Preface

Today I happened to see someone selling a laptop for $100. The specs were a Celeron N5095, 24GB of DDR4 RAM, a 512GB SSD (M.2 SATA, not NVMe), and it even came with a legitimate Windows Pro license.

RAM prices have gone up a lot lately, so 24GB at this price point is actually pretty decent. And I had always assumed the DDR4 in machines like this would be soldered on, but after opening it up I found that it’s replaceable. That alone made it feel especially worth it to me, since upgrading or adding more later would be easy.

I’ve used the N5095 CPU before. I previously had an M8S mini PC, also with an N5095, that I bought for about 1000 RMB and upgraded with 16GB of RAM. It had been running pretty well the whole time. Back then prices hadn’t gone up yet, so being able to get another full machine now for $100 with the same CPU and even more RAM felt like a pretty good deal.

That said, the two machines are positioned differently. The M8S was all about being tiny—it was marketed at the time as the smallest mini PC in the world, basically palm-sized, and it also had dual NICs. Performance was never its strong point, and now there are better options like the N100 anyway. This laptop, on the other hand, wins on having a lot of RAM, upgradeable memory, and a low price.

I bought it specifically to mess around with as a server. These days there are more and more personal agents and all kinds of AI-built apps, and they’re getting faster too. I’m definitely not going to complain about having more servers. So I barely even used Windows after booting it up—I went straight to installing PVE.

This post is split into two parts:

1. Transition phase: how I’m using it as a server right now, including how I installed PVE without an Ethernet port and what tuning I did afterward.
2. Later on: taking the motherboard out and putting it into an acrylic case to use it as a mini PC. I haven’t done this part yet, so for now I’m just writing down the plan and I’ll update it once it’s done.

1. Transition phase: keep the lid closed and use it as a server for now

1. First, a quick look at the BIOS

The first thing I did after getting it was go into the BIOS, check the basic parameters, and do a bit of tuning while I was there. It was a brand-new machine—I wasn’t even planning to peel off the protective film.

BIOS home page, Celeron N5095, 24576 MB memory

After going through it, I found that it was already in a pretty optimal state from the factory: VT (CPU virtualization) was enabled, and the performance-related options were set fairly aggressively too. It basically matched my preferences already, so there wasn’t much I needed to change.

2. I couldn’t resist opening it up first

Originally I planned to wait until after installing the system, but I really wanted to see what the motherboard looked like first. If I’m going to convert it into a mini PC later, I need to know whether the internal layout makes sense for that.

There’s a pretty handy little cover on the back—open it and you can immediately see the RAM and SSD. Since I had already opened that, I just went ahead and removed all the screws. Taking off the back panel took a bit of effort because the clips were tight, but I eventually got it off.

Prying off the back cover; the clips were pretty tight

Inside after opening it up: the large section on the left is the battery; the fan and motherboard are at the bottom right; both the RAM and SSD are on the motherboard

Once it was open, I found that the inside was actually very empty. Most of the space was taken up by the battery, and the motherboard only occupied a small section in the lower right. That was exactly what I wanted—the smaller the motherboard, the smaller the acrylic case can be later.

Two DDR4-2666 sticks, one 8GB and one 16GB, both socketed

The memory was indeed socketed: one 8GB stick plus one 16GB stick, both DDR4-2666. The SSD was a 512GB M.2 2280 SATA drive.

After taking a look, I put everything back together and started installing PVE.

3. No Ethernet port — how do you install PVE?

This machine is cheap, and it’s really cheap for a reason: there’s no Ethernet port. Never mind 10GbE or even Gigabit—it doesn’t even give you one wired port at all, and there aren’t many other I/O ports either. Later I might use a USB-to-RJ45 adapter, but right now all I had on hand was Wi-Fi.

