Topton Jasper Lake Quad i225V Mini PC Report

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

prdtabim

Active Member
Jan 29, 2022
171
66
28
I ordered this v2 unit [1] (N5105/no ram/no storage)on August 6th and it was delivered on the 15th to Germany without additional taxes/fees. They currently have a discount of 2EUR for every 25EUR spent, so only 157EUR for the plain N5105 unit.

So far this unit seems nicely built. I removed the huge amount of thermal compound and only left a decent amount on the 2 DIEs. They sit perfectly on the aluminium spacer with a slight pressure due to the fact that the mounting studs are a bit shorter than the spacer. Case gets warm/hot'ish' but not too bad for a passively cooled unit (some PoE switches get way hotter), CPU temp sits at ~55° and max I saw was 71° when switching to the BIOS direcly after the kernel relinking at the end of the OpenBSD setup. There were lower temps posted here, but this thing has a Tjmax of 105° so I couldn't care less about those temperatures...


I'm only having one problem:
Is anyone using the serial port on one of these boards? I can get the console output, but no inputs are recognized by the unit; neither at POST/BIOS, nor by the booted OS (OpenBSD). So I currently have to connect a USB keyboard to the unit...
I'm connecting via screen using a USB/Serial adapter and Cisco serial cable; both are working perfectly fine with other equipment (cisco and non-cisco /w rj45 serial), so I don't think the cables are the problem but the BIOS settings.

Im currently using those default settings:
View attachment 24106

(Terminal Type VT100+ gives me colors, but in BIOS the selected item is the same color as the background, so I switched back to VT100 for the screenshots)

Already tried enabling "recorder mode" (disabled should be right, but wouldn't be the first BIOS with inverted/wrong logic on some options...), various options of the "putty keypad" option (no clue what putty has to do with the console redirection...) and various non-standard combinations of data/parity/stop bits and flow control, but it seems this thing just completely ignores any input via serial console. Even OpenBSD with console on com0 doesn't get any input, so it has to be the firmware that discards it.
(yes, the cable/adapters are working. I'm using them nearly every week and I just double-checked by connecting to the com-ports on some cisco and juniper gear here)



[1] 157.19US $ 45% OFF|Solid-built Fanless Mini Pc 2.5g Soft Router I7 1165g7 I5 1135g7 Celeron N5105 4 Intel I225 2500m Lan Dp Hd-mi Pfsense Firewall - Barebone & Mini Pc - AliExpress
A bunch of questions / doubts ...
- Is the console port a RS232 ( 12V ) or a TTL ( 5V ) console ? I have a bluetech mobo that include a console port but that are no specs.
- In the settings of the terminal you try using "flow control none" ? not in bios but in the terminal emulator .

Putty is a terminal emulator ( ansi, vt100, vv220 etc ) that includes serial, telnet and ssh connections.
 

sko

Active Member
Jun 11, 2021
240
128
43
I know what putty is; I just never saw any options specific to a windows program appear in a BIOS...

the port is a standard rs232 over rj45 i assume - as said I get the full console output, only keystrokes are not registered by the unit.
Usually this is because of mismatching flow control, but default is disabled and that's whats set at the BIOS and emulator. I also tried explicitly setting -fn although screen doesn't enable it by default and It works OOTB with just "screen /dev/ttyU0 115200" with any other gear - only the topton unit will not register any input...
Also tried with FC enabled on both ends, but as said: no combination of parameters will give me keyboard input.
I'm currently suspecting either a non-standard wiring (RX pin) or a defective port.


EDIT:
Mystery solved - the pinout is different to cisco and most other gear using RJ45 serial ports. Coincidently, pin 6 is also TX on the topton, that's why I got console output, but the RX pin is at pin 5, not 3 (cisco et al).
Found an old Allied Telesyn cable while searching through my cable collection for something with a soldered DB9-connector where I could crimp on an RJ45 connector, but that cable worked 'as is' so I'm using that for now...


TL;DR
The RJ45 serial pinout for those units is NOT identical to (most) network gear!
At least the important pins (RX/TX, GND) seem to correspond with Allied Telesyn:
1661169455295.png
 
Last edited:

Ipse

New Member
Aug 21, 2022
8
4
3
I seemed to have fixed my vm reboots issue. Topton n5105 w/2usb 3 port, vga and hdmi.

