Two Fanless Intel Celeron N5105 4x 2.5GbE Options Reviewed

[Patrick Kennedy]

We review two Intel Celeron N5105 fanless firewall and router options with 4x 2.5GbE NICs that arrived as an update to our J4125 units

The post Two Fanless Intel Celeron N5105 4x 2.5GbE Options Reviewed appeared first on ServeTheHome.

Continue reading...

 

[elvisimprsntr]

Quote from the support folks at Protectli: Trusted Firewall Appliances with Firmware Protection

We are working on releasing our 6-port 2.5Gb Vault very soon. It should be introduced by early June, and will be shipping in volume by the end of the month.

 

[PigLover]

Power was not a winning point of this solution. At idle, we saw between 10-12W depending on single versus double SODIMM configurations. Maximum power consumption hit 24.5W even with the Topton configured (512GB/16GB) unit. There is clearly room to go up from there. This is a lot higher on both idle and maximum power consumption than the Celeron J4125 that we looked at previously.

For me this is the most interesting observation from this review.

Any thoughts on what is driving the higher power consumption? The N5105 should draw less power at idle than the J4125 due to its 10nm lithography. The RAM and NICs and almost everything else is the same on both boards.

The only material difference in the nvme vs SATA disk. Surely that couldn't drive a 100% increase in idle power (~10W vs ~5W nominal).

Very surprising. And a bit disappointing.

 

[glow]

It seems almost everything draws more power at idle nowadays. From my existing and ongoing testing, I know that my 3400G was just more efficient than my 3600 at being my general homelab server. My server is not 100% idle, but is not being pegged at 100% on any core, either. Closer to 1-2% on each thread. Just enough to wake the 3400g and 3600 out of any deep idle state.

The difference is ~17W at low load (server "idle"). If I loaded up a heavy, single threaded task both of them come out about even.

Recently, Asrock finally released a new bios supporting the 5000G series on my server's AB350 ITX board. I was looking at the 5600g to replace my 3600, and it did not seem to be much of an improvement at idle. Absolute idle figures seemed to heavily favor the 3400g, though many 3400g tests were done on 300/400 series chipsets and later CPUs were often tested on the X570 (hungry chipset that required active cooling for many boards). Nothing I can do about "low load" figures without the 5600g in person.

For me, it was still a ~17W tested difference at the wall, in favor of the 3400g vs the 3600. Disappointed, I started looking around at Intel solutions. In my searching, from the 10th gen onwards, Intel also started creeping upwards in idle power draw.

In the end, I just stuck with my old 3400g, and am considering selling my 3600, along with consolidating my Wyze 5070 router (~20W on its own, though it hosts an X520 DA2, which is almost 7W by itself) into the homelab server. Right now, the Wyse 5070 and the 3400g server are taking a total of 44W at the wall. With the 3600 in place of the 3400g, it was around 60W for both.

 

[PigLover]

I'm actually wondering if Topton and Kingnovy disabled the low P-states in the BIOS. Or even more ominously, if Intel has yet another P-state bug in the Jasper Lake lineup and has disabled them on the silicon (like the bugs in the 2011v2 server lineup that caused servers to "hang" randomly when idle).

 

[Stephan]

Smells like package state C2 only, instead of C6 or lower. Usual culprits are NVME SSDs which can't sleep deep enough and thus keep the PCIe link and the package alive all the time, LAN not plugged in or not deactivated, monitor resolution wrong, C-settings not AUTO but something else in BIOS, etc. pp. On desktop Intel, its sometimes the choice of PCIe slot for a card which enables deep Cx states. Intel i915 graphics also full of pitfalls which prevent deep sleep. Second monitor = C2 only. Realtek LAN = C3 max. On my laptop I can enable DIPM on SATA because the SSD is SATA, saves another few 100 milliwatts. On a desktop with spinning rust, if I enable it, ZFS will complain violently from errors. So bunch of options, but I doubt Patrick has any remaining will to live after surely 3-4 weeks trial and error just for one platform to get it to sleep better. Just to find out it won't work at all, because BIOS won't allow it whatever you do, because Intel developers found a bug with it you never will hear about.

Edit: After reviewing the pictures again, my money is on that Shi-hit NVME. Maybe he can try a Samsung 850/960 Evo or 950 Pro.

 

[allanxp4]

My unit (Topton n5095) is very unstable, from random user-mode crashes to random reboots based on what you use.

Windows Server and Ubuntu desktop: crashes in a few minutes if you use the intel graphics drivers and is idling or on low load.
Running Proxmox VE/Debian: OpnSense VM randomly crashes with a KVM Error or opnsense just reboots by itself without any warning.

It's weird because it can run prime95 for 12 hours nonstop, but crashes within 5 minutes if you have the gpu driver enabled, so what piglovers said actually makes sense - I'm going to check it later if mine has all C-states enabled, but i don't recall them being disabled on BIOS. But it could also be just a faulty VRM.

