Hacker News Reader

Comment by isoprophlex | Hacker News Reader

I'm happy that they're upgrading the pi, but it's not exactly lightweight computing in terms of power (they recommend a 27W PSU) or performance/watt (see: your cell phone)

To me the charm of the raspi ecosystem was always in giving you a very low power, "always on, almost no power draw" linux environment. Oh well, I guess there's always the older versions or the zero for that.

View on HN

Did you read the blog post? :)

"Raspberry Pi 5 consumes significantly less power, and runs significantly cooler, than Raspberry Pi 4 when running an identical workload. However, the much higher performance ceiling means that for the most intensive workloads, and in particular for pathological “power virus” workloads, peak power consumption increases to around 12W, versus 8W for Raspberry Pi 4."

Alright so 27W with heavy peripheral draw then? But 12 W still means it'll be drawing 2.4 amps instead of 1.6 when a random process decides to 100% it. I guess we can always underclock it.

The real unanswered question is, does it finally have a damn sleep mode so it can save power when idle.

Jeff Geerling measured 1.8W idle power draw.

Based on my experience with previous Pis, I bet you’ll be able to drop that even further by turning off unused board components at boot time.

True, I've gotten Pi 4s pretty low (< 1W), disabling the LEDs, Ethernet, HDMI, and USB controller, but it's never gonna equal a proper suspend.

How did you disable the LED? I mean I can disable LEDs via /proc but not in config.txt correct?

Add

    # Disable Power LED after boot
    dtparam=pwr_led_activelow=off
    # Disable SD card activity led
    dtparam=act_led_trigger=none
    dtparam=act_led_activelow=off
    # Disable the ethernet LEDs - these are Pi4 specific values.
    # Look in the docs for the values for other Pi boards.
    dtparam=eth_led0=4
    dtparam=eth_led1=4

to /boot/config.txt

Thank you!

It appears to depend on firmware or kernel but for recent kernels (recent 6.1 or 6.2) activelow is no longer necessary [1]. It seems to work on CM4.

PS: my bad I disabled the LEDs via /sys not /proc previously:

> # for i in /sys/class/leds/*; do echo 0 > "$i/brightness"; done

[1] https://github.com/raspberrypi/firmware/issues/1742

I've had this drive me crazy.

The required lines were different between the Pi 2 and Pi 4 I was using at the time, and the documentation was not updated (or I'd found old docs).

An exercise in trial and error.

1W is honestly not that great, though. A smartphone will do 10x that out of the box, and still pling for Facebook notifications.

A smartphone has two entirely separate computers in it, the one that runs Android or iOS, and the baseband controller. The latter is built to be very power-efficient while waiting for a radio signal. It will wake up the dormant application OS computer when a push notification comes in.

The same likely can be built on top of an RPi, using, say, wake-on-LAN signaling, or some GPIO as an interrupt source. You'll have to suspend your OS while idle for a prolonged time though.

I read that the "PC style" power button allows both soft and hard shutdown, so I suppose suspend and hibernate come into the picture as well. It's mostly a matter of OS support, the firmware seems to be there.

According to Jeff Geerling the Pi 5 doesn't support suspend in firmware, altough it might be possible with an update.

> soft and hard shutdown

My reading of that was:

"soft" = trigger an orderly OS shutdown, the same as if you executed `shutdown now` on the command line.

"hard" = cut power immediately, as if you just unplugged the power cable.

> The real unanswered question is, does it finally have a damn sleep mode so it can save power when idle.

I dream of DIY'ing a laptop, and RPi looks like a great platform for prototyping that... until you realize there's just no sensible way to put it to sleep. Hibernate + aggressive boot time optimization?

You know it's funny, my current laptop has a weird firmware bug (certified Asus moment) where it refuses to go into sleep mode. If sleep is triggered by any OS it will straight up just shut down completely. I've lost some work the first few times I used it out of habit.

Eventually though it didn't turn out to be much of an issue, SSDs boot real fast these days anyway and I can just do a full power cycle.

Meanwhile I'm genuinely annoyed that my Thinkpad takes a second or two longer to wake up from sleep than my Mac ;)

Is there a way to limit the processor power so that it never exceeds a certain threshold? Because some people might be okay with slower processor if it means less power usage.

I think at least with the Pi 4, you can limit the CPU clock by editing a config file, which has a power limiting effect.

If you set the scheduler to powersave, it downclocks all cores and saves power that way.

Yes, there is a low power mode. And you're forced to use that if you startup with an "under"-powered supply.

