Raspberry Pi 4 Benchmarks – Heatsink Edition

Orange Pi Development Boards

A few days ago, I ran some benchmarks in Raspberry Pi 4, and quickly found out that using the board without a cooling solution will cause serious performance issues, as in some cases my board was slower than Raspberry Pi 3 model B due to severe overheating.

After playing with LibreELEC yesterday, I’ve now reinstalled Raspbian Buster Desktop on the board, and fitted it with a largish heatsink and some old thermal paste.

Raspberry Pi 4 Heatsink
Click to Enlarge

So I’ll run benchmarks again with and without heatsink. I’ll only run sbc-bench this time.

SBC Bench Installation

Open a terminal windows or connect to the board through SSH and run:


Raspbian Buster will automatically fetch the latest operating systems packages upon first boot, but apparently not the latest firmware:


So I ran rpi-update to get the very latest firmware as well, and rebooted the board:


Normally, you should not have to do it, but Raspberry Pi 4 is really new, so I expect frequent changes at the beginning. Read warning in rpi-update:

WARNING: ‘rpi-update’ updates to pre-releases of the linux
kernel tree and Videocore firmware.

‘rpi-update’ should only be used if there is a specific
reason to do so – for example, a request by a Raspberry Pi
engineer.

Results with heatsink

Room temperature: 28°C. Uptime, idle temperature, and CPU freq:


Benchmark results:


Frequency/temperature monitoring during 7-zip:


It never went over the maximum 85°C temperature, and always under 80°C.

Results without heatsink

Now let’s remove the heatsink, reboot the board and wait a few minutes.

Room temperature: 28°C. Idle temperature:


So 4°C more at idle after about 3 minutes uptime without heatsink..

Benchmark results:


Throttling did occur, and indeed monitoring frequency and temperature during the benchmark shows temperature going close to 85°C, and real frequency dropping as low as 750 MHz to cool the system:

Comparison table and Pretty Chart

We can see that single thread benchmark are not affected by the presence of an heatsink, but the multi-threaded 7-zip compression is certainly impacted.

Benchmark Raspberry Pi 4 “Naked” Raspberry Pi 4 “Heatsink” Ratio
memcpy 2608.8 2540.1 97.37%
memset 3715.6 3541.8 95.32%
7-zip 4423 5397 122.02%
OpenSSL
aes-256-cbc 16KB
64891.56k 64782.34k 99.83%

The differences for memcpy, memset, and OpenSSL are just statistical errors. But Raspberry Pi 4 with heasink is over 20% faster for 7-zip compression as it does not throttle during the test.

Raspberry Pi 4 Benchmarks Heatsink
Click to Enlarge
Support CNX Software - Donate via PayPal or become a Patron on Patreon

20
Leave a Reply

avatar
5 Comment threads
15 Thread replies
0 Followers
 
Most reacted comment
Hottest comment thread
11 Comment authors
ExagaStefanoTLStheguyukJean-Luc Aufranc (CNXSoft) Recent comment authors
  Subscribe  
newest oldest most voted
Notify of
tkaiser
Guest
tkaiser

The sbc-bench output also recorded that swapping happened while running the 7-zip benchmark (RPi folks for whatever reasons love swap on SD card). If %sys exceeds 1% or %iowait exceeds 0% then benchmark execution suffered for sure. The generated numbers could be slightly higher with ZRAM or if no swapping would’ve been occurred.

The fastest way to switch to ZRAM on Raspbian/Debian is to:

A better variant is to rely on Stuart Naylor’s replacement https://github.com/StuartIanNaylor/zram-swap-config though

Also still undervoltage / frequency capping was reported. Based on your unboxing pictures it seems you’re not using the official USB-C PSU from RPi Trading?

Spemer
Guest
Spemer

Thomas, you are a legend.

blu
Guest
blu

As 7zip goes south with each consecutive run on ‘naked’, and the 3rd runs on heatsink/naked are scored ~5300/~4000 respectively, it’d be reasonable to expect a heatsink/naked ratio of well over 25% in the long run on this task.

Domih
Guest
Domih

Now, put a fan on top of the heatsink and do it again. Please. TIA!

Diego
Guest
Diego

Now dip it in liquid nitrogen and redo it again 😛

Member

I am also for the liquid nitrogen test. I think it’s the next thing for the rpi and pretty soon everyone will be doing it.

Domih
Guest
Domih

Hey, I love the smell of the spinning fan in the morning.

theguyuk
Guest
theguyuk

Flat bed Raspberry Pi waffle grill and Teasmaid.

TLS
Guest
TLS

Meh, liquid helium is what real overclockers use…

Diego
Guest
Diego

I’m getting old I got the impression. We should start a new OC xprize to reach -20 Kelvin.

Stefano
Guest
Stefano

Which Heatsink is used? where can i get it?

Exaga
Guest

Jean-Luc,

Thanks for carrying out this RPi4 test. I’m so glad other people have found the same I have have regarding the device. When I first started asking questions and alerting people [on June 26, 2019] of the large amounts of heat that this device produces, not many believed me, and I was labelled a “Troll” by some others. Anyway, I’ve fitted a 28mm x 28mm x 20mm aluminium heatsink to my Raspberry Pi 4 Model B with a 30mm x 7mm @ 5v fan on top of it. The system is really stable now, and very fast. Thermal status of the CPU is ~35’C when idle at the command prompt and ~45’C when compiling binaries from source code using gcc -j6. I’m running Slackware ARM, not Raspbian. I figured, if I was going to install anything it would be a bona fide OS and not some Debian rip-off that’s been hacked to bits so it will actually run on the device. The RPi closed-source firmware doesn’t impress me either. Anyway, I like the work you’ve done. Your results are very similar to my findings on the Raspberry Pi 4 Model B.

Cheers, Exaga