
If you use pfsense or opnsense, you can tweak the following tunables
Alternatively, you can also try to just set to ASPM to "Auto" as described here.
[spoiler/]

Hopefully you will find those options under the correct or similar type menu names, these unlocked bios options are a needle in a haystack!

I had default out the box roughly 12-13 watts idle and 17-18watts max on my CWWK 6 nic with 5 ethernet cables attached with bare openwrt.
With the above bios tweaks, its now 8.5-9.1 watts idle and 15-16 watts max.

Not quite that holy grail or my old Asus Router 6-7 watts but I am happy for 8.5-9.1 watts idle since its vastly more powerful I also checked speedtest and I max out my Mulvad VPN with Wireguard with Openwrtx86 on the cwwk mini router with the energy bios tweaks applied.(1gig bb)



If you do find stability or performance issues, it might be better to default bios or adjust one bios setting at a time to rule out the culprit.

 

[EasyRhino]

So I had noticed that when running baremetal opnsense, I wasn't getting full gigabit performance on my internet if I had zenarmor enabled.

Turns out I got a big performance boost by disabling flow control on the i226's. I had to add this to opnsense tunables:

dev.igc.0.fc	flow control	runtime	0
dev.igc.1.fc	Flow Control	runtime	0
dev.igc.2.fc	Flow Control	runtime	0
dev.igc.3.fc

and I can get pretty much full gigabit speed tests now even with zenarmor enabled.

 

[EasyRhino]

Has anyone been using opnsense and traffic shaping with these boxes?

When I set it up with my n5105, my download speeds drop to around 200-300mbps. From gigabit.

Admittedly I don't really need traffic shaping, I just didn't know if it was possible to run it and get near full speeds.

 

[tusk9541]

Just in case anyone is wishing to open up their unit (mine is a CWWK 6 nic unit build dated 1/2023), its very easy takes few minutes just keep the screws together in sections ie you remove 4 black ones you put them all in one area. I just put them in blue tack ball and away from my working area.

Some pics in case

4 black screws on top
View attachment 27248

4 black screws on other side
View attachment 27249

Inside 8 screws (mixed silver types)
View attachment 27250


You can take pictures of the insides of the units, I forgot to do the underside pic but that is the easiest one just remember there are screws hidden under the ram ddr slot and also there might be one under the nvme so its best to remove all ram and storage to get to those hidden screws before removing the entire mainboard from unit.


View attachment 27251

That was the cpu contact, I shined a torch but couldn't see a gap but you never know so ran prime95 for an hour, with all 4 cpu cores maxed it hit
around 48-52c max fine. Unit is hot but warm enough to hold, bios temp reported 47c so it feels stable and good this is with an included small fan
running on 7volt mod adaptor for silent cooling. Probably don't need the fan but ill leave it there in case + it cools the nvme and ram.

That looks much better than mine, I see that they've put screws on 2 corners of the copper block to apply pressure accordingly to the CPU, which mine doesn't have and haven't seen in other pics.

 

[s_p]

Hello guys

I am writing to you from Italy.
Unfortunately I found this forum and this thread too late and I had already bought probably the worst fanless mini pc from TopTon.
The current software configuration is Proxmox and 4 VMs (OpenMPTCP + 2 instances of ROOter + OMV) and 1 CT (AdGuard).
The hardware is:
N5105 cpu
i226-V NIC
Ram 16GB Kingston FURY Impact
500GN nvme Crucial P3 with heatsink for the OS and VMs
Samsung SSD 870 EVO 500GB for OMV.

I replaced the thermal paste with Arctic and put a thermal pad between the aluminum heatsink and the case.

Reading the various messages I set the bios with these:

Enable PL1 and set at 10000
Enable PL2 and set at 12000
Enable C-states
Enable acoustic noise reduction
Set PECI to off
Enable turbo
Set ACPI as auto

in addition i have installed TLP.


Despite all this, the CPU temperature is always at least 60°C and the nvme temperature at least 62°C

I'm waiting for a 120mm usb fan to put under the case, in the meantime could I do something to improve the temperatures?

 

[spengbab]

Hello guys

I am writing to you from Italy.
Unfortunately I found this forum and this thread too late and I had already bought probably the worst fanless mini pc from TopTon.
The current software configuration is Proxmox and 4 VMs (OpenMPTCP + 2 instances of ROOter + OMV) and 1 CT (AdGuard).
The hardware is:
N5105 cpu
i226-V NIC
Ram 16GB Kingston FURY Impact
500GN nvme Crucial P3 with heatsink for the OS and VMs
Samsung SSD 870 EVO 500GB for OMV.

I replaced the thermal paste with Arctic and put a thermal pad between the aluminum heatsink and the case.

Reading the various messages I set the bios with these:

Enable PL1 and set at 10000
Enable PL2 and set at 12000
Enable C-states
Enable acoustic noise reduction
Set PECI to off
Enable turbo
Set ACPI as auto

in addition i have installed TLP.


Despite all this, the CPU temperature is always at least 60°C and the nvme temperature at least 62°C

I'm waiting for a 120mm usb fan to put under the case, in the meantime could I do something to improve the temperatures?