Jeff Geerling on YT has a great in-depth review of these options in his latest video.

I read a lot of Jeff's blog, but his vlogs put me to sleep. He isn't charismatic and makes the worst jokes. People forget that a visual medium is visual. He channel is an example of great information that is ruined by the visual medium.

Can you explain their decision to recommend a 27W power supply then?

The new board supports more power hungry IO. If you want to use all of the IO available and push the new SoC then you're going to need a beefier power supply.

What happens if you use those beef options without a powerful PSU?

I don't think the newer models are supposed to completely supersede the older models. They are still available, and they still make them. If you have low power needs, you can use those.

Yeah my intel atom NUCs run at 5W idle and 10W load and they're a ton faster than a pi 4. Probably a pi 5 too. It's not very impressive.

And that's with a real SSD.

NUCs cost hundreds of dollars, though. And they are probably 8x bigger, if you start stacking Pis.

Rasperry Pi board easily cost hundreds of dollars, too, once the scalpers have bought up the little stock that is there.

I've only ever paid RRP and I strongly suggest you do to. Better yet, there are plenty of Pi 4 to purchase these days.

yes, these days. let's have a look at the last three years and see how it was then...

Why is that interesting? Are we expecting another global semiconductor supply chain problem?

Because people who _needed_ to buy a Raspberry Pi simply weren't able to do so.

Exactly the same as every other hardware supplier. For pi there was some kind of scheme for businesses who _actually_ needed the hardware in order to stay in business. No idea how that worked.

Either way, this is all history. Raspberry Pi boards do not cost hundreds of dollars as claimed.

psst https://rpilocator.com/

not much of use for the last three years.

This was used exactly for the last three years so you could get an alert when it became available.

Before that there was no need for a locator, just buy from whatever eshop, everyone had it in stock and at MSRP.

That site is one reason why I was able to buy Pis at list price these last three years.

I am so grateful to the folks behind that site. It's the only reason I was able to track down units when I needed to buy them.

you not checking the site != people not being able to buy them

I'm not saying that it was easy, or anything like business as usual. They've done interviews where they talked about the difficult decision to prioritize companies that would go under without new stock over casual hobbyists. It's one of those unenviable situations where there's no good outcome, just a possibly less-bad one.

Only if you are looking for a latest-gen or last gen machine. You can find some old NUCs for cheap, and there are lots of mini PCs or thin clients for around ~100. Yes about twice as expensive but more than twice as powerful (in both processor speed and features). Of course only if you plan to use it as a Linux computer, not for GPIO stuff.

This isn't super-practical for a commercial application which requires 70 identical machines.

Not only are they expensive and relatively large, machines that have had previous owners often have mystery issues which make them great for home tinkering projects, less so for something that can get you in trouble if it breaks down.

There's a reason companies buy new parts instead of employing teams to scour Craigslist for deals.

I hadn’t heard of these N100 systems before, but in a minute’s worth of searching I found this:

https://www.amazon.com/Beelink-Intel-N100-Computer-Desktop-D...

Faster processor, 16GB RAM, 500GB NVME SSD, with case. $165! That’s damn impressive, considering the RPi5 with 8GB RAM is suggested to sell for $80 (good luck getting it for that little). And Amazon can get it to me in two days.

Yeah it’s definitely bigger, but I wasn’t expecting these systems to be so cheap.

That’s not suppressing since the Pi is usually several process node generations behind.

The NUC's GPIO header: crickets

There was at least one NUC with a full GPIO header set. The DE3815TYKHE. I've done motor control, I2C and sensor IO with it and it was rock solid on that little NUC.

There was also the UP Board which was an Atom SOC with PI compatible GPIO. I believe that's still in production?!

Oh wow that's the first I'm seeing that, looks like it's still available though a bit pricey for over 1 GB of memory.

https://up-shop.org/up-board-series.html

Then again that is severely dated, the Atom was a very capable processor... when it launched, almost a decade ago.

Wow, I never knew they had x86 boards like this. The fact that its x86 is a huge deal itself in itself, due to more software/distro compatibility.

That's a easy problem to solve - https://www.adafruit.com/product/2264

There are other cheaper boards for USB to GPIO out here as well.

It solves it, as much as adding a trailer to your sports car solves your problem of not having a pickup truck. It works, but it's very inconvenient.

Eventually those boards run into limitations and then you have to just opt to go with serial to a microcontroller which misses the entire point of having a SBC.

