Frying An Arm Chip

[Kramy]

This warranty stuff should be simple:

Overvolt -> Not covered.
Internal mod -> Not covered
Overclock -> Covered

Nobody seems to think overclocking without messing with the voltage or OPP will cause a problem, so lets make that the official policy?

 

[fischju2000]

Those old ARM and other Palm CPUs were built with the 130 nanometer process, and they generated much more heat than the A8 does. It can't be compared.

http://www.gp32x.de/board/index.php?/topic/49301-change-in-mhz-ratings/ - This thread has a lot of info on those power profiles. As does this one: http://www.gp32x.de/board/index.php?showtopic=42953&st=11

For all I've read this is still true

It might reduce the life of the CPU but it would still last for many years.

To avoid significant device degradation for commercial temperature OMAP3530/OMAP3525 devices (0°C < Tj < 90°C), the device power-on hours (POH) must be limited to one of the following:
· 100K total POH when operating across all OPPs and keeping the time spent at OPP5 to less than 23K POH.
· 50K total POH when operating exclusively at OPP5.
· 44K total POH with no restrictions to the proportion of these POH at operating points OPP1 - OPP5.

OPP5 is 1.35v. I believe devs have gotten around 750-800mhz at OPP3, with 900mhz needing OPP5.

More threads with a lot of dev-talk on the issue:
http://www.gp32x.de/board/index.php?/topic/46002-overheating-cpu/
http://www.gp32x.de/board/index.php?/topic/46069-when-could-i-get-one/page__view__findpost__p__687085
http://www.gp32x.de/board/index.php?/topic/43293-is-this-something-to-be-concerned-about/page__hl__overclock%20mode

 

[Polossatik]

Let's chill out for a moment and look at some figures.

The "commercial grade" 600mhz package you guy's are using is rated to be used from 0 ° celcius to 70 ° celcius

From "OMAP3530/25 Applications Processor SPRS507F / FEBRUARY 2008 / REVISED OCTOBER 2009" , page 124

This seams to indicate that at 600Mhz that package can dissipate enough heat at 70° "ambient" to be "safe".

If you look at page 255 you see 2 values : OMAP35 30/25(CBC Pkg.)
-> Power (W) 0.92871
-> RθJA (Theta-JA) = Thermal Resistance Junction-to-Ambient, °C/W -> 21.89 ° C

Which basically means the package heats up 22 °C for each Watt consumed.

Now, the first question is how much Watt does a "soft overclocked" package uses (let's assume at the point it's still stable)
I would guess to see a modest non-linear watt figure until it's unstable, so let's take max 1.5 watt for around 750-ish MHz.
which means you need about 30 °c - ish lower than 70° C "ambient temperature" (=temp inside the case) to get that amount of heat gone.

So basically if you fully overclock the pandora, put it at 100% cpu and dsp and then put it in the backing sun, you might get into troubles without "upping the voltage".

Not going to happen fast.

The only reserve about above is the pop memory package that might add more heat,so I would suggest to have a chat with some TI engineers about this before further assuming stuff.
If you would now simply say "I'll look into it and make a statement after having the facts sorted out" then I doubt anyone can blame the OP team.

edit: doh, need some sleep, confused junction temp (90°C) with ambient temp (70°C)

 

[Asiyura]

The "pencil trick" was an overclocking technique pretty frequent on AMD CPU.
Those CPU where the same model, they tested the different CPU parts and their stability at different frequencies.

If the CPU passed all the tests without problems, they connected two soldering points on it to allow it to run at full speed.

If it ran all the tests without failures but failed to stay stable at max frequency, they would not connect those two points.
So the CPU would be sold as a lower frequency and cheaper.

The game was to use a pencil to make the electrical connection (by drawing it on the board) between the two soldering points to be able to use the full potential power of the CPU.
(it was less dangerous than playing with a soldering iron on a CPU...)
Sometimes, no problems would appear, but sometimes, you would have many crashes.

