- The Pi 3B has 10/100 Ethernet, 802.11n (single-band) WiFi, Bluetooth 4.1. Power idled at 1.4W and peaked at 3.7W.
- The Pi 3B+ removed the 10/100 Ethernet in favor of USB Ethernet (~300Mbps w/USB2.0). CPU cores were overclocked from 1.2GHz to 1.4GHz (so a heatsink is more necessary), with ~15% increase in benchmark performance. It added 802.11ac (dual band) WiFi and Bluetooth 4.2 w/BLE. Power idled at 1.9W and peaked at 5.1W. This is also the only 3-model supporting PoE (w/ extra HAT).
- The Pi 3A+ removed Ethernet and reduced USB to a single port. The RAM was reduced from 1GB to 512MB. Power idled at 1.13W and peaked at 4.1W. The A+ form factor is more compact. Overall the 3A+ is smaller, cheaper, and less power draw than the 3B+ (but not as low as the 3B).
The lowest power draw with acceptable performance is the 3B. For slightly more power draw and more CPU performance, go with 3A+. For "everything" (including PoE) the 3B+ is it.
If you want the 3A+ but don't need the video, want a smaller form factor, and half the power draw, the Pi Zero 2 W is it. Though the Pi Zero 2 W is supposed to be cheapest, due to demand it's often sold out or more expensive. The 3A+ is still cheap (~$25) and available, with the downside of the higher power draw and larger form factor.
(disabling HDMI, LEDs, Wifi, Bluetooth, etc reduces power draw more. in testing, the 3A+ drew less power than the Zero 2 W with everything disabled. all of them draw ~0.1W when powered off)
Is that with the same load? The chart in the article shows a Pi 3 and Pi 4 using the same idle power, with the 4 drawing more under full load. But the 4 can do more at full load, raising the question of what would be the 4's power usage running a load equal to the 3's full load?
Sometimes it’s more efficient to run a task at full CPU power for 25ms than low (non-idle) CPU power 100ms.
It wouldn’t surprise me too much if the 4 could run the 3’s full load in less power than the 3 does.
Most reports show the Pi 4 drawing ~2.8 W in idle headless mode, and the Pi 3B+ drawing ~1.9-2.0 W in idle headless mode. With full load the Pi 4 draws more (6.4W, to the Pi 3B+'s 5.1W) with the same test procedure.
But you do have to check the testing method; enabling/disabling hardware features changes the figure, and each additional USB peripheral draws more power. Otoh, to get a "max power draw" reading you have to enable everything and stress all CPUs at once, and then it will dip under thermal load.
The 3 was and the 5 is plugged into an energy monitoring smart outlet. Here's a graph of power for that outlet [1] for the last 30 days.
HA uses long term statistics for older than 10 days, which is why the 2/3 to the left shows much less variation. The switch from the 3 to the 5 is somewhere a little way into the right 1/3, where it is using shorter term data which has a much higher sample rate.
It does look like the average power goes up with the 5, but not nearly as much as I would have thought. The left 2/3 which was definitely all RPi 3, averages 2.48 W. From Jan 17 to the present, which was definitely all RPi 5, it averages 2.64 W.
If those reports of 2.8 W for an idle 4 are accurate I'm curious why my HA 5 is lower. I've got an unused energy monitoring smart plug of the same kind my HA Pi is on. I'll have to put the 4 on that and see how it compares.
I ran it on my Pi 3, 4, 5, Intel iMac, and on my cheap Amazon Lightsail instance. Here are the results, in seconds:
680.4 RPi 3
274.5 RPi 4
131.3 RPi 5
108.5 Lightsail
78.7 2017 iMac (3.4 GHz Intel Core i5)
45.2 M2 Max Mac StudioLooking at the instance creation page it looks like it is no longer available. The cheapest now has 2 vCPU and the same other specs.
If anyone would like to run it on something else for comparison I've put it Pastebin [1].
Normally, a script converts the puzzle into a header file that describes the puzzle in terms of the data structures the solver uses, and then the solver is compiled with that header included via #include, but for benchmarking I manually included the header, so it is just a single C file that you simply compile and run. Add a -q flag to have it just find and count the solutions but not display them. For this particular puzzle it is fine to let it show the solutions. For some puzzles with a ton of solutions it displaying them can take significant time.