There's an option on the bios to disable the GPU (GT) power managment. If you use it, it becames more stable on both Windows or Linux with GPU drivers, but still crashes after a few hours.

Tried different SSDs, different RAM, ran memtest86 for hours, nothing solved it.

Also, windows crashes in a way that shouldn't be very fun for people with epilepsy, flashing random colors mixed with random garbage data.

Either mine is a defective unit or there's something wrong with the silicon/motherboard (it also makes an annoying coil whine).
Let's see if they will just send another motherboard or if they will make me send the whole thing to China again.
If someone has similar issues, consider posting it here or letting me know.

 

[ReturnedSword]

A bit disappointed. The unit (N6005 variant) runs rather hot, definitely unpleasant to the touch, which eventually builds up heat on the reverse side of the motherboard. This causes the NVMe drive to overheat then throttle. Granted I’m using a Samsung PM981 which are known to run a bit hotter than other drives.

For some reason CPU-Z and HWiNFO reports my N6005 as an ES CPU. I haven’t been able to find any CPUID or Stepping lists for Jasper Lake so I can’t verify if it is in fact an ES CPU or not. I also experienced a crash in Windows 10 while running Prime95, which I haven’t been able to replicate yet.

The driver situation also isn’t great in Windows 10. I’m not sure about Windows 11 as I haven’t tried yet. In Windows 10, I had to manually hunt down about a dozen drivers that weren’t included on Intel’s support pages. In fact Intel doesn’t even post the Jasper Lake chipset driver package at all.

 

[allanxp4]

Yeah, these are definitely pre-production level stuff, had similar issues and took some time to find the correct drivers.

I don't know how good the j4125 are, but definitely not recommending the n5xxx/6xxx units for now.

A bit disappointed. The unit (N6005 variant) runs rather hot, definitely unpleasant to the touch, which eventually builds up heat on the reverse side of the motherboard. This causes the NVMe drive to overheat then throttle. Granted I’m using a Samsung PM981 which are known to run a bit hotter than other drives.

For some reason CPU-Z and HWiNFO reports my N6005 as an ES CPU. I haven’t been able to find any CPUID or Stepping lists for Jasper Lake so I can’t verify if it is in fact an ES CPU or not. I also experienced a crash in Windows 10 while running Prime95, which I haven’t been able to replicate yet.

The driver situation also isn’t great in Windows 10. I’m not sure about Windows 11 as I haven’t tried yet. In Windows 10, I had to manually hunt down about a dozen drivers that weren’t included on Intel’s support pages. In fact Intel doesn’t even post the Jasper Lake chipset driver package at all.

 

[Stephan]

Thanks for the update on this. First thing I do when a system crashes while running Prime95, I send it back. First thing I do when I see CPU has ES bit set and I didn't order that, I send it back. Because there will never be any microcode updates for that ES CPU and who knows what bugs the stepping has. Cooking the m.2 on top of that just creams my corn.

If it HAS to be stable, imho there is just no way around tier 1 motherboard shops like OEM (Dell, HPE, Lenovo), Supermicro, Asrock Rack, maybe Tyan. The only small x86 embedded shop I ever saw that produces stable platforms is PCengines from Switzerland. Their Alix line with Geode CPU kept on running well into its teenage years. I expect nothing less from the more recent APU2 line.

Here is such an example "stable" system, a fanless APU4D4 (1 GHz AMD Jaguar, 4 cores, 4 GB ECC RAM, 4x Gigabit Intel i211), 512 GB mSATA SSD, Dell aka Sierra Wireless AirPrime MC7455 4G card on mPCI to mini-PCIe adapter (middle), Atheros 802.11n AR9590 Wifi card (left), Bluetooth dongle (not visible, on back USB), GPS antenna (not visible to the right):


To run nicely, I patched the Linux kernel and assorted software, i.e. kernel 5.4 +wireguard +ath9k no regulatory limits, no power limits, no antenna reduction (this Atheros card was once integrated in a system with big antennas, tx power is therefore more limited than with the usual USA reg domain cards), no 20 MHz fallback with wifi neighbors, +gpio_keys_polled fix +pcengines-apuv2 fix to allow ACPI enumeration for LED control, but also make front button scan code work (reboot). Also managed to make gpsd for chrony time service precise enough without PPS just GPS NMEA, so that it is now usable as a sub-3-millisecond std-dev stratum 0 time source. BIOS is open source CoreBoot and schematics are available freely. Used to be orderable for years, now with chip shortage remains to be seen.

It's what I compare all those chinese fanless router boxes to. Not one has come close.

 

[ReturnedSword]

An APU2 is a bit too weak for my purposes. I have deployed a few before, and agree they are certainly great.

I'm getting an increasingly bad feeling about the Chinese Jasper Lake units. I'll do some more testing, but I doubt I'll change my mind unless I had missed something along the way.

