I don't know who started this while true bogus. This does NOT load cpu to max it only tricks it into staying at peak frequency. If this worked, all stress torture test apps would be out of business.

Contrary to popular belief, while true only does a JMP or a JE or JNZ or whatnot. A CPU is able to do a lot more complex instructions, with longer paths. Note the complete lack of cache, scheduling, swapping, memory access and whatnot from "while true". It is better suited to measuring idle consumption than full load.

If you want to know more about CPU load, look up tests ran on desktop OC machines. Start with LinPack from Intel, and continue through small and large FFT as well as Dhrystone and whetstone. Finally, Prime95 and SuperPI for limited test cases.

Additionally, above is a guide and does not compute to the watt. It is, however accurate to a decent 10 percent or so.

If you want to test CPU consumption, someone try a small buffer FFT, then we talk.

Oh one more thing:

Not all processors are equal. There is such a thing as an intended frequency where the darned thing is more efficient than the rest.

There is also a variation in manufacturing that separates man from beast. If they'd be the same it'd be too easy. And make overclocking contests very, very boring.

5 pages... no answer yet....

op: set the max to 500mhz and run an emulator

my criticism was about the fact that nokia engineer who knows a lot from HW. he has warned that overclocking may cause problems that can not be noticed. and there is nothing that proves that tweaking the cpu really makes device cooler.

yeah. well, what he said is nothing new. it's a well known possible downside to overclocking beyond manufacturer specification. but you also have to understand that chips are usually made with much higher tolerences. it's always guaranteed to run faster up to a certain point with negligeable problems. the fact that my one can run at 1.1ghz all day, watching movies, taking calls, browsing, gaming etc tells me it can handle quite a bit. I don't however run it at 1.1ghz all the time just in case as i cannot control the cooling level.

technically, it (cpu) will run cooler. not at 1+ghz but at 850mhz using ideal voltages, it should run cooler than stock voltage at 600mhz or have negigeable difference. Higher frequency does not mean it has to be worse for the chip. all depends.

anyways. considering Nokia said that even their stock kernel settings is not good for the N900 due to them overvolting, my undervolted cpu may make it last longer although either way, it should last us a good few years. And some other component will die before the cpu i'll bet.

according to the Nokia engineer Igor Stoppa, temperature is not the real issue (there seem to be safeguards) but overvoltage and data corruption:
http://talk.maemo.org/showpost.php?p...3&postcount=66
http://talk.maemo.org/showpost.php?p...&postcount=904

I have been overclocking all kinds of cpus and gpus for the past 15 years and have never had a chip die as a result.. Most of the overclocking have involved higher voltage as well.

The OS will always crash before anything burns.

I'll be amazed if it's different with the CPU in the N900..

I'm fairly certain that the chip is has the voltage and the frequency it has due to a compromise to make sure it's stable even if a device gets a not so good chip.

Off cause it can cause problems that'll not be noticed, like corruption, only way to check this would be a burn in test, but that could be something a bit dangerous without additional cooling I guess

Sure..

Not true. Shell does a lot of things doing that. In case of C/C++ it could be 'yes, it's true' but not for interpreter language. Especially for busybox which even doesn't precompile script, I think. It processes script line-by-line.

[QUOTE] A CPU is able to do a lot more complex instructions, with longer paths. Note the complete lack of cache, scheduling, swapping, memory access and whatnot from "while true". It is better suited to measuring idle consumption than full load.[/QUOTE}

Again, if you speak about "true load" you right. But if we speak about rough measurements of power consumption comparison of different frequency it is enough. And cache + scheduling are in business. (I hope swapping - not)

I'm pretty sure the load of an interpreter interpreting "while true" doesn't challenge a CPU (agreed, it's not a JZ, but still). Percent load in an OS is nothing more that the inverse of cycles remaining free and with an application demanding unlimited slices it's 0%. It's not loaded.

I disagree with the fact that the same load provides an accurate benchmark. For example, at 250 MHz, while true gets 100 cycles, at 500 MHz you get 200 cycles (maybe), which is great to determine frequency. Cool.

However, should I overvolt the little bugger, you still get 200 cycles. Nothing is determined this way save for frequency.

When a real load is applied, things start to happen. Voltage drops (and is droop corrected on most machines), heat is being generated, components lose conductivity, resistance increases, and more power is needed. As a result, you get a different response curve for a script versus the heavy stuff.

As I said, I have an amp meter hooked up to my work CPU. The difference in load between a script loop, a heavy executable (Photoshop) and Linpack is visible on a 3 cm needle when switching frequency. Granted, this is a powerful and complex machine, more than the ARM we have in N900.

But I'm getting WAY off-topic even from my initial post. I'll wrap it up.

It is my opinion as a rather old overclocker that:

a) Nokia set the said limit most likely to keep the CPU from heating up and/or eating an inordinate amount of power. This has been carefully considered.

b) While it's probably overclockable to a decent degree with little risk during normal usage, hard conditions pose a risk to the CPU.

c) Heat protection is great to have, but is not the end of the story. By the time thermal shutdown kicks in, components are stressed by heat as they increase in temperature locally. This may or may not be enough to cause damage through repeated roasting.

d) Executing a "while true" in an interpreter at room temperature gets you no meaningful result if you ever forget the N900 on a car dashboard, running a graphics demo over wifi.

Take these, as always, with whatever size of salt grain you deem necessary. It was never my intention to hijack a thread on TMO and turn it into an overclocking thread. Sorry about that :)

No, that was me -- "this Nokia Guy" corrected me.

No, that's not the 3430...

It doesn't do that. We went over this on the other thread....