Raspberry Pi 5 review – Part 2: Raspberry Pi OS Bookworm, benchmarks, power consumption, and more

A few days ago I finally went through the Raspberry Pi 5 kit I received last September going through the items and booting it with Raspberry Pi OS bookworm. I’ve now had time to perform more tests to check out the performance with benchmarks and test various features on the Raspberry Pi 5. So I’ll report my experience in the second part of the review and compare the Raspberry Pi 5 to the Raspberry Pi 4 and some other Arm Linux SBCs.

System information in Raspberry Pi OS Bookworm

Last time around, I installed the Raspberry Pi 5 in its official case, but for most of the testing, I decided to go back to the bare board fitted with its active cooler since it’s the best cooling option as we’ll see further in the review.

Raspberry Pi 5 Review Raspberry Pi OS Bookworm

Let’s first check some of the system information:


The Raspberry Pi 5 runs Debian 12 with Linux 6.1 as expected and the Broadcom BCM2712 processor (still shown as BCM2835 as in all RPi boards) does come with four Cortex-A76 cores.


inxi utility further confirms my board comes with 8GB RAM, the 2.4 GHz frequency advertised for the BCM2712 CPU, and lists the peripherals with two HDMI ports, Ethernet, WiFi, and Bluetooth, and we can also see Wayland and PipeWire as now used in Raspberry Pi OS as was announced during the release of the Bookworm version of the operating system. The microSD card capacity is reported correctly at around 32GB, and we can see 100MB of it is used for swap. There also appears to be a reference to the Raspberry Pi RP1 chip in the Network section, although it’s the same for the driver…

If anybody is interested I also saved the Raspberry Pi 5 Linux boot log.

Raspberry Pi 5 Benchmarks

Let’s start the Raspberry Pi 5 benchmarks with Thomas sbc-bench.sh script:


First, the good news is that there was no CPU throttling and the temperature never exceeded 66.1°C when using the active cooler in a room with an ambient temperature of around 28°C.

Let’s compare the memory bandwidth and 7-zip results against the Raspberry Pi 4 and other SBCs such as Khadas VIM4 (Amlogic A311D2), ODROID-N2+ (Amlogic S922X), Radxa Rock 5B (Rockchip RK3588), and others.

Raspberry Pi 5 memory bandwidth 7zip vs Rockchip RK3588 Amlogic A311D
Higher is better

The memory bandwidth of the Raspberry Pi 5 is much higher than the Raspberry Pi 4, but based on memcpy and memset results the Rockchip RK3588 platforms are still way ahead in that metric, and they also outperform the Raspberry Pi 5 in 7-zip benchmarks which benefit from a larger number of cores (Rockchip RK3588 comes with 4x Cortex-A76 cores, 4x Cortex-A55 cores). The Raspberry Pi 5’s 7-zip performance is about the same as the Khadas VIM4 and ODROID-N2+ SBCs, and about double the performance of the previous generation Raspberry Pi 4.

Raspberry Pi 5 AES 256 performance

The Broadcom BCM2712 is the first processor used in a Raspberry Pi board that comes with Armv8 Crypto extension and that shows in AES-256 benchmarks with the new Raspberry Pi 5 over 21 times faster than the Raspberry Pi 4.

Time to have a look a web browsing performance, especially since the Raspberry Pi OS Bookworm introduced an optimized version of Firefox besides the usual Chromium browser.

We’ll first use Speedometer 2.0 to check the performance in Chromium…

Raspberry Pi 5 Chromium Speedometer 2.0

… and in Firefox.

Raspberry Pi 5 Firefox Speedometer 2.0

Firefox is often significantly slower than Chromium in benchmarks, so Raspberry Pi Ltd and Mozilla have done a relatively good job optimizing Firefox for the Raspberry Pi 5 with Chromium achieving 63.5 runs per minute against 56.6 on Firefox. The chart below compares that performance against the Amlogic A311D2-based Khadas VIM4 (35.6 points), the Amlogic A311D-powered Khadas VIM3 (against 25.6 points), the Rockchip RK3588S-based Khadas Edge2 Pro, and a Raspberry Pi 4 overclocked at 2.0 GHz (21 points).

Raspberry Pi 5 vs Rockchip Amlogic SBCs Speedometer 2.0

Please note that web browsers are constantly evolving with performance optimizations (or bloat) getting added after each release, and we did not redo the tests on older platforms due to time constraints.

I also ran the glmark2 benchmark to test graphics performance on the Raspberry Pi 5.