Uh oh, this is the exact one I have coming I was wondering if you could get airflow through vents on the sides? Please report back when you've tried the 120mm fan!

You say it's at least 60c, what does it go up to under sustained load?

 

[Immortal]

Hello guys

I am writing to you from Italy.
Unfortunately I found this forum and this thread too late and I had already bought probably the worst fanless mini pc from TopTon.
The current software configuration is Proxmox and 4 VMs (OpenMPTCP + 2 instances of ROOter + OMV) and 1 CT (AdGuard).
The hardware is:
N5105 cpu
i226-V NIC
Ram 16GB Kingston FURY Impact
500GN nvme Crucial P3 with heatsink for the OS and VMs
Samsung SSD 870 EVO 500GB for OMV.

I replaced the thermal paste with Arctic and put a thermal pad between the aluminum heatsink and the case.

Reading the various messages I set the bios with these:

Enable PL1 and set at 10000
Enable PL2 and set at 12000
Enable C-states
Enable acoustic noise reduction
Set PECI to off
Enable turbo
Set ACPI as auto

in addition i have installed TLP.


Despite all this, the CPU temperature is always at least 60°C and the nvme temperature at least 62°C

I'm waiting for a 120mm usb fan to put under the case, in the meantime could I do something to improve the temperatures?

You're running Proxmox and 4 VMs so the system is pretty much never at idle, so 60C is pretty normal here without a fan and you will hit even higher under load even with a fan.

 

[s_p]

Uh oh, this is the exact one I have coming I was wondering if you could get airflow through vents on the sides? Please report back when you've tried the 120mm fan!

You say it's at least 60c, what does it go up to under sustained load?

after 4 minutes of stress test with 100% cpu, temperature is 64°C-66°C

 

[andrea87]

...
I replaced the thermal paste with Arctic and put a thermal pad between the aluminum heatsink and the case.

I'm waiting for a 120mm usb fan to put under the case, in the meantime could I do something to improve the temperatures?

What is your average cpu load with the four vms running?

The heatsink itself is not designed to be operated horizontally with no airflow, as it will saturate easily and have no convection taking the heat away from its "channels". Running it vertically might help a bit, but i don't expect a massive change, perhaps a few degrees.

What I'm more concerned at this point is the SSD, if you're running the cpu (and board) at 60-65°C I can imagine the SSD is doing 70-75°C at least in that tiny case. Install hddtemp, and try probing your ssd (use lsblk to get it's name) for it's running temperature.

If you have space between the cpu and heatsink, I would instead of using a pad between the spreader and the heatsink itself, use a thin copper / aluminum shim or just sand down the standoffs. If you do the latter, do it slowly, measuring the thickness each time with a calipber and keep the file parallel to the heatsink. It's unlikely you will mess up.

 

[howardkent]

What is your average cpu load with the four vms running?

The heatsink itself is not designed to be operated horizontally with no airflow, as it will saturate easily and have no convection taking the heat away from its "channels". Running it vertically might help a bit, but i don't expect a massive change, perhaps a few degrees.

What I'm more concerned at this point is the SSD, if you're running the cpu (and board) at 60-65°C I can imagine the SSD is doing 70-75°C at least in that tiny case. Install hddtemp, and try probing your ssd (use lsblk to get it's name) for it's running temperature.

If you have space between the cpu and heatsink, I would instead of using a pad between the spreader and the heatsink itself, use a thin copper / aluminum shim or just sand down the standoffs. If you do the latter, do it slowly, measuring the thickness each time with a calipber and keep the file parallel to the heatsink. It's unlikely you will mess up.

The infatuation with the CPU/SSD temps is misguided. If they are within thermal limits then there's nothing to worry about. Just because someone can diffuse their heat more effectively doesn't mean their machine isn't generating the same heat clock for clock as yours. Yours just runs a little hotter due to the thermals of your setup. Again if it's within limits, chalk it up as interesting but not that important, you're not overclocking.

 

[s_p]

What is your average cpu load with the four vms running?

The heatsink itself is not designed to be operated horizontally with no airflow, as it will saturate easily and have no convection taking the heat away from its "channels". Running it vertically might help a bit, but i don't expect a massive change, perhaps a few degrees.

What I'm more concerned at this point is the SSD, if you're running the cpu (and board) at 60-65°C I can imagine the SSD is doing 70-75°C at least in that tiny case. Install hddtemp, and try probing your ssd (use lsblk to get it's name) for it's running temperature.

If you have space between the cpu and heatsink, I would instead of using a pad between the spreader and the heatsink itself, use a thin copper / aluminum shim or just sand down the standoffs. If you do the latter, do it slowly, measuring the thickness each time with a calipber and keep the file parallel to the heatsink. It's unlikely you will mess up.

this is the hourly average, but the daily average is the same.

in reality the nvme is always 2 or 3 degrees greater than the cpu

I have no space between the cpu and the heat sink, I installed the thermal pad for greater thickness and safety

 

[Immortal]

Looks normal for N5105 without a fan. Just put a fan on it and you will be fine, there is no real way to run this cpu with like 100% load for long time and not hit 90-100C(so its gonnaa throttle) without a fan. People have unrealistic expectations. That Crucial nvme drive is perfectly fine till like 70-75C. So not really a problem here.