I'm pretty impressed by Intel's N100s, and I think they have more recent N200s and N300s but not as common as the N100/N95s/N97s.
The killer feature for a lot of home server and pet projects is hardware accelerated encode/decode; with modern codec support. That is a complete deal breaker vs the RPI 5.
Also, don't forget upgradable RAM, and storage without needing a HAT or anything like that.
N100 is probably the best CPU Intel has now, completely demolishing the low end. I have an N100 firewall, an N100 ASRock mini ITX board for NAS, an N305 CWWK NAS/home server board and they are all unbelievably good compared to older Atoms or RPis. Only 7840U is significantly better but also consuming much more energy.
Ditto to this. I have fallen in love with the N series chips. I use N100 mini PCs all over the factory floor as client machines for workers that manage job control and I am typing this on an N200 power PC that I have slapped a 48GB DDR5 SODIMM in. Working great.
Btw, ALL N95/N100/N200/N305 systems I have used have worked fine with 32GB DDR4 SODIMMs or 48GB DDR5 SODIMMs. Even tho Intel claims 16GB max for DDR4 and DDR5.
Also, the N305 chip is great. Basically two N200 slapped together. They didnt double the L3 tho which is annoying. But the 8 cores is fantastic.
I am currently running a 4 box Proxmox HA setup with with 3 N200 systems directly connected to each other with 2.5gbit links. Working great. I am hoping someone puts out larger DDR5 SODIMMs. these with 64/72/96GB would be fantastic :)
Oh, and I am using 118GB optane nvme drives as swap location and l2arc for the systems. regular NVMe drives for storage/OS
> GPIO:
RPI: 40-pin Raspberry Pi GPIO header
x86_64: Usually none, only available through USB to GPIO adapters
I have a mixture of RPi's, cheap AliXpress ARM boards, and a stack of Intel NUCs from a decommissioned dev cluster. The only one that consistently works are my Pi's. The GPIO breakout pins are they are used for embedded applications and not try to be desktop PC replacement (though they could do that) And they continue to receive updates for even my oldest. Whitelabel ARM boards might receive the occasionnel update.
Finally, I don't trust the thermal management of Intel enough to stick it in a unclimatized closet and expect it to be working a year later. It takes a lot of work to get an Intel chip to run at low power, low thermals. Something the Pi manages to do out the boxx.
This is a apple to oranges comparison that fails to take into account what makes the Raspberry Pi the Raspberry Pi.
> Finally, I don't trust the thermal management of Intel enough to stick it in a unclimatized closet and expect it to be working a year later.
I stopped trusting the thermals of the Pi after my experiences with the Pi 4. I swapped it in to some projects where a Pi 3 worked great, and the 4 thermal throttled itself the entire time.
I think the solution closes to pi, for x86, is probably… to talk to an Arduino over USB, hahaha. (Closest to the pi in the sense that there’s a healthy maker community behind it).
You hardly need gpio pins for anything these days, it's usually better to use usb native comms or a uart converter or an i2c bridge etc. The added bonus is that you won't burn the expensive Pi when something goes wrong with the peripheral.
ESPHome is usually the non-overkill solution when you just need a few sensors and actuators and don't have to also run the ML inference on the same node :)
I purchased a GMKtec G3 Mini PC with N100 processor for about NZ$300 from Amazon, although I note it would now cost me NZ$255 direct from the GMKtec website. This came with 8GB RAM and a 256GB NVME drive - both of which are upgradeable using standard components. It even has 2.5Gbe networking, Wi-Fi 6, and Bluetooth 5.2, and also included a full Windows 11 Pro licence.
It's near-silent, and I moved my Home Assistant installation to this from a RPi4b, which has made a noticeable improvement in performance. Memory usage sits at around 25% and the CPU usage rarely goes over 10%.
To say I'm impressed with this device is not giving it enough credit. Here in NZ, it would cost at least NZ$200 to get a Raspberry Pi 5 with case, power-supply, storage, etc. And that wouldn't come close to the performance and flexibility this box gives me.
Bought an N100 as a Jellyfin server about a year ago, for less tinkering/maker tasks these are excellent value, would have cost much more for a Pi solution if I could even get it to match the specs I don't feel it would be as reliable.
Has the Pi Foundation ever explained why they are saving $2 worth of memory chips to make these unbalanced versions with pedestrian amounts of memory? The Pi 5 should be firmly in the 8-16GB range given the compute power it has and the tasks it'll have to handle as a result.