 

[Dead1nside]

I know you're talking about ARM chips, but with limited experience of overclocking x86 if you set the FSB too high it'll just not boot or it won't be stable enough, start making errors and crash. It'd be the same for ARM I think, except ARM chips don't run as hot as x86. As I said in the other, heated thread, Squidge (I believe?) comment that he ran it at 500-600Mhz then at 800Mhz for an equal amount of time and measured only a 1 degree centigrade difference was encouraging as I think you can posit that 1 degree won't harm the chip due to heat, especially an ARM chip.

I found this by Googling, might help.

If you take a look at the latest OMAP3530.pdf datasheet, page 103 you will read TI's recommendations on voltage and the effects on total part life.

It looks like they don't recommend any overdrive on the DSP core which is probably good since it can pull some huge current loads under maximum use.

The ARM core has more flexibility for overdrive. Standard operating mode is OPP3 which is where TI guarantees 600MHz. We have seen 625 work fine, but 650 fails so TI is pretty much bang on with their promises.

The tests at 900MHz are at OPP5 which is maximum overdrive and will come with a disclaimer that your chip life will suffer if you leave it there all the time. I would consider it more of an experimental mode and when the SGX drivers are working fully and supporting the GPL, I'd rather see a focus on getting the clock speed back down to 600MHz or less for as much as possible.

So, if I overclock, I'll only get 44,000 hours of operation?
That's only 5 years!
I am depressed...

No, if you just overclock, you'll still get guaranteed > 100Khours.

However, if you then use OPP5-6 to make the overclocking more stable, then the hours will drop depending on how long the device spends in OPP5-6.

From everything I'm reading in the threads linked to by fischju2000, which are great, is that overclocking without changing from OPP3 is not detrimental to the *lifespan* of the chip. I'm no expert but some of the guys talking in those threads are certainly more qualified, and it appears to be coming from their interpretation of the documents. From comments by people who've had contact with the Pandora boards it appears that they don't have trouble getting to 720Mhz (the ones tested). Perhaps qualifying any warranty in terms of not altering the power profile or the voltage is a good way to do this. Everything I read seems to suggest that this will not have a bad effect on the lifespan of the chip.

 

[fuzzypinkunicorns666]

According to TI's datasheet for the OMAP3530, running it at the max voltage (1.35v, referred to as OPP5/OPP6) halves the expected lifetime from 100k hours to 50k hours (which is still over 5 years of constant usage) and no other adverse effects are noted other than not being supported in the extended tempurature versions. So at least according to the datasheet, overvolting it seems like it should be okay as long as you don't use it in >90C environments and you don't have a >5 year warranty. Unless you can increase the voltage even higher than the datasheet mentions, that is.

NOTE: I am not an electrical engineer and claim no expertise on the matter.

 

[Grench]

This does bring in another twist... I don't see how this question is going to get answered without some trial (and potential error).

Heat:
The main thing that can kill chips is heat. Even if the processor will run stable at 800Mhz outside the case in a bench set up, that does not mean that it is healthy for it to run that way stuffed into a pocket case with no ventilation. To test if the OpenPandora console can hold XXXMhz stable without frying, the test would pretty well have to happen in an OpenPandora case, right?

Testing:
Is there a temperature sensor in, on or near the CPU? If there is a temp sensor on the ram (as rumored) then a unit could be set up to run at 500, 600, 720, 800, 880, 940, etc... and build a temperature curve. The TI processor spec should have an allowable max operating temperature. Figure out what Mhz correlates to what die temperature, subtract 10% and call that the 'Maximum supported overclocking rate'.

Application:
Set the sensor in software to a safe threshold - if it hits that threshold it shuts the system down. If you let the end user changes that threshold, then they're voiding the warranty. Don't allow over volting. Keep it simple. This would also help guard in case someone sets the lid switch to not shut it down, closes it, sticks in in their pocket or other insulated scenario. If the sensor doesn't exist, then scratch this thought.

