How to Overclock Raspberry Pi 4 to 2.0 GHz

Yesterday I played with my new “ICE Tower CPU cooling fan”  for Raspberry Pi 4 which cools the board just great but is clearly over the top. Regular reader and commenter m][sko mentioned it was possible to overclock Raspberry Pi 4 to 2.0 GHz after a firmware upgrade. That looks like a perfect task for my new fansink!

I’ve upgraded both the firmware and operating system:


then edited /boot/config.txt as root (sudo) to add the following lines to boost the maximum frequency to 2.0 GHz:


Do not try to set force_turbo=1, as while it would allow to further boost the voltage, it will change a fuse in the chip, and it will void your warranty.

We can now reboot the board to validate the settings, and check those are applied:


Good. At idle the frequency is lower, but that normal:


Voltage is set to 1.0335V:


over_voltage will decrease or increase the max voltage in decrements/increments of 0.025V steps. The range -16 to 8, or 0.8V to 1.4V.

Let’s run sbc-bench.sh as we did previously, and monitor the temperature with rpi-monitor:


Let’s go to the log to check the maximum temperature and actual frequency during 7-zip test:


Broadcom BCM2711 processor Arm Cortex-A72 cores were indeed clocked at 2.0 GHz, and the temperature went as high as 53.1°C, so we still have a good safety margin here.

Raspberry Pi 4 Overclock 2.0 GHz Temperature
Click to Enlarge

rpi-monitor chart actually shows a peak at 55°C, but still 30°C from the 85°C temperature limit of the processor.

Let’s see how much performance we’ve gained against running Raspberry Pi 4 at the default 1.5 GHz frequency.

Raspberry Pi 4 @ 1.5 GHz Raspberry Pi 4 @ 2.0 GHz (overclocked) Ratio
memcpy 2636.2MB/s 2547.2MB/s 0.97x
memset 3707.9MB/s 3651MB/s 0.98x
OpenSSL AES-256-CBC 16K 64744.11k 86338.22k 1.33x
7-zip 5454 6905.33 1.27x
Raspberry Pi 4 Overclocked Benchmarks
Relative performance of Raspberry Pi 4 @ 1.5 GHz vs Overclocked Raspberry Pi 4. Note OpenSSL score has been divided by 10 to show all results in the same chart

Memory benchmarks are basically the same meaning the bandwidth is not limited by the processor, but for other benchmarks, we get 27% to 33% performance improvement which mostly matches the theoretical 33% improvement we ought to expect.

If for some reasons, you’d like to have higher memory bandwidth, you could mess up with sdram_freq and over_voltage_sdram as explained in RPi overclocking options.

Support CNX Software - Donate via PayPal or become a Patron on Patreon

21
Leave a Reply

avatar
9 Comment threads
12 Thread replies
1 Followers
 
Most reacted comment
Hottest comment thread
10 Comment authors
MarkwillyDomitheguyuk Recent comment authors
  Subscribe  
newest oldest most voted
Notify of
blu
Guest
blu

Now that’s a good use of that fan.

Robert
Guest
Robert

I was expecting that you submerged that thing in liquid nitrogen when I read the headline 😅

theguyuk
Guest
theguyuk

Would someone really use it long term sat open on desk with the ice tower or should it be tested in a case?

Domi
Guest
Domi

I’m waiting for the 4GB model. I still have a few giants CPU coolers from the late 90s in my stash. I think I’ll be able to adapt one of them.

theguyuk
Guest
theguyuk

Let us all know how you go on and can you test when in a suitable case.

Domi
Guest
Domi

There are very nice ATX full towers now :)) This being said these coolers were work of art at the time, I’m thinking in particular of one which had a full circle of copper plates (not the horizontal one Zallman.) Almost a crime to hide them in a box.

Member
Michal Lazo

Another nice idea for article is
raspbian 32bit (debian buster 10 ) vs ubuntu 18.04 64bit on raspberry pi
there is download for ubuntu 18.04 64bit for raspberry pi 4
https://jamesachambers.com/raspberry-pi-ubuntu-server-18-04-2-installation-guide/

as it is kind of interesting as for example x264 encoding is like on 3/5 speed on 32bit

dimtass
Guest
dimtass

Nice one. I think that the frequency in the table header is inverted.

Still, I think that those high temps and need for such big heatsinks are limiting it’s usage. Maybe underclocking is better than overclocking in this case so you can use it without heatsink. For example a nice benchmark would be, how much do you need to underclock rpi4 so it doesn’t throttles with a simple passive heatsink and how well it performs at that frequency.

tkaiser
Guest
tkaiser

> I think that the frequency in the table header is inverted.

Are you referring to this?

‘Fake’ is what the Linux kernel thinks would happen. But Linux on the RPi is just a secondary OS running on the ARM guest processors. And a guest OS like Linux on the RPi has no idea what’s really going on: https://github.com/raspberrypi/linux/issues/2512#issuecomment-382703153

The Linux kernel will display 2.0 GHz even if the ARM cores are just running with 600 MHz due to undervoltage or throttling.

That’s what the ‘real’ column is for. Here the primary operating system running on the VideoCore (ThreadX) is asked about the real clockspeeds using /usr/bin/vcgencmd measure_clock arm. No idea why ThreadX reported once 2001 GHz since Jean-Luc configured arm_freq=2000.

dim
Guest
dim

No I meant the table in the end of the post with the memcpy and memset tests.
_But it’s my mistake_.
I expected those test to be faster @2.0GHz and I thought it was inverted.

dim
Guest
dim

Thanks for the info, I haven’t realised that Linux is a guest OS. Interesting and concerning at the same time. I’ll have a look at it.

tkaiser
Guest
tkaiser

See starting at ‘The real brain of the Pi is not open source’ here https://ownyourbits.com/2019/02/02/whats-wrong-with-the-raspberry-pi/

Domi
Guest
Domi

Thanks for the link!

Paul M
Guest
Paul M

Mickmake did a good review where he tried to see just how little power it took to boot a pi.
https://youtu.be/AHIld3cjmTM

Domi
Guest
Domi

Alternative article title: How to make your Pi 4 run for two weeks instead of two years :))

Member
Michal Lazo

We didn’t change voltage too much so I don’t see any problems 🙂

Member
Michal Lazo
willy
Guest
willy

Your stability issues might possibly be caused by light or other radiations that can reach the silicon while they were blocked by the aluminum before. Make sure no area around is visible from outside. The edges are not supposed to be sensitive but the hole square centimeter of silicon can be sensitive to intense light. Just try to photograph it with a flash during a build and you’d be surprised to see it hang or crash.

Member
Michal Lazo

It looks like when I exchange thermal tape to proper thermal paste everything works just fine and temperature is much better. It looks like 3M thermal tape from isn’t big win after all.

willy
Guest
willy

That doesn’t surprise me. I found as well in my build farm that thermal tape isn’t that good.

Mark
Guest
Mark

Mine peak at 65 degrees, passively cooled with official Pi micro USB power adapter converted to usb-c with adapter

http://ix.io/1QAG