Frying An Arm Chip

[Gary13579]

Sorry, my ADHD is kicking in extremely bad and it's hard to pay attention/read. But if it's possible to fry the A8 via OC/OV, then it should be clearly easy to test. If the majority of chips don't overheat, and instead just crash, then clearly any Pandora's people fry will be defective units.

The key thing is OCing to the point where it doesn't crash, then abuse the hell out of every chip in the Pandora. Listen to music over Bluetooth while downloading a 20gb file over WiFi while computing pi to 5 billion digits while playing speakers at max volume while charging with a USB device plugged in pulling 500mA. Keep in mind the Pandora has zero ventilation, so it needs to be in a case or similar object.

 

[Elanzer]

It's not going to happen, a very large majority of people who go on about the dangers of overclocking on the internet either: do not know how to overclock and/or have never done it _properly_ , have limited technological knowledge in the first place, or simply pass on word from other people like the above thumping the dangers without citing any reliable sources of people actually frying their hardware and under what conditions.

Not to say the dangers of overclocking are not there if you do something stupid like jack up all the voltages to breaking points for no reason, but any modern CPU with internal thermal throttling is a very big challenge to kill with overclocking while using in-spec voltages, and a ARM CPU that generates next to no heat is even less of an issue.


I hate to sound like an elitist, but 95%+ of people on "overclocking forums" are near-clueless on the issue, the same is going to apply to the people here who say that the ARM CPU will get damaged with no sources to cite and likely no proper experience to cite.

edit: Also with overclocking embedded devices, you also have to consider that there can be other issues related to "bricking" a device with overclocking. It's possible that setting an overclock can be simply unstable on a piece of hardware (ie: it works for a couple seconds, then suddenly the device is "fried") and the piece of hardware had saved that overclock the moment it is applied, always booting up with an invalid/unstable setting, effectively "bricking" the device with overclocking but NOT ACTUALLY DAMAGING THE CPU.

There are many people who have "fried" ARM/MIPS CPU based routers with overclocking, but simply because they set an invalid/unstable setting and the device will no longer boot for the setting to be changed back, however using a JTAG cable to reset the firmware would fix them.

 

[atomicthumbs]

My GP2X went to 310 megahertz and ran a stress test for ten minutes. It got fairly warm IIRC, but it didn't break until quite later until I tried to upgrade to Open2X and botched it.

 

[DaveC]

There is allot of talk here about how ARM can handle OC and there should be no problem. The problem with that is that this TI chip is not just an ARM, there are a whole lot of other things in there (GPU etc) that MAY fry out even if the ARM itself is fine. If the ARM survives but the GPU or DSP blow,then you still have a borked SoC.

The only way to know for sure is to TEST a few units with a thermocouple attached and to measure the amp draw (voltage may remain the same but current can spike) while running code that is working out all sections of the chip.

 

[Sphinxter]

There is allot of talk here about how ARM can handle OC and there should be no problem. The problem with that is that this TI chip is not just an ARM, there are a whole lot of other things in there (GPU etc) that MAY fry out even if the ARM itself is fine. If the ARM survives but the GPU or DSP blow,then you still have a borked SoC.

The only way to know for sure is to TEST a few units with a thermocouple attached and to measure the amp draw (voltage may remain the same but current can spike) while running code that is working out all sections of the chip.

Like I always say, "Science Demands!".

 

[trooper]

Just out of interest Craig, how many gp2xs have you had returned which have been fryed by overclocking? (I know you send them to GPH to test why they failed)

I'm not aware of any, I don't think the chip can get hot enough. I guess if you started to fuck about with the voltage by hand (ie doing wacky hardware mods) you could kill it.

That said it was a pretty common GP32 mod.

Yep, And i still have that GP32 Blu, Still pencil modded, Still use it, Clocks to 208mhz stable, And still working 100%.

It was a guaranteed 166mhz GP32 Blu, Purchased from you back at the end of 2004 if i remember rightly.

Trooper

 

[skeezix]

I've let my mk2 dev board sit running at 900mhz all night (with a book on top of it to try and keep some heat in), no problem. but thats hardly a useful test. (need ot have it running video decoding in software or something on the opengl tap to get it worked up a little, and cover it with glue to keep all the heat in

Charging the pandora makes the battery warm, much warmer than it gets from running at 900.

The question is not so mucdh _frying_, but killing it before it would have died normally, however..

And usually its overvolting which leads to overheating and death in my experience, but I will leave it to experts.

jeff

 

[Tom`]

The question is not so mucdh _frying_, but killing it before it would have died normally, however..

Good point, but TI's info says 100,000 hours at OPP3 or 50,000 at OPP5 - as long as this is over a year, then a 1-year warranty is reasonable, right? (50,000 hours is around 5.7 years of continuous use.)

 

[Alpha2]

Hi, layman here.
Thought that crossed my mind:
Assuming the thing is overclocked and as many say it simply shuts down. On the surface this seems like it'd be harmless but is it also possible that even though it crashes that it sustains some form of diminished capacity or loss of power over time?