Disable IME. Intel Management Engine
Limit C states to C3
Set P1 power limit to: 10000
Set P2 power limit to: 12000

Seems the CPU was spiking temps really high when I had P2 limit Set to 22000. I lowered it to 12000 and disabled IME as well as set limit to C states to C3.

Not sure which fixed it but I'm at 24 hours now and no VM lockups or mystery reboots. Will give it another day before I confirm this fixed issue. I did run a few stress-ng tests and so far so good.

I take PayPal donations at ron@stopspamming.net.
J/K. (I bet I get spammed to hell) I hope I was able to help others. Let me know if this helped you!
Newbie here... I spent the weekend reading through the forum posts and like many others I re-pasted the CPU, made BIOS changes and stabilized the temperature at 40-42C (ambient 26C) .
I installed a bare metal pfSense which works well, tweaked the manual CPU settings there as well - so BIG thanks for all who contributed with their experience and findings, otherwise this thing would be a fire hazzard.

Two questions though for the hive mind:
- running a stress test from pfSense (only 60sec, CPU at 100%) the core temperature jumps almost instantly to 85-90C then immediately after it's over, drps back to 45C. To me it looks like the case/radiator thermal inertia is so big, that it can't cope with the increase in temperature. Should I be worried? Doesn't this also suggest my thermal transfer is not ideal?
- secondly, with these tweaks I haven't seen the cores ever going in Turbo mode - which BTW is enabled in BIOS. I limited PL2 to 15w, is that what's preventing it from even attempting to boost the clock? I expected to jump but not be able to stay at 2.6-2.8 GHz for long....based on YT videos (Mr Guo).

Cheerios!
 
Last edited:
  • Like
Reactions: dazagrt

Covert_monkey

New Member
Jul 5, 2022
12
2
3
Finally pulled the trigger and got one from Changwang 4V i266 version. Has anyone got that yet?
Also, has anyone tried to replace the existing thermal compound with liquid metal ?
 

xShARkx

Member
Jun 12, 2022
78
49
18
Finally pulled the trigger and got one from Changwang 4V i266 version. Has anyone got that yet?
Also, has anyone tried to replace the existing thermal compound with liquid metal ?
No one got the V4 yet, it got release about 10 to 14 days ago. It took topton 8 days to ship mine and its still in china.
 
Last edited:

sko

Active Member
Jun 11, 2021
240
128
43
Another small update:

Running the v2 with one of the already mentioned DSLRKIT 24W PoE splitters works so far.
BUT: the splitter does not negotiate a power level via lldp (despite being advertised as 'active'), so all 802.3at-conformant switches won't provide more than 15.4W by default. You either have to set a static power level or with cisco you can use the 2x-mode ('power inline port 2x-mode').

The actual power consumption reported by the switch ('show power interface gi n/n/n detail'):
Code:
[...]
Actual consumption
Measured at the port: 20.0
Maximum Power drawn by the device since powered on: 22.3
[...]
Those values are with default BIOS settings (after clear cmos) for CPU/power saving and nothing special in terms of frequency scaling / power saving modes enabled in the OS (OpenBSD 7.1).
I've only installed 8GB RAM (crucial CT8G4SFRA32A) and an old Samsung SM951 128G NVMe and only one NIC + the console port are connected.
Maybe the integrated GPU can be completely disabled or at least I saw an option in BIOS to scale down its frequency, which might save a considerable amount of power.

The 22.3W maximum is drawn during the relinking/kernel rebuild after booting, so I don't expect much more load CPU-wise in production; but enabling the other NICs might bring it above the 24W the splitter is able to provide on paper...

edit: each network port accounts for ~0.5W at idle (i.e. only plugged in and "up", but with no actual load)
 
Last edited:

oneplane

Well-Known Member
Jul 23, 2021
845
484
63
For the serial port: if you enable console redirection you need a redirection driver before you can use the serial port in the OS again. If you disable console redirection, the serial port works as normal in the OS, but the BIOS doesn't always show up in text more (only graphical mode most of the time). Also, if you can see the BIOS via the serial port, you already have the correct hardware, so the cable and settings are fine. You'd obviously have to change the settings (but not the cable) if the baud, bits and flow control is configured differently in the OS. If the input also doesn't work in the BIOS, it might simply be a mis-matched TX-RX connection, but if you see output, RX-TX is already good.