And this leads us to...
The Jealousy RMA:
If a Pandora gets shipped that can not overclock to X (720Mhz or 800Mhz) and stay stable then is it faulty and subject to RMA/replacement? You know the drill - someone is going to get one that only works at the advertised 600Mhz and feel inferior to someone else who was able to get 800Mhz stable and get all in a huff about it.

************
Slight deviation - Everyone keep in mind just how much/many things this device can do WITHOUT over clocking it. Overclocking does not need to be done in order to have fun with or enjoy the device. I don't know about anyone else, but I plan to use/enjoy the device 'out of the box' and push it's stock limits to where I feel constrained before I even consider overclocking it.

 

[Patrick R Ludvigsen]

longquote

Oh yes, someone else found this, I was thinking about this the whole 2 threads.

 

[kingoddball]

Hey Craig: Why not test an MP unit yourself - until the point of total failure?

Get one of the Pandora's that you have made (which WE will have the SAME SoC on our Mass Produced units) and see how it runs at 800-850mhz permanently.

Get an MP unit knock it up to some high MHZ and run it for a day or two straight. Just put a few movies or leave Mupen running on a loop or something to test it under heavy load.

Using such a generic question of "Can an ARM chip be fried from OC'ing" is far too vague. andora2ut4:

All we see on video's are Dev running their devices at 800mhz+ and it's an absolute tease IF we cannot.

I want to run my Pandora at 800MHZ+ (same as Devs have doen) when playing N64 and other high end apps. It will most likely be underclocked when using normal desktop or watching movies, etc.

 

[Poem58]

i think all of the links are dead, but still trying to show it was the old Intel's or early Pocket PC ones...

Can I overclock my PDA?
WARNING...If you don't know what Overclocking is, then it is probably not for you. Though in the end, this should make your PDA go much faster, it can ruin it. Since you are voiding your warranty by trying this, you will not be able to get your device fixed without paying, should you have problems in the future. Casio E-115 -

​

Like you said the links don't work, and the warning is just that - a warning. We saw this in the early GP32 days too.

But was it actually possible? We need to see some real evidence. Surely with all those devices out there someone would have documented something? There is tons of stuff about frying X86 CPUs.
​

the problem is Craig, it was so long ago I can't find anything. But I KNOW it happened but maybe it wasn't a CPU meltdown? maybe something else and it was assumed? After all mali said he could no longer turn it on. Does that Have to be CPU death? Or could that be something else?. I know there were reports of it killing your PDA just as much as I know the first time I kissed my wife, but I can't find a link for it, so I cannot validate the reason. It was about 7 years ago. I would guess the old forums were not kept for posterity once those PDA's became antiques. I'll look some more. if it will satisfy you, and someone can find me the software, I will do it to my Axim. I really don't want to lose it, but if it's the only way I can prove it...
nah, I'll look some more for a link or something.

edit : The thing we're missing here is despite not being able to find a link, that is a good thing. It just goes to show that there isn't any easily searchable way to find evidence that an arm processor has melted down in the past 7 years. Not enough so that anyone has bothered to post it in any significant numbers. I think this is what Craig might have been really looking for, the fact that you can't seem to find evidence, not that my old Axim CPU is possible to fry. Clever old boy! (as they say in the movies)​

 

[kingoddball]

@ Poem58: You could very well be right. There is not much evidence even as low as no evidence (xScale intel SHIT does not count).

Did Craig mean this? I does not seem like it though.
Seems like he is seriously trying to get us to find instances of ARM cores being damaged by OC'ing. I don't care about what's been done on older chips.

I WANT my Pandora to be able to run at 800mhz+ the SAME as the Dev units. I want to play some N64 and other programs that use higher speeds.

 

[twobit]

I can't seem to find much on the dangers of overclocking, only that "it's dangerous to overclock". Maybe suggestions that life expectancy (for the device) is lower, which I imagine is true if it's operating above recommended settings. My Intel E8600 is overclocked to 400x10 @ 1.3v (3.3ghz to 4ghz), but I don't epxect it to stay healthy for a long time. It'll still last a number of years if I'm taking care of it, but yeah, I don't expect it to live forever working hard for me.

