Will Linux phones get decent processors in the future?

Hardware Librem 5
43

CPU Brute force test: performance:

Rasberry Pi 5 Broadcom 2.5Ghzx4
234.790 k/s

Librem 5 Freescale 1.5Ghzx4
446.688 k/s <---- L5 Winner

1 Like
44

Wow :open_mouth:

Is that because L5 is optimised?

1 Like
45

What command was used to produce the output in each case?

2 Likes
46

Yes.

aircrack-ng -S

But if you use another bench like sysbench L5 also win on some aspects.

1 Like
47

That doesn’t look right to me.

I ran that command on a Pi 4 (I don’t own any Pi 5 units) that has 4 x Cortex-A72 @‍1.5GHz, and it consistently produced ≄1130 k/s. I can only assume that a Pi 5 would do even better, or at least no worse.

If this is a Pi 5 that you own, I would check it for configuration and for thermal throttling.

FWIW, using the usage monitor on my desktop suggests that this is a multi-threaded test i.e. all CPUs were at 100% while the test ran. Admittedly that is x86. (However, as it happens, that makes no difference to the three-way comparison of these specific ARM chips, since they all have 4 CPUs, as long as it is consistently either single-threaded or multi-threaded with the number of threads equal to the number of CPUs.)

Edit: My Librem 5 (which I guess is the same as yours and is 4 x Cortex-A53 @‍1.5GHz) was consistently ≄430 k/s i.e. slightly slower than yours for some reason but not dramatically different. That may give an indication of a certain amount of variability in results - but nowhere near enough to explain what’s happening on your Pi 5.

2 Likes
48

Yes it is weird results, i not sure what going on as all the cpu(4) go full as shown htop in rasp 5.

49

One other comment: For the Pi 4 test, I ran the test via ssh. That is not ideal because then you are forcing the output to be sent across the network (and encrypted). Ideally, all testing would be conducted locally logged in on the device and for the test on the Librem 5, it was indeed done locally. Would it make a detectable difference? No idea.

50

On raspberry pi 5 i did the test local, in Librem 5 i did first remote(ssh) and later local on L5, and results on L5 remote and local is same.

However i did a bech on sysbench on Librem 5 and Rasp5 where Rasp5 win;

carlosgonz@raspberrypi5:~ $ sysbench cpu --threads=$(nproc) run
sysbench 1.0.20 (using system LuaJIT 2.1.1723681758)

Running the test with following options:
Number of threads: 4
Initializing random number generator from current time


Prime numbers limit: 10000

Initializing worker threads...

Threads started!

CPU speed:
    events per second: 10321.51

General statistics:
    total time:                          10.0004s
    total number of events:              103234

Latency (ms):
         min:                                    0.37
         avg:                                    0.39
         max:                                   10.13
         95th percentile:                        0.38
         sum:                                39979.13

Threads fairness:
    events (avg/stddev):           25808.5000/227.24
    execution time (avg/stddev):   9.9948/0.00

Are you on Crimson in L5?

Looks like a issue only in aircrack-ng bench.

2 Likes
51

No, byzantium (at the current time).

1 Like
52

The Steam Frame VR rig runs a Snapdragon 8 Gen 3 (SM8650) Arm SoC built on a 4 nm process, with an 8‑core CPU configuration. That’s Arm64, but with FEX it runs x86
 and with Proton it can even run (evil) Winbloze games through FEX faster than on native Winblows on x86 hardware. I dream of a Librem with that sort of performance.

I know, I know
 blobs and non-open hardware. But if Purism could sandbox it, those SoCs will be in production for a long time for the Frame and probably relatively inexpensive (rumor has it that the Frame will cost under $1000 and they will sell tons of them) and the work Valve put into optimizing battery life could be there for community use.

Just sayin’
 a man can dream. :last_quarter_moon_face:

2 Likes