Regarding TaoBao: this is the same as all direct-from-china shops, but with even less international focus. Most sellers in China that just want to move product will sell on multiple platforms:

- Aliexpress
- Alibaba
- DealExtreme
- TaoBao
- Shein?
- JD?

There are a bunch more, a lot of them are owned by the same company, but there is some variation. Aliexpress seems to be the most international-oriented one (regarding their marketplace services). You will not get a (cuddly?) retail experience, mainly because it doesn't matter at their volume and doesn't matter considering their local (physical) sales. Even if an entire county were to say "the shop is so bad, we'll never order there again" it won't make a real dent. Keep this in mind, compensate accordingly and it can still be a useful source. Oh, and like most shops, they can be somewhat vulnerable to social pressure but not everything we understand as social pressure is seen the same way so that's not really a guarantee or anything like it. It's a different world.
 
Last edited:
  • Like
Reactions: n00b and abq

dazagrt

Active Member
Mar 1, 2021
195
97
28
- running a stress test from pfSense (only 60sec, CPU at 100%) the core temperature jumps almost instantly to 85-90C then immediately after it's over, drps back to 45C. To me it looks like the case/radiator thermal inertia is so big, that it can't cope with the increase in temperature. Should I be worried? Doesn't this also suggest my thermal transfer is not ideal?
This needs to be a balance of thermal mass and rate of dissipation, doesn't it? If you can't increase the dissipation rate by using forced air speed across the surface area, then you are only left with increasing the surface area; which is more fins or spreading the heat out more efficiently via the contact material or heat pipes. It's something I had planned to have a look at when the new unit comes.
 
  • Like
Reactions: ReturnedSword

skimikes

Member
Jun 27, 2022
83
79
18
It was guests hanging. Lots of info here: VM freezes irregularly

pfSense was more stable than Linux variants but still inevitably froze. I used the recommended Proxmox settings recommended by pfSense (BIOS, not UEFI): pfSense® software Configuration Recipes — Virtualizing with Proxmox® VE | pfSense Documentation

Tried all kinds of settings but none stopped the freezes and it's a common issue amongst numerous people as demonstrated in the Proxmox forums. I think the devs have narrowed it down to a linux kernel issue but I've since moved on to ESXi since I needed a stable platform so I can't undertake the troubleshooting requested by Proxmox devs (basically running the mainline kernel).

There's detailed logs I've provided in that thread. Often nothing was logged to screen or file so I had to setup remote logging via UDP - netconsole for linux based VMs and remote syslog for FreeBSD/pfSense. I also opened bugzilla reports on Proxmox and kernel.org detailing these issues and logs attached.

Best to read the Proxmox thread I've linked to for more information.

kernel.org bugzilla report: 216349 – Kernel panic in a Ubuntu VM running on Proxmox
Proxmox bugzilla report: 4188 – VMs freeze on Intel N5105 CPU
Did you try with C-states disabled in the BIOS of the N5105 unit so that the cores aren't allowed to go to sleep? I have an N6005 unit sitting around. I had issues with it in the beginning and one of the things I changed was disabling C-states. After that, I no longer experienced pfsense hangs in Proxmox. I'll re-enable C-states and see if I can get it to hang.

Have your guests be stable on ESXi?
 

oneplane

Well-Known Member
Jul 23, 2021
845
484
63
I suspect the configuration of the management engine (which in CSME versions does some power and thermal management as well) and the firmware-driven core configuration is just completely "free for all, do what you want" causing problems with both wakeups and package limits.

If we can figure out the configuration of a different vendor's product that has a bit more of a 'retail' status, we can probably create better profiles for the Jasper Lake hardware from Aliexpress as well. Maybe we'll see this SoC used in some enterprise IoT gateway device or basic NAS soon.
 

jarodmerle

New Member
Aug 19, 2022
4
2
3
Edmond, OK
Did you try with C-states disabled in the BIOS of the N5105 unit so that the cores aren't allowed to go to sleep? I have an N6005 unit sitting around. I had issues with it in the beginning and one of the things I changed was disabling C-states. After that, I no longer experienced pfsense hangs in Proxmox. I'll re-enable C-states and see if I can get it to hang.