I agree that its not the ideal solution. Still would be better to use a SBC like the Raspberry Pi or other boards. But was just pointing out that if GPIO needed to interface with slower hardware like relay boards or sensors , then there is an option.

If you want the GPIO, that's a good reason to go Pi. Nothing equal to the software support inside the raspberry pi eco system for it's embedded controls.

If you want a small PC for media/homelab server/cheap desktop, they don't make any real sense anymore.

Yes, if I look at those new Intel N100 based mini PCs with a 15W TDP and their power / price / consumption ratio it would seems Intel took notice of the threat and reacted accordingly.

Especially now that the RPI kind of need a fan and you need to buy the power supply, the storage and the case. Well the RPI has GPIO but for small home server use case nobody cares...

>Well the RPI has GPIO

GPIO can be added via tons of USB to GPIO boards out there.

But I agree - lot of people tend to buy the Raspberry Pis for home servers instead of just opting for used mini-PCs from secondary markets. Even a 7-8 year old Intel CPU in those mini PCs will vastly outperform a Raspberry pi. Even the Raspberry 5. Plus, better I/O options and storage with mini PC.

> GPIO can be added via tons of USB to GPIO boards out there.

Depends on the use case. USB adds a few orders of magnitude more latency and jitter versus what's probably just APB.

Which only matters in very niche usecases though.

Most of the stuff just flashes LEDs or reads a switch. Doesn't matter for that. And for more intelligent stuff there's i2c or api which have their own interface boards.

Yeah, the intel n100 is absolutely fantastic. but even at bargain basement price its at least 2-4x the price:

https://www.amazon.co.uk/TRIGKEY-Mini-PC-Desktop-Computer/dp...

I have the above, it's max powerdraw is <5 watts, even at 50% cpu its <3watts. That comes with ram, SSD, case and motherboard. so to get the pi5 to the same state would need an SD card (boo hiss poor speed.)

From what I've seen the pi5 is 1/3rd faster than the intel j5005. (in pybench at least.)

but comparing to a real intel NUC, of course its going to be faster, the NUC costs an entire order of magnitude more.

The SD is the worst for reliability, but if it's like rpi4, you can put UEFI on the SPI flash and come up on non-SD boot devices without SD now.

But that's all kind of crazy when cheaper, faster SBCs commonly simply boot to reliable, on-board, eMMC.

Yeah, still happy I went with the Odroid M1 for a small home media server. It’s not fast, but it does everything I need out of the box, has a very nice aluminum case, and doesn’t use much power.

eMMC isn't that reliable. It's basically an SD card on a chip.

Difference is they're more optimised for random writes than large files like most SD cards, but you can get such SD cards too ("High Endurance" models)

When was the last time the eMMC on your phone failed? They are very reliable.

I have had several SDs fail sooner or later since rpi1 though which is why I personally won't be using them any further.

I don't know how the N95 and N100 compare, but I picked up a Beelink mini PC with an N95, 16GB RAM and 500 GB NVME for $195 on a sale the other day.

The Pi 5 looks to be $112.

By the time I buy a PSU, SD card, case, RTC battery, etc, I'm definitely not saving money buying a Pi.

Disclaimer: Canadian dollars.

I picked up a Beelink because the Raspberry Pi 4 was unavailable. I use it as a Linux desktop (next to my M1 Air) and a Jellyfin media server. It has replaced the Raspberry Pi for some purposes, but the Pi still has a place when hacking on hardware due to the GPIO.

I bought an N100 for 160 USD including an SSD off of AliExpress. Thats 2x the cost of an rpi4. I benchmarked it and cpu perf is roughly 2x the rpi however i/o was easily 10x without any sdcard or USB shenanigans. Bonus because it's an Intel chip I can use the regular x86-64 os builds instead of some goofy fork. That's very compelling if you don't need a gpio or any of the raspberry pi accessories for your use case.

It's great for my farming robot though. We have plenty of power but want something small and reliable, and the Raspberry Pi Compute Module works wonders for this. More power will be welcome once they release a Compute Module version of the 5.

Is there a reason you're not interested in those NVIDIA Jetson modules? Even excluding the GPU compute, I had thought the Orin boards were significantly faster than a Pi 4.

So for what we’re using it for, which is handling Wi-Fi communications to a central server, handling CAN bus communications to four motor controllers, dealing with RTK GPS and calculating path following parameters for GPS paths, the raspberry pi is perfect. They are cheap, have excellent documentation, and until recently availability was no issue.