Ah yeah seems like the bulk prices I was looking up[0] are in bits, not bytes. Apparently ISPs aren't the only ones trying to make themselves look better through these shenanigans.
Would've only cost them $17 to go with 2x4GB modules, but limited board space and all that I guess. A 12GB version would've been nice to have for $7 more at least, since it's the max single module.
Yeah, the same way as chip density nm values have always been a thing and now the definition has been stretched so thin that it's lost all meaning? This is pure "big number better" marketing bullshit that has no place in a world where even L1 cache is measured in KB.
Bits came first, and it has a very clear meaning. Measuring memory chips in bits is not a marketing thing -- every RAM chip has been measured this way from the very beginning. Including the ones in your computer right now.
Potentially you are only familiar with RAM modules?
> L1 cache is measured in KB
L1 cache isn't a discrete component. It is measured differently because the audience for that measurement is different.
I don't see a point in labelling something that isn't addressable bit by bit in bits. The lowest unit anything running on it can use is a byte and it's perfectly divisible by 8, there's no reason to use bits. I'm not going to count the number of atoms in my drink or the how long my trip to work is in fuckin nanometers.
Doing something the way it's always been done is not a very convincing excuse, if we always did that we'd still be drawing on cave walls and foraging for berries.
Chips are sold to audiences who do care about the bit-by-bit implementation.
> anything running on it
You've skipped a few layers of abstraction. Chips aren't sold to people who implement software. They're sold to people who implement hardware. The bits matter when you're designing the pieces below the software.
Is there any compelling reason to run 32-bit code these days? If so, then 4GB might be a good fit. But nevertheless, these were originally intended as low-spec hobby machines, though they obviously grew way beyond that.
It's not like they ever stop making old versions (except for the original prototype), there are perfectly valid uses for a Pi 3/4 or even a Pi 2 still today. If you need something more basic, those are good options and far more efficient/reasonably priced too.
The Pi 5 seems to be aimed more at the high end heavily demanding tasks, after all it uses 3x more power than the Pi 4. It's not meant to replace it and it should be equipped accordingly for its role.
32-bit code uses less memory, so it makes sense for embedded systems that can't have and don't need a lot of RAM. Doesn't make sense for an application class CPU though.
The problem with the later generation raspberry pi's is that almost everything that made the early generations special have been eroded, to the point where almost every project where the GPIO is required is better served by an mk 2 then an mk 5
The real problem is that where the mk 1 and mk 2 was fully utilized with the mmc/sd cards and usb power supplies people had kicking around in closets and a case made from scrap wood, this slowly stopped being the case with the later models meaning that it was no longer a real alternative to an ESP32 for electronics hobbyist, and that all of those "near mandatory" addons pushes the prices up near that of more powerful N100 based devices.
I've bought/set up at least ten of these cheap mini PCs (~$200) over the years and at least half of them had their built-in WiFi just stop working after some time. Something to be aware of if you're not going to be using it wired. Can always swap over to USB WiFi if it stops working anyway.
I've never this problem with a Raspberry Pi (have a RPi 2 running on WiFi since it came out!) but haven't bought as many of them.
They report idle power consumption of around 3-3.5 W for the pi and 7.5-10 for the N100. Not sure if that’s measured or from the spec sheets. This seems sort of surprising to me, I thought pi’s were known for their decent power efficiency while running, but not super impressive idling.
N100 can hit lower then raspberry pi 5 while still being 2-4x more powerful.
However whole system is slightly more complicated (how much memory? Nvme ssd vs usb flash drive) but they are overall comparable.
I have given up on the Jetson nano, support is absolutely atrocious with the latest image being Ubuntu 18.04. If you want to learn CUDA, rather do it on the desktop with any cheap GPU, or get an Orin nano if you really want to go that route. That said, I expect the Orin nano support to go the same route.
That headless quad-core intel mini PC uses more energy sitting idle than my entire 8C/16T AMD framework laptop sitting idle, including powering the display.
I find that very hard to believe considering the N100 has a TDP of 6W and in benchmarks draws <8W when idle. Ryzen 5 7640U's numbers are 5x that. The laptop's display itself will consume more energy than the entire mini PC on idle.
Plus the manufacturer published TDP numbers are with a baseline load, not 100% idle and as optimized as possible (because at that point you might as well turn the device off).