Have your guests be stable on ESXi?
I've had similar issues to those @gyrex reported with my HUNSN NRJ02 unit with the N5105 SoC and found last week that I am not even able to enable C-States and boot successfully (into Proxmox VE). So for me, even with C-States disabled I have lackluster stability with VMs failing roughly every 5-7 days on average. I decided to give up on it and will be selling mine and replacing it with an older, but higher-end micro PC..
 

TH_user

New Member
Aug 22, 2022
11
5
3
This is my first router appliance build. Intent is to produce a baremetal Pfsense box.


Purchased the box above and I'm looking for M.2 and DDR4 recommendations. Patrick's review indicated the need for low power to stay away from some of the transient issues being experienced. So I've been looking for DDR4 with 1.2 v. Consider 2Rx8 for 16 Gb.

I considered the Samsung 970 EVO Plus but others have reported issues and I'm not sure with only 2 lanes a more budget friendly M.2 would meet the need. Have also looked at KingSpec which STH found underwhelming and WD Black. Considering 250 Gb.

Finally can anyone confirm if the Noctua A4x10 fits this case and if the board has a pin out for the fan? The marketing shows a solid bottom plate and I don't know how much air would be moved. I'll probably wait on this until I get my hands on the appliance to confirm.

As I understand my stated use case I'm looking for solid reliability more than speed of execution. I'm in Thailand so anything on AliExpress is available to me but some other supply chains may be difficult.
 

DanielN

New Member
Jul 6, 2022
6
1
3
Interesting. Have you ever had freezes? What OSs are you running on your VMs?

Data like this is useful in narrowing down the issue with the N5105.
FreeBSD based appliances and Debian right now, had no issues with Ubuntu but switched to Debian for my needs. No freezes. Are you sure your N5105 CPU isn't a ES or QS chip? I think people are probably using the N5105 fine and aren't making noise because they aren't having issues. I have N5105 version 3/slim incoming and I will let you know if I experience the same issues.
 
  • Like
Reactions: stevenla

ReturnedSword

Active Member
Jun 15, 2018
526
235
43
Santa Monica, CA
This needs to be a balance of thermal mass and rate of dissipation, doesn't it? If you can't increase the dissipation rate by using forced air speed across the surface area, then you are only left with increasing the surface area; which is more fins or spreading the heat out more efficiently via the contact material or heat pipes. It's something I had planned to have a look at when the new unit comes.
Exactly this.

Take for example, a high end socketed passive heatsink: Noctua NH-P1. The P1 is the absolute best socketed passive heatsink on the market today, yet it cannot handle 65W CPUs once the CPU enters turbo state for an extended period of time, much less higher turbo state levels. In the case of the P1, this can be mostly mitigated by installing an optional fan, but then that would make the heatsink “not fanless” anymore, in which case other Noctua HSF are better suited since they were designed with fans as a prerequisite.

For the various chassis variants discussed in this thread, the main issue is that the chassis is not designed well. There is I could say minimal effort in taking account of thermal mass vs thermal dissipation. The chassis itself actually has a decent amount of thermal mass. The issue is that the heat once “sinked” is not dissipated well/fast enough due to the fin arrangement and how the heat is transferred to the chassis initially (through the copper/aluminum plug). It seems to me that whoever designed the chassis only has a very superficial understanding of thermaldynamics, which is a bit annoying as I had a discussion with my niece who is in secondary school and taking secondary school level physics, and she was able to immediately identify the main issues.

That being said, the SoC itself can handle quite a bit of heat. The storage drive and RAM, on the other hand, may not. Most consumer storage drives top out at 85 C, and running at that temperature for extended periods of time is not great. RAM can start having errors as well if it was not designed for high operating temperatures. Depending on who you ask, the main board itself may also not be designed for extended usage in elevated temperatures. The chassis variants that can install a 15x10mm fan can help with this, though that’s not an ideal solution.
 
  • Like
Reactions: dazagrt