Jetson boards are much more expensive. We do plan on making a computer vision add on for our robot which uses the Orin or similar to process images, but I find the raspberry pi perfect for sort of stitching the whole system together.

Would be interesting to see benchmarks of different PI generations at the same power level, assuming the older ones have dynamic power mgmt features too.

The worst part about it is they require a nonstandard, USB-C standard-violating proprietary power brick that does 5V5A.

None of your 100W Anker or Apple power bricks will supply this. They'll do 12V3A, 12V5A, 20V5A, but not 5V5A.

Yet another piece of nonstandard USB-C equipment.

> When using a standard 5V, 3A (15W) USB-C power adapter with Raspberry Pi 5, by default we must limit downstream USB current to 600mA to ensure that we have sufficient margin to support these workloads. This is lower than the 1.2A limit on Raspberry Pi 4, though generally still sufficient to drive mice, keyboards, and other low‑power peripherals.

This is also very uncool, since powering it through the GPIO header with a capable PSU won't trigger the PD signal and makes it impossible to draw any meaningful current through USB? I hope this blockade can be worked around in the boot config.

The blog post says you can override it.

How much power is safe to send through a GPIO pin? I'd imagine you'd want to make use of the Pi's 2 5V pins at that point.

I mean theoretically dupont connectors are rated for 2.5 A so two of them could get it done, but the current will never be completely equal across both and one may get overloaded. Maybe the safest option would be to find some kind of barrel-jack-to-usb-c dongle.

In practice I doubt it would be a real problem since you'd need to max out both the USB draw and CPU load at the exact same time to get the full draw.

It’s worth noting that this is _allowed_ by the USB-PD spec, but you’re right that 5V @ 5A is not common because it is not required. At least it’s better than the Pi 4 was at launch? :D

It would have been better if it could accept 12V3A which many USB-PD adapters can do. Or better yet just a XT30 connector and 5V5A.

12V is an optional voltage, a 25W device is supposed to use 9V 2.8A. Any USB-PD charger of 25W or more is guaranteed to provide that.

The charger is allowed, the Pi 5 is not.

I pulled up the actual USB power delivery standard[1] USB_PD_R3_1 V1.8 2023-04.pdf, and 5V 5A is perfectly compliant, albeit optional.

On page 805 you find Table 10-2 SPR Normative Voltages and Minimum Currents, which specifies that a USB PD source with a rating of 15 < x ≤ 27 watts *shall* support 3A at 5V, however it *May* advertise up to RoundUp (PDP/Voltage) to the nearest 10mA. Requires a 5A cable if over 3A is advertised. 27W/5 rounded to the nearest 10mA comes to 5.4A

[1]: https://www.usb.org/document-library/usb-power-delivery

The problem here is that a power supply offering 5V 5A is compliant, but a device requiring a 5V 5A source is not.

If a device needs 25W it is required to accept 9V 2.8A, if it wants to be PD compliant. This is precisely because 5V 5A is optional for a 25W source, so a device cannot rely on it being present.

It doesn't need it, on the blog they say that if you use a standard 15W USB supply, it (by default) limits the board's USB output current so it can't cause a power failure at 100% CPU.

Oh, I see, so 100% CPU with no USB devices can run fine on 5V3A? That's nice to know. I could just use a powered hub for USB devices then.

I wish they had an auxilary DC power connector of some sort though that could just power the Pi on straight up + and - from a DC power supply. Ideally anywhere from 5-12V.

yeah, i miss the 5-10W early days

You can still buy older models of Raspberry Pi if you don’t need the performance of the new ones.

what if you'd be looking for new low power hardware ?

At this point something like beelink mini pc fits my needs much better - I don’t need tons of gpio ports, etc.

At 27W that is a strong pass vs a NUC. If I go with a low power x86 NUC, I can run all software on it, and probably insert a SSD + additional RAM, for probably ~15W.

That'll be peak power draw though: you have to spec the standard power supply to cope with the CPU running @ 100% & simultaneously feeding the 1G ethernet and all the USB ports at max power.

Jeff Geerling measured 11W peak board power running benchmarks. Idle power was measured at 1.8W.

They claim it will draw less power than the Pi4 for the same workloads.

Max tdp rating is not idle power draw.

Difference being if your workload now demands it, the increased power rating allows faster race to idle, and overall gain in power efficiency.

Peak power draw is 12W, not 27W for the Rpi5 though.

It's 12W, not 27.

Reply on news.ycombinator.com