Has recommended Mini PC hardware stagnated?

[natevw]

I've been using a Minisforum UN100C which is an Intel N100 w/16GB of DDR5 to run Frigate. That had been mostly working but was running into two issues:

• kept hitting the 16GB and would find cameras offline and slow to browse the UI because processes kept getting OOM killed and/or swapping. might be in part due to my configuration (probably need to move my detect feeds to lower-res instead of the same feed as recording… and also bump a small HassOS install I co-located over to another machine…)
• but the main problem is the fan in this is going out and the time between re-oiling it and it getting noisy and overheating keeps getting shorter and shorter.

I did grab a supposed replacement fan before the country I live in kind of self-sanctioned itself wrt to getting stuff from overseas, but that one didn't work out and I don't want to end up paying huge fees now for another try. But each time this comes up I figure this machine was kind of old/cheap already when I got it and that was two+ years ago. So surely I could get something with more RAM and even better processor? But that's where I keep running stuck…

I'm not a huge PC enthusiast and most of the sites I used to reference have been crowded out by AI slop which I don't know what to believe. So again with this starting point of a 16GB N100 that I purchased for $160 back in 2023 September, I search for Mini PC now and I see:

• nearly everything is still 16GB max. still not seeing very many 32GB options!
• there's kind of some other CPU options besides the N100 but it's really unclear what's what:
  • there's the N95 and N97 which sound "less impressive" (97 < 100) but supposedly the N97 has faster speeds (at higher TDP…)?
  • there's also the N150 which sounds better (150 > 100) but again supposedly is an older/lesser CPU and only advantage is cost and perhaps lower TDP? [but some are selling N150 as "upgraded N100" so maybe the site that turns up on results is just auto-generated by a system prone to making stuff up in a way that just sounds legit?]
  • and then what might be somewhat new is I'm seeing more Ryzen-based ones which are often advertised as "better than N150" (in what way? but not better than the N100 that I already sorta have if I can fix the fan?). still generally only 16GB of RAM…

So basically when I got to find a Mini PC now they seem about the same as the one I bought two years ago, only now with inflation/tariffs so that the price isn't even really cheaper. Maybe I'm just old school and I guess I've been reading about the end of Moore's law for a while now but I thought at least we were getting more cores and more memory and better codec handling and whatnot…. Just weird to see that the N150 processor in everything now is supposedly an older architecture than the N100 that was cheap a couple years back.

Anyway sorry this got long. I guess I would boil this down to questions:

• am I right that basically the cheaper Mini PCs available now aren't much/any better than the ones already out several years ago?
• is there an easy line-up between the N95 vs. N97 vs. N100 vs. N150 especially re. Frigate performance specifically?
• how good are the various Ryzen(s) as well again specifically re. Frigate performance?

[inkeep-ai-assistant[bot]]

---

Based on the available documentation, the Beelink EQ13 with an Intel N100 CPU is currently recommended as the preferred mini PC for Frigate(1). The documentation notes it "Easily handles several 1080p cameras"(1) and has dual gigabit NICs for easy isolated camera networks(1).

Regarding your specific questions:

Current Hardware Recommendations

The documentation does show the N100 as the primary recommended CPU for mini PCs running Frigate(1). For inference performance, the N100 achieves approximately 15ms for small-320 models and around 25ms for 320x320 detection(1). The documentation also notes that the N100 "Can only run one detector instance"(1).

The Intel N150 is also listed in the hardware recommendations with slightly better performance: approximately 15ms inference time, with t-320 at 16ms and s-320 at 24ms(1). This suggests the N150 does offer some improvement over the N100 for Frigate workloads.

Memory Concerns

Your 16GB memory issue is a real concern. The documentation explains that shared memory sizing is critical for Frigate(2). For multiple high-resolution cameras, you need to calculate the minimum shm size using the formula provided, and with 1080p+ cameras, memory requirements can add up quickly(2).

AMD Ryzen Options

For AMD-based systems, Frigate supports ROCm for object detection on AMD discrete GPUs(3). The AMD 780M achieves around 14ms for 320x320 detection and the AMD 8700G achieves approximately 20ms for 320x320 and 40ms for 640x640(1). These are competitive with Intel options for Frigate workloads.

Hardware Acceleration

One important note: the documentation shows that Intel CPUs from gen8-gen12 can use either preset-vaapi or preset-intel-qsv-* for hardware acceleration(4). For the N100 specifically, users have reported success with preset-vaapi configuration(5).

Bottom line: The mini PC market for Frigate hasn't dramatically improved in the past 2 years. The N100 remains a solid choice, though the N150 shows marginal improvements. Your memory constraint (16GB) is likely more limiting than the CPU for your use case with multiple cameras.

📚 Sources:

1. Recommended hardware
2. Installation
3. Supported Hardware
4. Video Decoding
5. [HW Accel Support]: Frigate on intel N100

---

Have another question? Just tag @inkeep.

[natevw]

lol, speaking of AI slop….

Yes I had already read that Recommended Hardware page. Afaict the Beelink EQ13 if it were in stock [the documentation link brings up some other model instead] would be an N100 with 16G DDR4?

So this the latest and greatest for running Frigate these days is a computer with the same processor and the same amount but slower RAM than I bought as basically some bargain bin closeout a more than two years ago?!

[NickM-27]

The current docs recommend a basic starting point. There are plenty of options out there with more RAM as the RAM is customizable. For example here is a current platform that would work very well https://www.amazon.com/MINISFORUM-Pro-125H-Barebone-Computer-HDMI2-1/dp/B0FH21FSZM?th=1

[natevw]

Thanks, though to be clear still looking to spend closer to $150 than $550 though (and figure out my inefficiencies). Maybe to rephrase my question in terms of comparison:

• would an N150 machine be better or worse for Frigate than an N100 machine?
  • (is the CPU difference itself the source of the "known compatibility issues" of the EQ14?) edit: found Beelink EQ14 support #18850 (reply in thread)
• would a N95/N97 machine be better or worse?
• how about a Ryzen 5 3500U (which is claimed "better than N150"), again for Frigate workloads specifically?

[NickM-27]

In general the N# means higher is better, so N150 would be best.

AMD platform has significantly worse performance for machine learning so Intel is definitely recommended.