If anyone actually wants a Pips solver to solve Pips puzzles, there's a copy of mine along with the scripts that convert puzzles from the JSON format the NYT provides for download to the header file for the solver and some other stuff as a shar archive on Pastebin [2]. Pastebin's download link messes up something with that, so click the raw link and copy/paste from the browser.
Note: I'm still doing a lot of experimenting with it, both in the solving algorithm and in how the display the results, and am not at a point where I'm able to accept outside contributions. There are other Pips solvers that have been put on Github (and even posted on HN) that are (1) more sophisticated than mine, and (2) almost certainly more open to contributions. Just use mine if for some reason you want a fairly simple (aside from the damn display stuff...) brute force solver in C and aren't going to expect anything else from me :-)
126 Intel N5105, gcc -O2 (Debian)
98 Neoverse N1, gcc -O2 (Debian) - Oracle Cloud
63 Snapdragon 8gen3, clang -O2 (Termux) - Xiaomi 14
59 Ryzen 7600, clang -O2 (MSYS2 / Windows 11)
48 Ryzen 7600, clang -O2 (Debian)
45 Ryzen 7600, gcc -O2 (Debian)(Also I was just looking at the pricing, you don't get any extra CPU cores until the $84 tier!)
It’s a single-core 700MHz ARMv6 chip with 512MB of RAM. It's a fossil—a Pi 5 is 600x faster (according to the video). But for the 'low-bandwidth' task of routing some banking traffic or running a few changedetection watches via a Hetzner VPS (where the actual docker image runs), it’s rock solid. There’s something deeply satisfying about giving 'e-waste' a second life as a weekend project.
Well, it's still running today on the original SD card. At noon today it processed its 1,055,425th record in the database.
Still, if it ever crashes, I'll just tear it down. :)
One nice thing is I can print to the CUPS server even if the printer is off
https://www.raspberrypi.com/news/printing-at-home-from-your-...
* They have full sized HDMI ports
* They will happily run using any random old USB charger and not overheat.Then again we use a kW or two to microwave things for minutes on a daily basis so who really gives a shit.
Enough energy to run that thing for an entire year in under 1/2 a gallon of gasoline.
When you can pretty easily offset the entire yearly energy use by skipping a mow of your yard once, or even just driving slightly more conservatively for a few days... I'm not so worried about the power use.
In my region - it's about $3.50 in yearly power costs.
This site[0] claims a Prius Prime XSE gets 1.42 miles/kWh. Or (1.42 miles /1000Wh)*2 = 0.0028 miles. Which is ~14 feet, which is significantly more in line with my expectations (though still high)
[0] https://www.motortrend.com/reviews/2024-toyota-prius-prime-x...
The easiest way to do the calculation would be, assuming a Prius Prime can do M mi/kWh on battery power, is to calculate 0.155 mi/day x 1/M kWh/mi x 1 day/24h = 0.0065 kW = 6.5/M W. That gives us W which can directly be compared with the 2 W he gave.
Also, 1.42 mi/kWh seems way low for battery power operation. I'm pretty sure that is for mixed gas/electric operation, expressed in MPG-e (47.9) and mi/kWh for convenient comparison to pure EVs. (You can convert between MPG-e and mi/kWh used the conversion factor for 33.7 kWh/gal.
It has a 13.6 kWh battery and a 39 mile all electric range, which suggests M = 2.9 mi/kWh. Plugging that into 6.5/M W gives 2.2 W.
M is probably actually a little higher because the car probably doesn't let the battery actually use 100% of its capacity. Most sites I see seem to say 3.1-3.5 mi/kWh.
On the other hand there are some losses when charging. On my EV during times I've the year when I do not need to use the heating or AC the car is reporting 4.1 or higher mi/kWh, but it is measuring what is coming out of the battery.
When calculated based on what is coming out of my charger it works out to 3.9 mi/kWh. This is with level 2 charging (240 V, 48 A). Level 1 charging is not as efficient as level 2.
If we go with 3.1-3.5 mi/kWh, and assume that is measured on the battery output side and that the loses during charging are about 8%, we get 2.9-3.2 mi/kWh on the "this is what I've getting billed for" side. If we use the average of that and plug into 6.5/M W we get 2.1 W.
