Yeah, it's strange that a such highly integrated SOC doesn't have something simple as a temp sensor. C'mon, it even has TV input!
Sad, I'll I have my finger IR sensor. I don't think it gets so hot, at stock speeds it barely gets warm.
I'm tired of disassembling it!
Before running benchmarks, remember to kill phoenix process! (screen will freeze, don't panic)
pidof phoenix | xargs kill -9
To make it easier, so you can just type "stop" to kill phoenix process:
echo "pidof phoenix | xargs kill -9" >/usr/bin/stop
chmod +x /usr/bin/stop
sync
I wrote a small Pi calculation programCopypasted some code from internet and adapted for my needs, making a Pi benchmark.
Cross-compiled to arm, worked great.
My idea was, being very smalll, it should avoid cache misses and show the real cpu performance difference.
There was very little difference between O1, O2, O3 and Os optimizations,less than 1%.
From O0 it was 10% faster. I had to compile with static libraries, as it complained about LIBC libraries version.
And it's very obvious:
# ./pi 1000000
Starting PI...
Pi: 3.141592653589794004176383168669417500495910644531250000000000
Loops: 1000000
Time: 9932mS
# ./pi 1000000
Starting PI...
Pi: 3.141592653589794004176383168669417500495910644531250000000000
Loops: 1000000
Time: 5626mS
Almost 200% cpu power!
In this case, DDR speed made no difference, clearly stating my cache supposition.
# ./pi 1000000
Starting PI...
Pi: 3.141592653589794004176383168669417500495910644531250000000000
Loops: 1000000
Time: 5624mS
Benchmarked it for 10 minutes, no stability issues. Running
top showed 100% cpu usage
# pi 100000000
Starting PI...
Pi: 3.141592653589794004176383168669417500495910644531250000000000
Loops: 100000000
Time: 568621mS
However in memory operations the DDR performance boost is huge:
# time dd if=/dev/zero of=/dev/null bs=1M count=8192
8192+0 records in
8192+0 records out
real 0m 35.62s
user 0m 0.02s
sys 0m 35.59s
# time dd if=/dev/zero of=/dev/null bs=1M count=8192
8192+0 records in
8192+0 records out
real 0m 24.76s
user 0m 0.00s
sys 0m 24.76s
I used the scope for few hours with no issues.
[Irony on] With this cpu power, you could start mining bitcoin and it would self-pay in about 10 years! [Irony off]
I made a video. Notice how the FFT update rate almost doubles with the overclock!
`
Now I'm figuring out how to write to the display buffer.
We need DOOM running in this!