Raspberry Pi 4 Can Be Safely Overclocked To 2.15 GHz

szczys writes: When the Raspberry Pi 400 (a keyboard form-factor single board computer) was released last week, the company hinted at overclocking. Testing has now shown that the heat spreader used in that design does an excellent job. The chip was already clocked at 1.8 GHz, versus the stock 1.5 GHz in the original Raspberry Pi 4 Model B board. But it can be safely overclocked to 2.15 GHz, as can the Compute Module 4 with an adequate heat sink.

At 2.0 GHz, the Pi 400 got up above 60 C and showed signs of continuing to warm up even after 50 minutes, but it was nowhere near throttling. So I tried 2.2 GHz, at which speed the CPU refused to boot entirely. Backing down to 2.15 GHz, it ran just fine, so I left it for three hours. It settled in at a cozy 62.5 C, which is warm, but well within specs.

I ran the CM4 with the larger heatsink at 1.8 GHz to give some basis for comparison to the cheap heatsinks. What a big difference a big hunk of aluminum makes! It settled in at a comfortable 68 C or so. Even pushing it up to 2.15 GHz and leaving it for a couple hours, it stayed just a hair below 70C (158F) -- a safe margin on the throttling threshold -- and only a few degrees warmer than that huge heat spreader in the Pi 400.

Further reading: The Verdict After Hackaday's Teardown of a Raspberry Pi 400: 'Very, Very Slick'.

Nice.

[thegreatbob]

Brings back thoughts of my Athlon XP 1700+ starting to crap out at around the same frequency, though that required an enormous (for the time) heat sink, and was surrounded by hardware drawing an extra 100+ watts on top. I wonder which has better per-core IPC...

Si life

[OrangeTide]

Has the life of the silicon been characterized under these overclocking scenarios? The BCM2711 is 28nm I believe, and it would be pretty reasonable to compare its life to similarly sized wafers of the same process node. (I don't have that data, sorry). If overclocking drops a 30 year+ Si to 10 years, maybe we don't care. If you run it hard in a cluster and some portion of them fail after 2 years, that might be pretty annoying.

Re:

[thegarbz]

Has the life of the silicon been characterized under these overclocking scenarios?

Yes. The answer is, you'll end up throwing perfectly fine Pi4s in the trash because they are disposable cheap little processing bricks.

If you run it hard in a cluster and some portion of them fail after 2 years, that might be pretty annoying.

I'm not sure what's more annoying, having a pi fail while running as a cluster, or simply running the cluster in the first place. The only use case for this is to simulate clusters in an educational environment, and even then setting up a swam of VMs is probably a better idea.
You're not going to be running a cluster of these flat out for any reason doing any kind of importan

Re:

[OrangeTide]

I'm not sure what's more annoying, having a pi fail while running as a cluster, or simply running the cluster in the first place. The only use case for this is to simulate clusters in an educational environment, and even then setting up a swam of VMs is probably a better idea.

a pair of RPi4 boards not a bad way to learn how to setup vSphere and ESXi [vmware.com]. But given that performance is less critical than learning and experimenting in such a setup, there is probably no need to overclock. (education is important work)

But can we underclock it???

[Smid]

So we can run it at lower power and without a fan?

CPU is not the be all and end all. My Pi4 was bought because of the USB3 and Gigabit eithernet, it made it a perfect low power smb/nfs/deluge/subversion server, and great for scripts to pull things from iplayer at 4am, digital radio puller, and now its going to have an SDR stick detecting 433mhz signals around the house, auto triggering lights and sending alarms to my phone.

None of this is CPU intensive. It's an IO octopus of mighty proportions, who need spe

Re:

[sunami88]

But can we underclock it???

In my experience (with a 3A+), underclocking a Pi causes it to run hotter. Not joking. Never figured out why.

Re:

[thegreatbob]

Not sure if it implements core throttling, and if it does, dynamic voltage - but I experienced stuff like this with older laptops... try to limit the CPU speed, stuff takes longer, leading to e.g. CPU stays non-idle longer (work done per unit of power consumption is worse), screen stays on longer while you're work, hard drives don't get to spin down ('member when hard drives weren't constantly tickled by the OS?), power consumption from the chipset, memory, etc. remains the same... netting possibly worse ba

Re:

[toddestan]

It depends on what you're doing with it too. Sometimes a faster CPU will run cooler, because it can get its work done quickly then go back to a low power idle. If you slow it down so it actually has to run at 100% a lot and has less opportunities to idle, it can actually use more power in the end.

Re: But can we underclock it???

[kenh]

The latest firmware on the RPi 4 allows operation without a fan or heat sink. The RPi 400 (Keyboard Pi) has a massive 2.5" x 7" (estimate) sheet of aluminum bonded to the SoC.

The RPi 400, as discussed here, can be overclocked without issue or additional cooling to over 2.1 GHz.

The traditional RPi 4 can be run normal-clocked without cooling.

Re:

[nickovs]

The short answer is yes, you can underclock. By default the Pi4 will scale the clock speed depending on load, so what you are really doing is changing the maximum clock speed. If you are trying to save power then you might want to look at both reducing the maximum frequency and also dropping the voltages for the CPU and SDRAM. If you look in the overclocking instructions [raspberrypi.org] you will see that you can set negative values for the over_voltage* settings to bring the voltages down as well as changing the maximum CP

Uhh, ok?

[sunami88]

All Pi's are a little different and some take to overclocking better than others. Some have slightly shitty wifi. I remember running my Pi 4 (4gb) at ~2.1ghz or thereabouts not too long after they came out.

After all the complaints about heat they released a bootloader (firmware?) update that seemed to lower the threshold for overclocking, at least on mine.

That said, I've since grabbed an 8 gig (because I have a legitimate problem) and it runs all day long at 2.0ghz...

So I'm not sure how this is news, and

YMMV

[sremick]

First of all, The RasPi 400 is a rather different beast than the RasPi4. The 400 is a complete package, while the Pi4 on its own does not have a case or cooling or anything.

Regardless, a bit presumptuous of the OP to state such a thing categorically. Anything above spec is going to be "YMMV". My Pi4 can't go above 2GHz without locking up within a short amount of time, usually within an hour. Even at 2GHz I've had at least 2 lock-ups so even that isn't 100% safe. And I have heatsinks and active cooling.

so much promise, so many ideas, too little time

[jm007]

this is great news! now mine will be able to sit around collecting dust even faster than before!

That's cool, but

[t4eXanadu]

It's idling at 68c. What does it run at under load? My pi 4 runs at a cool 48c with heatsink and fan, but it gets pretty toasty under significant load. I suspect it won't last too long overclocked if you're doing anything CPU intensive for long periods.

does not - will not - work with all processors

[lkcl]

ok, so when a wafer is manufactured (10in usually) the distribution of the ions and the base properties of the actual silicon itself are non-uniform. this ultimately results in some of the dies being capable of running at higher clock frequencies than others.

a typical mass-produced processor they simply don't bother to "speed-grade" them. the Allwinner A20 was listed as nominal 900mhz but *SOME* of them were capable of running at up to 1.2 ghz.

it is only really Intel and AMD that bother to do extensive te