920 points look to be on the low side, as a Rockchip RK3588 hardware such as the NanoPi R6S mini PC router achieved 4,525 points with glmark2-es2-wayland… [Update: I was unable to find glmark2-es2-wayland  yesterday, but it’s just an apt install:


The score is indeed different, and does not look as bad. But 2,036 points is still significantly lower than the 4,000-4,500 points achieved on Rockchip RK3588 hardware.

]

The WebGL Aquarium demo works fine with Chromium and Firefox, although the frame rate is significantly better with the Chromium browser.

WebGL Aquarium Chromium Raspberry Pi OS Bookworm
1000 fish, 48 fps – Chromium

 

WebGL Aquarium Firefox Raspberry Pi OS Bookworm
1000 fish, 35 fps – Firefox

For reference, in the NanoPi R6S review, I mentioned that “the demo renders smoothly with 1000 fish at 60 fps, and it’s still good with 5,000 fish at around 30 fps”, so the Raspberry Pi 5 is indeed not quite as performant as Rockchip RK3588/RK3588S system when it comes to graphics either.

So far, we can say the benchmarks show that Raspberry Pi 5 is a clear step up compared to the Raspberry Pi 4 with typically 2 to 3 times better performance, and even over 20 times for cryptographic workload (like AES, TLS, etc…), but it does not quite reach the performance of the popular Rockchip RK3588 CPU although it can get close for some specific workloads.

microSD storage and USB 3.0 benchmarks

I built iozone for the Raspberry Pi to run the storage benchmarks on both the provided 32GB Class A1 microSD card as part of the Raspberry Pi 5 kit and a USB 3.0 NVMe SSD drive.

Results for the microSD card:


SD card performance is supposed to be improved thanks to the RP1 chipset, but it’s hard to see with that specific microSD card. Reads are up to about 89MB/s and writes max out at 19MB/s. Random read and write speed is also important, and it looks decent which is probably why I didn’t notice any massive slowdowns due to I/Os during use. But note that (more expensive) competing boards with built-in 32GB or 64GB eMMC flash are usually much faster and more responsive.

I also tested the USB 3.0 bandwidth (5 Gbps) with an ORICO enclosure fitted with an Apacer NVMe SSD formatted with EXT-4.

Top USB 3.0 port:


Bottom USB 3.0 port:


In both cases, we got about 388MB/s sequential read speeds and 411MB/s write speeds, or about as expected for a 5 Gbps USB link. It’s unclear why writes are faster than reads as we disabled caching in the iozone command.

Gigabit Ethernet, WiFi, and Bluetooth

We’ll now use iperf3 to test gigabit Ethernet and WiFi 5 network performance. I’ll also test Bluetooth with my phone and a Bluetooth headset.

Here’s the Raspberry Pi 5’s gigabit Ethernet download speed using UP Xtreme 11 Edge mini PC‘s 2.5 GbE port on the other side of the connection.


Now for upload:


Both are great and around 940 is expected in either direction. Let’s now try a more demanding full-duplex (bi-directional) transfer:


Again excellent.

Let’s now test WiFi 5 (at 5 GHz) connected with a 433 Mbps link to Xiaomi Mi AX6000 router.


Download:

Upload:


The performance is fairly good with 224 Mbps uploads and 259 Mbps downloads, but if you want better network performance Ethernet should be used when possible. As a side note, high-end platforms with WiFi 6 can deliver over 1 Gbps transfer rates with that testbed.

Finally, I tested Bluetooth with my Android smartphone. The good news is that pairing worked flawlessly. The bad news is that Raspberry Pi OS does not know what to do with my phone with the error message:

Pairing successful – this device has no services which can be used with Raspberry Pi

Raspberry Pi OS Bluetooth Smartphone

So I can’t transfer files or do anything else for that matter.

Let’s try with a Bluetooth audio headset…

Raspberry Pi 5 Bluetooth Audio

It works! I could listen to the audio from a YouTube video, but the volume control is not working right. I can only mute the audio or turn it on at 100% at whatever position it is on the slider. However, if I use the volume buttons on my headset, then the volume slider is properly synchronized in Raspberry Pi OS… That means I can control the volume from my headset, but not from Raspberry Pi OS except for mute/unmute.

YouTube and 4K video playback

I then tested YouTube videos up to 1920×1080 resolution at 30 fps in both Chromium and Firefox. 4K (2160p) resolution was not an option despite selecting a 4K video.

RPi 5 YouTube Firefox
Firefox
RPi 5 YouTube Chromium
Chromium

The video played smoothly in both web browsers with only a few frames dropped according to the “Stats for Nerds” overlay window.

