Release CPU stress test

[gunrock]

I had the same issue with heat and charging.


My CC with semi-broken wifi got 942mhz at OPP5 for about five hours (it was time to get up and dismantle the test site before the twins got up!).

 

[Linux-SWAT]

Three solid tests. One week per OPP.


All CC Pandoras:


Non-big USB


OPP: 3 4 5


MHz: 784 856 933


Non-big USB & non-bluetooth


OPP: 3 4 5


MHz: 773 852 920


Non-mini USB


OPP: 3 4 5


MHz: 763 841 911


I can confirm that at least one whole week is needed to have an accurate result.

 

[_wb_]

How much difference did the whole week make compared to the first result that was stable for 2 hours? Did it make something like 2MHz difference, or 20MHz, or more?


Also, are these 3 different units? What do you mean by "non-big USB" etc?

 

[Linux-SWAT]

It's about a 20 to 30 MHz difference. Huge.


Yup, 3 OP. They all have a default. Non-big USB means that the big USB port doesn't work.

 

[Linux-SWAT]

Another tested unit:


CC Noisy non-Wifi


OPP: 3 4 5


MHz: 784 865 931


On this one, the OPP3 was the fastest to reach the stable value of all units i've tested.

 

[Linux-SWAT]

On more report.

1GHz unit.

OPP 3 and 4, same maximum stable: 1198 MHZ

 

[Linux-SWAT]

For the first time, i had a "hang detected, aborting".

What is it ?

 

[Klumpen]

I would like to be able to overclock to 800 stable.

Is it safe to switch to OPP4?

I really could use the MHz that i am missing with OPP3.

 

[Linux-SWAT]

I would like to be able to overclock to 800 stable.

Is it save to switch to OPP4?

I really could use the MHz that i am missing with OPP3.

Well, test your unit...

860@OPP4 is a good start.

 

[_wb_]

For the first time, i had a "hang detected, aborting".

What is it ?

I get those when I'm doing other stuff while running the stress test, like heavy SD card transfers, wifi transfers or even just sound. I suppose it means the cpu stress test didn't get access to the cpu for too long (e.g. because it is too busy handling interrupts from that other stuff), so it has to do the test again to be reliable - it isn't really a stress test if the cpu is not being stressed. But I guess notaz can give the real explanation

 

[notaz]

It was designed to detect deadlocks due to overclocking-caused memory corruption and such, but I guess it's also useful for what _wb_ said.

 

[Linux-SWAT]

It happened while i did nothing, and also at a low near-stable frequency

 

[Linux-SWAT]

Another tested unit:

CC non-Wifi.
OPP:   3     4     5
MHz: 803 882 946

 

[Linux-SWAT]

I've noticed a "bug" on many of my tests.

The latest precise example:

(re)Tested my CC, OPP5=930, OPP4=854.

According to my own statistics, OPP3 will be below 800, so i begin to check at 810, to keep a little margin. One hour later, still at 810. What i had to do to trigger the errors, was leveling to 840, then errors appeared, and MHz quickly dropped to ~797. Now, three hours later, i'm at 790.

This behavior can happen at almost every OPP. Sometimes, errors just never show up.

 

[Binky]

Mine's settled at 798MHz @ OPP3 after 3 hours, if you'd like another data point. (a summer 2012 rebirth unit)

No overvolting

 

[Linux-SWAT]

So, an update to the CC i reported in post 82:

http://boards.openpandora.org/index.php/topic/10086-cpu-stress-test/?p=188099

OPP     3      4       5
MHZ   780  853   930       
 

 

[alucart1984]

What are the voltages for OPP3, 4 and 5?  What does OPP stand for anyways?  Getting my Pandora next week can't wait!

 

[urjaman]

"The set of discrete tuples consisting of frequency and voltage pairs that the device will support per domain are called Operating Performance Points or OPPs."

Eg. Officially the 600Mhz units will do 125Mhz @OPP1  (I  dont remember voltage, quite low), 500Mhz @OPP3 (1.2V) and 600Mhz at OPP5 (1.35V). I dont remember OPP1-2 or OPP4  (OMAP3530) voltages now off-hand, but generally here in the forums OPP refers only to the official voltage point without needing to remember the voltages. DM3730s (1Ghz units) have a different set of OPPs (and IIRC "max out" at OPP4).

 

[Binky]

So pandoras are 50% faster than datasheet specs? (500 vs 780) @ opp3

Where does that extra performance come from?

Does overclocking to 780MHz do (long term) horrible things to pandoras?

 

[alucart1984]

"The set of discrete tuples consisting of frequency and voltage pairs that the device will support per domain are called Operating Performance Points or OPPs."

Eg. Officially the 600Mhz units will do 125Mhz @OPP1  (I  dont remember voltage, quite low), 500Mhz @OPP3 (1.2V) and 600Mhz at OPP5 (1.35V). I dont remember OPP1-2 or OPP4  (OMAP3530) voltages now off-hand, but generally here in the forums OPP refers only to the official voltage point without needing to remember the voltages. DM3730s (1Ghz units) have a different set of OPPs (and IIRC "max out" at OPP4).

WOW!  Thanks for that answer. so basically overclocking would work like a computer.....the lower the volt the lower your max OC will be but lower battery draw.  Higher voltage = higher possible OC with more battery draw due to higher clock and voltage.  Interesting.  So once I have my Pandora will I be able to set each program to a specific clock?  So like with NES Emulation I would think a clock rate at 500Mhz or lower would be feasible but N64 would need more OC than stock.