Nice! Even though I've got a Proxmox serve at home running on a real PC (but it's not on 24/7), I do run my DNS, unbound, on a Pi 2. It's on 24/7 and it's been doing its job just fine since years.
Fast forward. We’re getting ready to role out our next generation. It’s based on the Pi Compute Module 4 (the CMs are basically just the basic Pi and you put your own carrier board for peripherals under it). It is amazing. It has easily 20x the power, 20x the RAM, better temp specs and such, a great eco system, uses about 30% less power, and about 1/5 of the price. The only thing we’re not sure about yet, is the robustness of the BLE with the onboard radio chip.
It’s amazing how far these things have come. For low volume product builds, it’s hard to find a reason not to use one of the CMs.
It's funny how Raspberry Pi started out for an educational market, and accidentally revolutionized the embedded market.
Used 1L mini PCs (EliteDesk, ThinkCentre Tiny, etc.) with i5-8400T/8GB/256GB go for $50-100 on eBay. You get x86 compatibility, NVMe support, real Ethernet, and no thermal throttling. Running an EliteDesk 800 G4 with 2.5GbE adapter and 2x2TB NVMe for home server duties. Draws ~15W idle, handles everything I threw at it. The Pi would need USB adapters for any of that. Pi still wins for GPIO projects and actual embedded use. But for "small Linux box" use cases, the used business mini PC market is tough to beat.
When I did that on Pi3 when it first came out you could crash the system because the thermal throttling wasn't fast enough (the temp sensor was on the GPU not CPU). When I reported the issue on the pi forums the answer was essentially "why would anyone ever want to do that"
But still haven't gotten a full run on any Pi prior to the 4 B.
The more things change, the more things stay the same.
With all due respect to Raspberry Pi and everything they’ve accomplished in the educational and hobby space,
I felt that one in my bones. I suspect a lot of people with embedded experience who worked with Raspberry Pi over the years feel it too.
I can still barely believe they got Grand Theft Auto 5 running on that thing.
I am interested in this, I have been using Raspberry Pis for various projects and home servers since the original - Currently one is hosting my navidrome music server, my password manager, and several other local network servers.
I feel the upgrade each time, and then get used to it, as I suppose we tend to do. I still remember the upgrade from 1 to 2 being the most impactful to me personally though. (I think maybe because that's when game emulation became viable?)
A Youtube tab, web browser modern enough for YouTube, and OS modern enough for that web browser, all fit in 1GB of memory? Wow.
YouTube is an absolute clown show. It's so bad that I'm certain Google devs are actively making it terrible on purpose. I use Newpipe on an older (but not that old) tablet. Whenever Google breaks Newpipe and I have to use a browser, it takes like 30 seconds just to load the page.
Decoding video is trivial when you have hardware decoders.
The graphs are interesting but, really, if you’re considering your readers rather than SEOing for last decade’s search engine technology, you should lead with them and discuss the findings afterwards.
I.e., get to the point quickly and then unpack the detail.
It’s interesting seeing where the incremental vs revolutionary improvements have occurred. CPU-wise, a huge leap with the 3 and then solid but steady improvement with 4 and 5. But the most meaningful jump in GPU performance seems to be 4 -> 5, and I’d be really interested in what that maybe opens up in terms of console emulation.
Anyway, fewers ads, please. Scanning through the article on mobile felt like playing hopscotch in a minefield.
The UPS says 35W for all of it, but I’ve always been too lazy to unplug devices to see how it breaks down. I’m also not sure how accurate the measurements are, especially under a load that low.
I’d be willing to believe the mini PC draws less than the other components at this point.
My applications have remained the same for many years my octoprint and retropie don't require more FLOPs as time goes on but I'd really enjoy a modern board that has fewer headaches. Works on any normal USB port instead of requiring specialized power supplies, doesn't brown out and reset as much, doesn't heat up as much, etc. I suspect "a pi 3, but now with fewer headaches" would sell better than "a pi 3 but even more headaches and bigger numbers that you don't want".
Except it's not even fixed function blocks, it's the 12 core VideoCore IV GPU running software that does the decoding.
VideoCore is the real Raspberry Pi, the ARM block running Linux was just a subprocessor that VC controls.