Let’s now try to play some local 4K H.265 videos as well since the Broadcom BCM2712 SoC is supposed to support H.265 video decoding. So I connected a hard drive with some 4K video samples and played them by clicking on the file which opened VLC video player:

  • Beauty_3840x2160_120fps_420_8bit_HEVC_MP4.mp4 (H.265, no audio) – OK
  • MHD_2013_2160p_ShowReel_R_9000f_24fps_RMN_QP23_10b.mkv (10-bit HEVC, 24 fps, no audio) – First frame only for the duration of the video
  • Fifa_WorldCup2014_Uruguay-Colombia_4K-x265.mp4 (4K, H.265, 60 fps) – Video and audio OK

While two of the video could play fine if I enter full-screen mode, I can see a thick black bar at the bottom of the video (but no such thing at the top), and changing the aspect ratio does not help.

4Kp60 HEVC video playback CPU usageCPU usage while playing a 4K H.265 video at 60 fps is rather low which should indicate hardware video decoding is indeed working. For comparison, If I switch to a 4K H.264 video at 30 fps, the CPU usage is higher as software video decoding is used instead.

Raspberry Pi 5 H.264 software video decodingIt should be the same for other video codecs, as I understand Broadcom BCM2712 only supports H.265 hardware video decoding, and all other codecs must be handled by software.

Raspberry Pi camera

I still have a Raspberry Pi camera module 3, but I had to skip testing because the camera connectors on the Raspberry Pi 5 are now smaller… It’s not such a big issue, as I understand it’s simply a matter of getting a $1 cable and you should be good to go with the new Raspberry Pi SBC and existing official camera modules.

Note that Broadcom BCM2712 does not support hardware video encoding, and it is now handled by software as well.

Thermal considerations

We’ve already seen the Raspberry Pi 5 offers a significant boost in performance compared to the Raspberry Pi 4, but we did so with a large heatsink and a PWM fan part of the “active cooler”. The fan is not particularly noisy, but some people may prefer to have a fanless system. Would that work with just the heatsink? What happens when we use the official Raspberry Pi 5 case? Let’s find out.

Note that usually perform testing in a room with 28°C ambient temperature which may impact the performance of the system. Just to show how ambient temperature may impact the CPU temperature I left the system run idly overnight (19:00 to about 10:00) just running sbc-bench.sh -m monitor temperature and rpi-monitor to draw the temperature chart.

overnight idle temperatureThe CPU temperature dropped from about 55°C to 50°C following the ambient temperature delta from about 26°C at 19:00 down to 21°C at 6:00 and going up from there.

Let’s check out the temperature chart starting from idle and running sbc-bench.sh script using the active cooler.

Raspberry Pi 5 Active Cooler Temperature Chart
Raspberry Pi 5 with active cooler

The idle temperature was about 46-47°C and went up to around 66°C during CPUminer. No throttling occurred and we are far from the 85°C limit, so the active cooler perfectly did its job.

The heatsink is quite large covering most of the board, so let’s disconnect the fan and see what happens:


Raspberry Pi 5 Temperature Fan Off

The idle temperature was quite higher at around 60°C, and we reached 85°C with both 7-zip and CPUminer, and throttling occurred with the CPU frequency dropping as low as 1,500 MHz from the usual 2,400 Mhz. That was only during short periods, so if your workload only requires bursts of performance, it should be possible to use the heatsink only. I’m also confident some will come with fanless cases for the Raspberry Pi 5.

Raspberry Pi also sent me the official case. So let’s remove the active cooler, stick the small heatsink from the kit on the Broadcom BCM2712 processor, and connect the fan…

Broadcom BCM2712 heatsink fan case

before installing the cover and reconnecting everything.

Raspberry Pi 5 case raspberry pi os bookworm

Time to run sbc-bench.sh again:


Raspberry Pi 5 temperature chart case fanThe good news that is CPU throttling did not occur. But the idle temperature is a bit higher at 56°C, and under load, it peaked at about 75°C. That also means the official case performs its cooling duty just fine, but the fan may be activated more often than with the active cooler because of the much smaller heatsink.

Now let’s turn off the system, disconnect the fan, and run the script with the Raspberry Pi 5 in its case without any active cooling.


Raspberry Pi 5 temperature chart case no fan

It does not look good at all. I waited for about 20 minutes before starting the benchmark, and the idle temperature was already around 67-68°C, but when we ran sbc-bench.sh the temperature quickly shot up to 85°C and even peaked at about 88°C.

The benchmark ended about 15:30 and I first thought there must be some background process keeping it at an elevated (85°C) temperature. But htop did not show anything…

Raspberry Pi 5 htop

sbc-bench.sh monitor mode also reported the high temperature and virtually zero CPU load:


It’s like the temperature sensor was stuck at around 85°C… and it would not come back down…

So I opened the case and reconnected the fan while the board was still running (probably not recommended) and the temperature dropped immediately…

Fan Broadcom BCM2712 CPU temperature

So if your plan was to use a tiny heatsink and put the Raspberry Pi 5 in an enclosure without a fan you can forget about it. That’s why Raspberry Pi provides an optional active cooler and the official case comes with an integrated fan, although it should perfectly be possible to create a fanless metal case for the board.