If I'm reading this correctly, I think the article is citing 'typical' (i.e. what other people said) idle power consumption measured at the wall for similar N100 PCs.
TDP something different: it is typically a term to describe the amount of heat that needs to be dissipated from a chip while under load. While it is a number that is closely related to the power consumption, it is neither the same thing, nor would those numbers ever cite the total consumption of the system. (because, for instance, most components in a computer are passively cooled)
It depends on your use-case. If I was making a file server then idle power consumption is all that matters because both of these machines will saturate a gigabit ethernet link[0] and a home file server is going to be idle 90+% of the time. If I was making a build server instead then I'd actually be using the additional CPU capacity.
[0] actually I wouldn't use either of these machines for a file server due to the lack of ECC, but you get the point.
A desktop system doesn't need to optimize idle power. So I'm willing to bet none of the components (cough intel) was picked for their low idle power consumption. :)
If I wanted another minipc, I'd be looking for one designed like a laptop. The vast majority of consumer computers spend most of their time idle, even when it is something we are directly interacting with at the time (like laptops). In my experience, this is even more true for minipcs. I'm wondering what magic has gone into laptops and hasn't made it into electron guzzlers like the intel minipc in this article.
Running powertop on battery while sitting idle running sway, waybar, and alacritty with my "brightness" setting set to 85/255 but my amdgpu.abmlevel is set to 3 which reduced the "actual_brightness" to ~27/255
PowerTOP estimates your power consumption, it does not actually measure it. Also, this neglects AC/DC power conversion loss. For a more accurate apples-to-apples comparison, disconnect the battery and use an actual power meter (e.g. Kill-a-watt) at the AC adapter.
powertop is just relaying whatever value my laptop's battery is reporting for current+voltage, read from sysfs[0]. Are you suggesting my laptop's battery is merely estimating current and/or voltage? I don't know much about the hardware in my laptop's battery, and I'm not succeeding in finding information online, but I'm going to need a source to believe that considering how trivial it would be for the battery to report real values for current and voltage.
Regardless, your point about AC/DC conversion losses is good considering the article almost certainly was measuring from AC seeing as how it doesn't have a battery.
N100 though allows the PC to be fanless, which is attractive to those of us who hate noises that persist all day (more precisely, that contain a frequency that persists all day, fatiguing the part of the inner ear specialized for that frequency).
The older generation thin-client PCs also tend to only support SATA M.2 SSDs, which are becoming a liability now with the prices rising sharply relative to NVMe due to lack of demand, especially if you want a model with DRAM. Just about anything with an N100 will support NVMe, so you can load them up with a 2TB SSD pretty cheaply.
+1 for fanless. I got one of these optiplexes as they were recommended everywhere. They are indeed more powerful but very closed down. For example one cannot even control the fan speed on them and I hear it whine at idle all day long. If I try to unplug the fan from the motherboard, the whole computer won't boot.
Also, what I could gather from a youtube presentation is that they are electricity hogs (compared to a 5W Pi) unless you get a >9th gen box which has some better low power states (which may or may not be supported by your distro).
I personally own one of these, and unless you are running it with a very intense load, it's pretty inaudible. Active cooling on a 35W part is not very loud.
I'm glad that computer worked for you, but I bought a computer once because multiple people online reported that the model (Intel NUC8BEH) was almost completely silent, but ended up hating it because of its cooling fan's constant whining noise even though the noise was not loud, e.g., not loud enough to interfere with my understanding someone talking.
They might be right for some people, I admit. Just looking on Amazon though, right now in my area, the average price is $150-$200 for an N100, sometimes higher [the best-selling N100 on Amazon is $200]. For the money (see my $120 one), you can get much better performance, greater expansion, a metal case, and even though the power draw is higher... unless you are in, say, Germany, the cost savings of the computer will pay off any savings from the greater efficiency. For what it's worth, it's also a Dell and not a Chinese brand.
Depends on what the use case is. You can't possibly fit that chonker in lots of spaces that a Pi 5 or an N100 will easily fit, and they're much lighter to carry. And the power consumption is atrocious in comparison for battery use.
The killer feature for a lot of home server and pet projects is hardware accelerated encode/decode; with modern codec support. That is a complete deal breaker vs the RPI 5.
Also, don't forget upgradable RAM, and storage without needing a HAT or anything like that.