Following is an article about a Motorola Droid being overclocked to 1ghz. It's an ARM Cortex A8 processor that's stock 550mhz. I can't see that they're using any voltage mods, just bumping up cpu speeds using software. There's probably more to it, but I'm not having any luck finding anything. There's no source code for the application they're using provided. I see that it's reliant on kernel versions for the software.

http://www.redmondpie.com/how-to-overclock-motorola-droid-to-1ghz-9140391/

Seems that the ARM processors are still ahead of the game compared to Snapdragon in the Nexus Ones.

I jumped in before reading the entire thread, so I'll go back and see what else is posted. Probably got it all sorted and my post is redundant/useless.

Cheers,
Twobit

P.S.
I want my....
I want my Pandora
- sung to the tune of Money for Nothin' by Dire Straits

 

[kloplop321]

I just know that any arm device I OC works when it works, it may get warm, but it WILL CRASH before anything BAD happens. that is my experience and I am sticking with it.

 

[frostedfires]

I just know that any arm device I OC works when it works, it may get warm, but it WILL CRASH before anything BAD happens. that is my experience and I am sticking with it.

I agree with you thats all that happens when I use them... Craig u shouldn't worry too much about it because the chance of someones to break within a year because of OC is super duper uber rare IMO. Also almost the whole community wants to OC their pandora so if you tell us we can't OC it then a lot of customers will be dissapointed to do something that isn't realy dangerous.

 

[zounder1]

I'm going to go out on a limb and say testing one Pandora to some O/C level such as 800 is utterly useless statistically. You might by fluke test on a machine with a more robust chip that can handle the load, while the majority may not handle this fine. Or it might be indeed true every chip can handle that amount of O/C. Testing one machine most certainly will not determine this. A reasonable sample size of 5-10% would be required to get anything approaching statistically significant. Are 5-10% of the folks in the order queue ready to sacrifice their order in the name of true useful testing? I highly doubt it.

So I would strong discourage OpenPandora from doing an O/C burn in on just one unit and publishing the results... given the complaining and short fuses that seem prevalent lately, you'll just end up getting attacked later when someone gets a unit that can't hit 800 or fails when they run it up to 800. And even if you clearly state that results may vary, I guarantee you'll get someone complaining about their machine not O/C like the test and demanding a refund/exchange.

I personally am horrified at the way a lot of folks here act. Don't give them more fuel to flame the fire by making any promises about the ability to overclock.

 

[alucart1984]

Heres another question to add to the thread. How would you even know someone overclocked their Pandora? How would one be able to tell a Pandora was ever or is overclocked if it is dead? Just a thought. I agree with what ever someone said earlier, if TI warrants over clocking then so should OP but if not then OP shouldn't either.

 

[Patrick R Ludvigsen]

This reminds me of the chaos with Asus N10s

 

[Gary13579]

I don't even really get this topic. Take a devboard, overclock it without overvolting as high as you can go, run some extreme computational program on it, and throw it in the oven @ 350F. Take it out and see if it's cooked yet.

 

[TrashyMG]

I don't even really get this topic. Take a devboard, overclock it without overvolting as high as you can go, run some extreme computational program on it, and throw it in the oven @ 350F. Take it out and see if it's cooked yet.

add some peppers and onions....

 

[WizardStan]

I've spent the last hour trying to track down anyone that had overclocked their NDS and had it catastrophically fail, to no avail. There's reports of getting it up to 1.8x clock speed and the worst that happens is it crashes during some games (possibly corrupting save games, or your file system for homebrew). Reboot and the system restarts as normal. I can find no reports of it failing. Same with the GBA overclock mods. This is simple overclocking by replacing the crystal, however, and not overvolting.
The only information I can find at all about overclocking any ARM processors are success stories, including a few voltmod ones (a lot of GP32 ones, actually)