x264 Benchmark and ARM64

Table of contents

x264 Benchmark

A few days ago, GAT posted he bought a Raspberry Pi 4 B to run Pi-hole1 on it.
Before that, he wanted to test his x264 Benchmark and built the needed components on his RPi4B, where he used the current ARM64-image.

Well, compiling x264 on ARM64 brings the ARMv8 NEON instruction set extensions, which speeds up the benchmark alot.
On a BCM2711B0 (1.500 MHz) with ARMhf-image a x264 Benchmark run was done in 4h 54m 14.302s, where my BCM2711C0 (1.500 MHz) with a ARM64-image only needed 4h 20m 47.592s, which is roughly 12 % faster.

Time to get my Prometeia out of the basement, again…

Setup

Since I knew, my CM4 was able to clock at 2.500 MHz with a little bit of cooling, the x264 Benchmark is a bit more challenging, so I had to go to sub-zero temperatures.
I decided to use my CM4 instead of a RPi4B, since it was already on my table and I wanted to try the mounting of my Prometeia on it.
Turns out, it’s pretty easy. A few zip ties and a little bit insulation later, the system was ready to start.

If you see this, the image is broken :(

A new record

I started the CM4 and waited for the Prometeia to cool down.
This time I could mount the evaporator directly on the SoC, without any copper extension I had to use of the normal RPi4B.
It resulted in a lower temperatur, which was about -44 °C, instead of -36 °C. Of course, the BCM2711C0 is naturally cooler than the B0, but on idle the type of SoC should be irrelevant.

With -44 °C on idle I overclocked the SoC to the currently maximal speed of 2.500 MHz (over_voltage=12) and started the benchmark.

If you see this, the image is broken :(

If you see this, the image is broken :(

After one crash in the middle of the first pass, the next try was successful and it finished the run in 2h 53m 24.375s.2

If you see this, the image is broken :(

Condensation

I was lazy and used only a little bit insulation around the CM4 itself.
On the underside of the IO Board I saw condensation. Enough to know I should have insulated it with a mat of Armaflex or something else.

If you see this, the image is broken :(

When the 2.500 MHz-barrier is lifted and I have to use my dry-ice cooling-pot, I definitely will insulate the whole thing, that’s for sure.
Or maybe my current delivery from ElmorLabs3 will solve this problem, too.


  1. Pi-hole ↩︎

  2. x264 Benchmark Thread at VoodooAlert ↩︎

  3. ElmorLabs ↩︎