Not suggesting in any way that this is the case since I dont know anything about engineering, I just wonder if that's not something that might be put in the worry basket while we're in the neighborhood.

 

[WizardStan]

Hi, layman here.
Thought that crossed my mind:
Assuming the thing is overclocked and as many say it simply shuts down. On the surface this seems like it'd be harmless but is it also possible that even though it crashes that it sustains some form of diminished capacity or loss of power over time?

Yes, it does. The TI specsheet suggests that running it at stock voltage will get you about 20 years of nonstop, 24/7 standard usage out of it; overvolting it will drop that down to about 5 years of nonstop, 24/7 overclocked usage.

 

[Jonttu]

I have a friend who has OCed his Droid to over 1.3 Ghz. He's been running stable for a couple of months now. I'm sure it is possible to fry a chip. With too much voltage anything is possible. I'm pretty the only thing overclocking will do is increase temps just by a wee-bit, decrease battery life, and shorten the overall life span of the device. Besides I'm sure the chip by itself has a safety mechanism to turn off if the coretemps or voltages are too high.

 

[Mr B]

Doesn't it all depend on TI, as supplier of the OMAP parts? If they will guarantee the OMAP parts to OpenPandora when they've been overclocked, OpenPandora should be happy to pass that guarantee on. If TI will not, then OpenPandora should not - regardless of any suspicions we may have about the silicon suppliers motives.

Thats not the way it works. You buy chips. After that your on your own. Warranty will be provided by Craig and the others, meaning it will go out of there pockets.

B!

 

[dentrado]

As fischju2000 posted earlier:

OMAP35x Extended Temperature and
Reliability Information:

To avoid significant device degradation for commercial temperature OMAP35x 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.

from TI: http://focus.ti.com/asia/download/TechDay09tw_OMAP_track_1_OMAP35x_Anderson_Hsiao_TI.pdf

So according to TI even overvolting is supported but will shorten the lifespan of the Omap, however it will still last for longer than the warranty. (As Tom said 50000 hours is 5.7 years non-stop usage.)

 

[mali]

I think the general problem lies elsewhere. You have to take all kinds of heat generators into account. So if you set up a test Pandora, use an empty battery, connected to mains, external harddrive on the USB port streaming stuff, running a Quake 3 demo loop. Ambient temparature maybe 35°C. Voltage and core clock to maximum. If it doesn't break after a few weeks, it should be totally risk free.

Such a test should be made anyway before shipping to spot problems, imo.

 

[sold]

How 'bout we wait until the team release a utility that will allow us to safely overclock/underclock? That way, we all get the device, the devs can run tests and then they can give us the app and the go ahead (or not). If you choose to mess with the clock speed prior to being told by the OP team if its ok or not and you kill the CPU, you get no warranty. Also, anyone overvolting or overclocking past the safe zone (somwhere between 750-900Mhz?) will be voiding their warranty by disregarding the safe overclock and then they get nada. Naaaa-daaaaa!

But it seems to me from what I've read that the A8 is a resilient little bastard, so under a reasonable overclock it seems like it will be ok. But then again, Im no expert.

 

[ldesnogu]

There is allot of talk here about how ARM can handle OC and there should be no problem. The problem with that is that this TI chip is not just an ARM, there are a whole lot of other things in there (GPU etc) that MAY fry out even if the ARM itself is fine. If the ARM survives but the GPU or DSP blow,then you still have a borked SoC.

I think the general problem lies elsewhere. You have to take all kinds of heat generators into account. So if you set up a test Pandora, use an empty battery, connected to mains, external harddrive on the USB port streaming stuff, running a Quake 3 demo loop. Ambient temparature maybe 35°C. Voltage and core clock to maximum. If it doesn't break after a few weeks, it should be totally risk free.

This is exactly what I had in mind: if you just overclock and run some test that exercises only the CPU, you have only checked a small part of your chip. *All* of the SoC has to be stressed at the same time and this includes at least: integer core, VFPlite, NEON, SGX, C64x+. If you don't stress it this way, the test is useless. As far as I know no such program exists in the wild, any volunteer to develop one?

 

[Magic Sam]

Hi,

Slightly off topic:

Judging from the reply I received from WizardStan on that other thread, it seems there is something I didn't quite understand:

What is the real "default" speed (without OC) of the pandoras we will get ?

Is it 600MHz, or is it 720MHZ as written on TI website ?

[/off topic]

Bye and thanks, Magic Sam

 

[mali]

The default speed is 600MHz.

 

[Tor]

Weren't the Xscale processors mentioned in this thread manufactured before RoHS came into play? My understanding is that RoHS soldering (lead-free) is a bit more sensitive to temperature changes and could be worse affected by high temperature - i.e. you should be more careful these days. But don't take my word for this - I have no reference atm. to point to.

 

[peca]

Is it 600MHz, or is it 720MHZ as written on TI website?

AFAIK, chips was 600 MHz "certified" (or class), but default speed will be set to 500 MHz. Sorry, I have read it more that once, but I haven't note any source.