Power consumption

I measured Raspberry Pi 5’s power consumption with its active cooler using a wall power meter.

  • Power off – 1.7 Watts
  • Idle
    • 3.0 Watts (headless, WiFi only)
    • 3.6 Watts (Ethernet + WiFi, USB mouse & keyboard, HDMI connected)
  • “4K” YouTube Video in Firefox (full screen @ Full HD – avc1 (H.264) codec) – 5.5 to 6.2 Watts
  • “4K” YouTube Video in Chromium (full screen @ Full HD – avc1 (H.264) codec) – 5.7 to 6.8 Watts
  • Stress test on all 4 cores (stress -c 4) –  8.8 Watts

My Raspberry Pi kit comes with a USB PD power supply that supports 5.1V/5V or a little over 25W. The maximum in my test above is 8.8 Watts so that’s a lot of legroom when it comes to the power draw. So I tried to increase that by playing a 4K H.265 from a USB hard drive and run iozone on an external SSD while running stress -c 4 at the same time.

Raspberry Pi 5 Hard drive external USB

It did manage to peak at 16.8 Watts (and stabilize at 15.9 Watts), and so for that use case, a 5V/3V USB power supply as used with the Raspberry Pi 4 would not have been sufficient. For what it’s worth sbc-bench.sh does not report any drop in voltage under this specific stress test:


That makes me believe it should be possible to use a 5V/3A power adapter with the Raspberry Pi 5 for many use cases unless you are connecting USB peripherals that draw a lot of current like mass storage devices.  Raspberry Pi confirms that saying:

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.

I’m running out of time to test this in detail, however.

Conclusion

The Raspberry Pi 5 SBC is a great upgrade to the Raspberry Pi 4 if you feel the earlier SBC is sluggish with two to three times the performance depending on the workload. The Broadcom BCM2721 is also the first “Raspberry Pi” processor with Armv8 crypto extension catching up with the competition and delivering a 20 times boost in performance for AES-256 against the Raspberry Pi. If you are familiar with the Raspberry Pi ecosystem and want something faster, the Raspberry Pi 5 is almost a no-brainer.

Having said that, there are some changes to the layout that make previous cases incompatible, the MIPI CSI and DSI connectors are smaller and users must purchase new flat cables, a new power adapter is potentially needed, there aren’t any official fanless solutions yet, and some features like H.264 hardware video decoding, or any sort of hardware video encoding are now missing since the BCM2712 can handle that in software. That’s more flexible but also consumes more power.

More advanced users that are mostly vendor-agnostic may be disappointed with the performance and features compared to platforms based on Rockchip RK3588 processor with a similar price point, support for 8K video output, 8K/4K AV1/VP9/H.264 hardware video decoding, hardware video encoding, an AI accelerator, and more. Raspberry Pi hardware still probably benefits from a larger community and possibly better software support although the current Raspberry Pi OS Bookworm has a bit more bugs than I expected such as Bluetooth audio volume issues or a thick black bar under full-screen videos in VLC.

I’d like to thank Raspberry Pi for sending a complete Raspberry Pi 5 kit for review and testing. You should be able to purchase the board and accessories from any local distributor for $80 (8GB RAM as reviewed here) and up.

Share this:

Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress

ROCK 5 ITX RK3588 mini-ITX motherboard
Subscribe
Notify of
guest
The comment form collects your name, email and content to allow us keep track of the comments placed on the website. Please read and accept our website Terms and Privacy Policy to post a comment.
43 Comments
oldest
newest
Gnattu
Gnattu
6 months ago

The idle power consumption looks very bad. The board at its powered off state consumes more power than similar SBCs in idle state is…. ridiculous. The 3W idle is even more than some x86 systems.

tkaiser
tkaiser
6 months ago

> The board at its powered off state consumes more power than similar SBCs in idle state is…. ridiculous.

Was already the same with RPi 4 (1.2W when ‘powered off’).

Jeff Geerling
6 months ago

I just posted some more info that I _thought_ I had already posted to my sbc-reviews thread regarding the poweroff power consumption: https://github.com/geerlingguy/sbc-reviews/issues/21#issuecomment-1793836873

(The short of it is if you set POWER_OFF_ON_HALT=1 and WAKE_ON_GPIO=0, you can get it to 0.01W, or 0.09W with PCIe peripherals attached.)

Stuart Naylor
6 months ago