The problem is that the PVE installer doesn’t support Wi-Fi. During installation, it can’t connect to wireless and there’s nowhere to enter a Wi-Fi password, so once the installation finishes you end up with a machine that has no network. The proper way is to build a custom install image yourself, or find another way to configure networking afterward, but I really didn’t want to go through all that hassle. I had already made a PVE boot USB drive when setting up another machine before, so I wanted to just use it directly.

So I used a slightly hacky trick instead: I used an Android tablet as its network card.

Connect the tablet to the laptop with a USB cable, then set the USB mode to "USB tethering"

I connected the tablet to the laptop with a USB cable and enabled USB tethering. On the laptop side, that shows up as an extra “wired network card.” The PVE installer can detect it, so the installation uses that interface, which naturally means there’s no need to enter any Wi-Fi password.

After installation, the laptop and the tablet were on the same subnet, so the tablet could directly open the PVE web UI. From the terminal in the web interface, I configured Wi-Fi (static IP, Wi-Fi SSID and password). After a reboot, it showed up on my home network normally, and I could unplug the tablet.

4. Tuning after installation

After that I did a round of tuning, mostly based on the setup I already use on my other PVE mini PC so the two machines stay consistent:

1. Do not sleep when the lid is closed: right now it’s being used as a server with the lid shut, so I changed the lid-close action to do nothing and disabled all sleep/hibernate functions.
2. Switch to the no-subscription software repository: the default PVE enterprise repo requires a subscription, so I switched it to the free community repo and updated the system to the latest version.
3. Disable services that a standalone machine doesn’t need: PVE high availability (HA) and cluster communication services are only useful when you have multiple machines in a cluster, so on a single machine I just turned them off.
4. Memory and logs: set swappiness to 10 to reduce swap usage as much as possible; set a 200MB cap for system logs so they don’t just keep growing forever.
5. Auto turn off the screen: the console goes black after 60 seconds of inactivity.
6. Add it to the tailnet and monitoring: installed Tailscale so I can still connect back to it while away from home; I also added checks in monitoring for whether it’s online and whether the web UI is reachable, so I’ll get alerts if it drops offline.

I also checked a few other things along the way: CPU VT-d (needed for device passthrough) was working; the battery doesn’t support setting a charge limit, which is a bit of a shame, but since I plan to remove the battery later anyway, it’s not a big issue.

At this point, it can just sit there quietly with the lid closed and act as a server. In actual use, it’s fairly quiet and low-power. Whether for NAS, a soft router, or running other services, it’s all pretty suitable.

2. Later on: take the motherboard out and use it as a mini PC

As a server, I don’t really need the screen, keyboard, touchpad, or battery. It just needs to sit there quietly and do its job. So the later plan is to take the motherboard out on its own and install it into an acrylic case to use it as a mini PC.

Before doing that, there are a few things I need to confirm:

1. Whether the BIOS has “power on after power loss.” A mini PC needs to be able to boot by itself when power comes back after an outage; otherwise someone has to physically press the power button every time. Laptop BIOSes don’t always have this option, so I need to check first.
2. Whether it can boot from the adapter alone after removing the battery. Some laptops simply won’t power on without a battery, so this needs to be tested first.
3. Cooling. Right now it’s inside a laptop chassis with an airflow path that was designed for it. Once it goes into an acrylic case, the thermal conditions will change, so I’ll need to run a full-load test and check temperatures.
4. Networking. At that point I plan to use USB-to-RJ45 and switch it to wired networking, which should be more stable than Wi-Fi.

I’ll come back and update this part once it’s done.

Closing thoughts

I’m not really going to comment on the laptop’s build quality itself—that’s not the part I care about. But a new machine still does feel nice, even though I took it apart and reinstalled the OS before even peeling off the protective film.

For $100, with 24GB of upgradeable RAM and an N5095 CPU that I’ve used before and know is good enough, I think it’s a really worthwhile little server for tinkering.

All articles in this blog, unless otherwise stated, are licensed under @Oreoft . Please indicate the source when reprinting!