 

[Zinc64]

A bit disappointed. The unit (N6005 variant) runs rather hot, definitely unpleasant to the touch, which eventually builds up heat on the reverse side of the motherboard. This causes the NVMe drive to overheat then throttle. Granted I’m using a Samsung PM981 which are known to run a bit hotter than other drives.

For some reason CPU-Z and HWiNFO reports my N6005 as an ES CPU. I haven’t been able to find any CPUID or Stepping lists for Jasper Lake so I can’t verify if it is in fact an ES CPU or not. I also experienced a crash in Windows 10 while running Prime95, which I haven’t been able to replicate yet.

The driver situation also isn’t great in Windows 10. I’m not sure about Windows 11 as I haven’t tried yet. In Windows 10, I had to manually hunt down about a dozen drivers that weren’t included on Intel’s support pages. In fact Intel doesn’t even post the Jasper Lake chipset driver package at all.

Got the TopTon N6005 myself last week.
Just checked, and mine is showing as an ES CPU as well.
Does this mean I should return it?

My NVMe temps were hitting 63C during Windows Updates.
Put a cheap heatsink on it and it does normal temps now, 29C @ idle.

It's been completely stable under Windows 10 so far, but you have to find drivers on your own.

 

[elvisimprsntr]

Does not pay to be an early adopter.
This is why I will wait for Protectli: Trusted Firewall Appliances with Firmware Protection to complete their own stress testing before making them available for sale.

 

[mj1003]

My unit (Topton n5095) is very unstable, from random user-mode crashes to random reboots based on what you use.

Windows Server and Ubuntu desktop: crashes in a few minutes if you use the intel graphics drivers and is idling or on low load.
Running Proxmox VE/Debian: OpnSense VM randomly crashes with a KVM Error or opnsense just reboots by itself without any warning.

It's weird because it can run prime95 for 12 hours nonstop, but crashes within 5 minutes if you have the gpu driver enabled, so what piglovers said actually makes sense - I'm going to check it later if mine has all C-states enabled, but i don't recall them being disabled on BIOS. But it could also be just a faulty VRM.

There's an option on the bios to disable the GPU (GT) power managment. If you use it, it becames more stable on both Windows or Linux with GPU drivers, but still crashes after a few hours.

Tried different SSDs, different RAM, ran memtest86 for hours, nothing solved it.

Also, windows crashes in a way that shouldn't be very fun for people with epilepsy, flashing random colors mixed with random garbage data.

Either mine is a defective unit or there's something wrong with the silicon/motherboard (it also makes an annoying coil whine).
Let's see if they will just send another motherboard or if they will make me send the whole thing to China again.
If someone has similar issues, consider posting it here or letting me know.

Hey- did you ever figure out what was causing the random crashes? I'm having the same issue on multiple Topton units I ordered, both in N6005 and N5105. Both have Samsung 980 Pro NVME and plenty of 16 to 32GB memory.

Edit: I should mentioned, during Proxmox bootup, I get the following errors, where it hangs for a minute or two then continues booting Proxmox.

 

[ReturnedSword]

Got the TopTon N6005 myself last week.
Just checked, and mine is showing as an ES CPU as well.
Does this mean I should return it?

My NVMe temps were hitting 63C during Windows Updates.
Put a cheap heatsink on it and it does normal temps now, 29C @ idle.

It's been completely stable under Windows 10 so far, but you have to find drivers on your own.

Sorry I didn’t see your reply here. YES. Immediately initiate a dispute and ask for a refund or return. Don’t be like me who tried to work with Topton, and only found out later they were killing time until the dispute window ran out. If the window runs out you’ll have to escalate it to AliExpress, or your payment method.

 

[elvisimprsntr]

Our mutual friends at Protectli: Trusted Firewall Appliances with Firmware Protection are getting closer to releasing their 6-port 2.5G vault.

Looks like a much larger heat sink for those power hungry Intel chips, than the sub-$300 stuff direct from China.

Fortunately we are getting closer to releasing our 2.5Gb/s firewall.

Protectli Vault Pro VP4630 - 6 Port Intel® i3 - Protectli US Store

VP4630: 6-Port firewall micro appliance/mini PC, built to run open-source software. Includes US-based support & 30-day money back guarantee.

protectli.com

 

[Patrick]

Wow! I have the 6x 2.5GbE Core i7-1165G7 unit coming in and it was $529 shipped and that has a MUCH better processor. I may see if I can get a Protectli just to compare.

 

[elvisimprsntr]

Wow! I have the 6x 2.5GbE Core i7-1165G7 unit coming in and it was $529 shipped and that has a MUCH better processor. I may see if I can get a Protectli just to compare.

But if it’s thermally constrained, is it really “better?”

 

[cesmith9999]

I have the 6x 2.5GbE Core i7-1165G7 unit coming in

where did you order this from?

Chris

 

[ReturnedSword]

where did you order this from?

Chris

On AliExpress. Both Topton and KingNovy are selling it.