OMAP5432 at home !


pardon my ignorance, but there was a way to run the CPU at 0.5 GHz?  (i assume that it would require less power, right?)  if we get proper 3D acceleration, that would still probably be plenty for certain games/activities...

since it's a kernel thing, that means you'd have to reboot to get to these other frequencies, right?

The values that you can set the frequency to is defined in the device tree, it could always be added back by editing the device tree source and compiling a new device tree binary file.. You can easily change frequency on the fly with this command ex: sudo cpufreq-set -f 1.5Ghz , however it needs to be one of the frequencies defined in the device tree.  It's a bit less feature friendly than how we do it on the Pandora as on the Pandora we can set it to any speed within a range.


 
 
Last edited by a moderator:
I would imagine you could still set the clock speed to anything you like,  but without that OPP you won't trigger a drop in operating voltage by dropping it that low.

Or at least that's what I understood that to mean, based on how the OPP stuff works on Pandora. Am I misunderstanding something?

-Neelix
 
I would imagine you could still set the clock speed to anything you like,  but without that OPP you won't trigger a drop in operating voltage by dropping it that low.

Or at least that's what I understood that to mean, based on how the OPP stuff works on Pandora. Am I misunderstanding something?

-Neelix

Notaz share this earlier:
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/arch/arm/boot/dts/omap5.dtsi?id=05e7d1a57a268b491d0e24952d219a73c62bd3df
It is a part of the device tree setup for the omap5... speeds are hard limited to the values set in this device tree setup.
 
It is very easy to modify the device tree to add back the "missing" values. You do not even need to recompile the kernel binary, just the device tree (DTC is available separately in Debian). And you may be able to use "device tree overlays".

Then, the cpu-freq tool can under- and overclock the CPU. If you are very courageous, you could even invent a 700 MHz or 1200 MHz or even 2 GHz OPP...

The bigger problem is if these modes can harm/damage the CPU. In the best case the kernel or user space or e.g. GPU will simply hang up sporadically. In the worst case the operation voltage does not match and adds thermal stress.

This is what the "descoped" probably means, i.e. TI has not further studied their chips to approve additional operation points.
 
oo, did not know about cpufreq stuff.  should be useful in other pursuits!  thanks for the info all.
 
I wonder how low we can set the voltage. With 500MHz, it should still be as fast or even a bit faster than a 1GHz Pandora, so lowering the voltage would improve the battery life, which would be great for stuff like classic console emulation.

That said, we don't know yet how well powersaving of the OMAP5 works. Maybe it's unnecessary to lower the voltage and speed to 500MHz or even lower.
 
I wonder how low we can set the voltage. With 500MHz, it should still be as fast or even a bit faster than a 1GHz Pandora, so lowering the voltage would improve the battery life, which would be great for stuff like classic console emulation.

That said, we don't know yet how well powersaving of the OMAP5 works. Maybe it's unnecessary to lower the voltage and speed to 500MHz or even lower.

Well that is essentially what it used to do... when it ran at 500Mhz, it lowered the voltage to 880mV (880000 uV). I suppose lower voltages could be tried. 

 

diff --git a/arch/arm/boot/dts/omap5.dtsi b/arch/arm/boot/dts/omap5.dtsi
index 3bfda16..a4ed549 100644
--- a/arch/arm/boot/dts/omap5.dtsi
+++ b/arch/arm/boot/dts/omap5.dtsi

@@ -45,7 +45,6 @@

 

operating-points = <

/* kHz uV */

- 500000 880000

1000000 1060000

1500000 1250000

>;
 
Last edited by a moderator:
I wonder how low we can set the voltage. With 500MHz, it should still be as fast or even a bit faster than a 1GHz Pandora, so lowering the voltage would improve the battery life, which would be great for stuff like classic console emulation.

That said, we don't know yet how well powersaving of the OMAP5 works. Maybe it's unnecessary to lower the voltage and speed to 500MHz or even lower.

500MHz Pyra is like 100Mhz Pandora? Sounds nice. Is this because of the 2 cores of the OMAP5? Well, if the target of 10hrs usage per charge is meant at max Speed (1.7 GHZ) then it could lead to alot more battery time for average tasks. :)
 
Last edited by a moderator:
setting the clock speed would be a great option for the Pyra overlay.

500 MHz - Ultra Low Power [if it won't break the device]

1.0 GHz - Low Power

1.5 GHz - Performance

1.7 GHz - Overkill [if it won't break ...]
 
Last edited by a moderator:
I wonder how low we can set the voltage. With 500MHz, it should still be as fast or even a bit faster than a 1GHz Pandora, so lowering the voltage would improve the battery life, which would be great for stuff like classic console emulation.

That said, we don't know yet how well powersaving of the OMAP5 works. Maybe it's unnecessary to lower the voltage and speed to 500MHz or even lower.

500MHz Pyra is like 100Mhz Pandora? Sounds nice. Is this because of the 2 cores of the OMAP5? Well, if the target of 10hrs usage per charge is meant at max Speed (1.7 GHZ) then it could lead to alot more battery time for average tasks. :)
Actually benchmarks I've done the OMAP5 @500Mhz is faster than my 1Ghz Pandora Overclocked to 1280Mhz. It being dual core does play a big part, but generally the A15 is a better, faster core.
 
Last edited by a moderator:
Is that with multi-threaded benchmarks?  What about single threaded?  Two OMAP5 CPUs @500Mhz outperforming one OMAP3 CPU @1280 suggests one OMAP5 CPU @1000, which isn't really that much of a performance boost.  The A15 boasts 3.5 DMIPS per MHZ (up to 4.0) while the A8 only claims 2.0, and while benchmarks may not actually give us the advertised 75-100% increase I would still expect more than 28%.
 
Is that with multi-threaded benchmarks?  What about single threaded?  Two OMAP5 CPUs @500Mhz outperforming one OMAP3 CPU @1280 suggests one OMAP5 CPU @1000, which isn't really that much of a performance boost.  The A15 boasts 3.5 DMIPS per MHZ (up to 4.0) while the A8 only claims 2.0, and while benchmarks may not actually give us the advertised 75-100% increase I would still expect more than 28%.

Benchmarks is poor wording... I was just timing compiling a program several times, with different amount of jobs(so multi-threaded) with the same compiler flags on the OMAP5 @500Mhz and the Pandora fully overclocked... If you or someone else could suggest something more comprehensive that even some chob like me could setup then I could do a better study.
 
I tested make and make -j4 on a x86 dual core, make -j4 was almost exactly 2 times faster so I think it's a reliable test.

Anyway, there's the taskset command to select core usage if you want to test more precisely.
 
I used these values in the dtb :

1290000, 1700000

Values I got there :

https://github.com/omapconf/omapconf/blob/master/arch/arm/mach-omap/omap5/opp54xx.c

Cpufreq listed up to 1.7GHz :

available frequency steps: 500 MHz, 1000 MHz, 1.50 GHz, 1.70 GHz

I did a make -j4 to force full CPU usage but I only reached 1.5 GHz.

cpufreq stats: 500 MHz:82.65%, 1000 MHz:5.18%, 1.50 GHz:12.17%, 1.70 GHz:0.00%

Here's the thread on IGEP forums :

https://www.isee.biz/support/are-these-1-7ghz-values-correct

Did you try "cpufreq-set -g userspace; cpufreq-set -f 1700Mhz" ?
 
Back
Top