Frying An Arm Chip

[petelewin]

But truthfully the question does dig a bit deeper than just simple difference in MHZ. Since the reality may be that there is not quintessential difference between "our" 600mhz chips and TI's 720mhz chips. Since it may just be an arbitrary label applied to squeeze extra revenue out of the same chip fabrication process. Or it could mean the chips are actually tested to run at 720, and the 600's will only clock up to, but not quite 720. The fact is, we don't know. There is no way to know, and TI won't tell anyone because it would defeat the purpose of the labeling to tell anyone.

True, we can't know for sure - but the chances are that these are speed binned. That does not mean that the 600MHz parts will not run above 720MHz. Firstly to be classified as a 720MHz part it will need to run at 720MHz plus margin at the worst case temperature (well above where you are likely to use them in a Pandora) and worst case voltage. Secondly, the % of parts that test above 720MHz is unlikely to match the % of the market that will pay for higher rated parts.

 

[mali]

^Also keep in mind that no binning existed yet when they bought the 4000 SoCs for the first batch.

 

[petelewin]

^Also keep in mind that no binning existed yet when they bought the 4000 SoCs for the first batch.

Or at least they hadn't announced the 720MHz parts.

 

[yodaofborg]

^Also keep in mind that no binning existed yet when they bought the 4000 SoCs for the first batch.

Or at least they hadn't announced the 720MHz parts.

Errm, you must be new because we already had the *binned* chip discussion, if I recall correct the chips were binned at 500mhz, correct me if I am wrong.

 

[mali]

There was just one kind of 3530 when they bought the SoCs for the first batch. The 600MHz certified version. The 720MHz version was introduce mid(?) last year.

 

[WizardStan]

Errm, you must be new because we already had the *binned* chip discussion, if I recall correct the chips were binned at 500mhz, correct me if I am wrong.

You are wrong. You're right, we've had the binning discussion a few times before, but it is always "maybe this or maybe that."
It will be impossible to know what is really going on without asking TI, and they'll never tell. We can make educated guesses, and once people get theirs and start overclocking (warranty concerns aside) we'll have a little better idea as to what might be the reason, but nothing will ever be known for certain.

 

[yodaofborg]

Yeah, I guess you are right Stan.

 

[Dervish]

Specifically don't know how a chip would burn up before it would crash. But to find out what that crash point is you'd need to test each chip (process, voltage and temp considerations).

Easiest way I'd consider for doing that is to run a program which just runs though loads of multiply accumulate operations (likely to be the critical path for the chip) and slowly increments the processor speed. Now the problem would be know at what frequency it crashed at. I'd guess that when the pandora crashes it freezes rather than reboots(is possible it does though). However without knowing the details would the screen freeze in a crash or not (even if it was connected to TV out)?

I'd hope (for the purposes of this) that it would freeze allowing the user to read off the crash point and then input that (with a safety margin) on reboot.

If there is a thermocouple on the chip then obviously linking that into the program would be great to stop damage (maybe even predict the crash). Sorry if this rather obvious method has been proposed before but this is a damn long thread!

EDIT: Actually forgetting what this is, surely could link it via usb to another comp which captures updates on the current frequency then when it breaks just read off the last one?

 

[petelewin]

Easiest way I'd consider for doing that is to run a program which just runs though loads of multiply accumulate operations (likely to be the critical path for the chip) and slowly increments the processor speed. Now the problem would be know at what frequency it crashed at. I'd guess that when the pandora crashes it freezes rather than reboots(is possible it does though). However without knowing the details would the screen freeze in a crash or not (even if it was connected to TV out)?

I'd hope (for the purposes of this) that it would freeze allowing the user to read off the crash point and then input that (with a safety margin) on reboot.

The nature of a crash due to overclocking is going to be unpredictable. The processor deadlocking is quite likely (does OMAP3530 have a watchdog?), but you may also get corrupt data, resets, or just about anything else. I don't think exercising all the possible critical paths is as easy as running loads of MACS either.

The 'crash point' could be recorded by writing the frequency at the end of each successful trial to a file. To be safe it should close the file after each write, and open it again after the next successful trial.

 

[petelewin]

There was just one kind of 3530 when they bought the SoCs for the first batch. The 600MHz certified version. The 720MHz version was introduce mid(?) last year.

Agreed. I meant that TI could still have been speed binning, and storing the fastest parts. They could have done this to get more data on the yield at different frequencies before selling a different frequency part. A quick search however suggests the 720MHz parts are all rev D, so maybe not.

 

[Exophase]

while the majority of pandora owners happily run their unit at 600MHz/OPP3, happless few pandora owners get random crashes a few times a day at the cited clock/OPP. are those pandora owners entitled to an RMA? or will you go ahead and tell them the chip maker rates OPP3 at 500MHz and guarantees nothing for 600/OPP3, ponentially leaving the guys with the impression they got lemons (even though technically they did not)? perhaps you can advise them to go with OPP5 (more on that below)? i guess we can both agree that by being liberal with overclocking, OP put their product above the level of TI support.

The question has never been whether or not the warranty should guarantee functionality at an overclock, obviously it shouldn't. The question is whether or not the warranty should be voided if you do overclock. 600MHz should not be considered an overclock based on Open Pandora's product specifications, so it's on them to accommodate this (and yes, any failure to do so should be covered by RMA). What I was trying to say is that if random crashes are dangerous enough to void the warranty then we have a serious problem, because any program can potentially randomly crash without first being overclocked.

If OP has to pick between OPP5 at 600MHz and decreased life span vs OPP3 at 600MHz and instability then I would think they'd have to go with the former. They promised 600MHz, which implies that the thing performs at that. They didn't promise anything about power draw or lifetime outside of "will work for a year", and can feel free to defer to the TI stated usage for that.

my bad, somebody mentioned 600MHz was 'considered stable at OPP3', and i did not bother to re-check with the specs. OPP5 is clearly a state of maximal allowable overdrive, which means even though the chip maker considers the device to be stable at that OPP (mil-grade devices excluded), it practically halves the chip's projected lifespan when used at 'normal' load levels ('power-on hours' really does not say much here). now crank that usage level to extreme - make it reqularly crunch while being roasted on your car's dashboard daily, and that POH figure drops to god knows what. surely, the device will likely last its warranty time, but how many user would knowingly spend it in this manner?

So where do you draw the line exactly? We're talking about an operating range up to 90C; that's what the power on hours are speced over. I have a hard time imagining that ambient temperatures up to 40C at the most would do an awful lot to tip it over the breaking point, and I think it'd be more of a matter of it failing to work at all past this point rather than having a diminished life span. I doubt any kind of feasible environmental conditions will do much to reduce the operating hours of the chip.

 

[nikkopt]

Quote from another thread:

You could use eFuse's to tell whether the cpu had been overclocked or not.I doubt the OMAP3530 has that though.

I was reading somewhere that the OMAP chips do have efuse, but not sure if it can be programmed in anyway to check for overclocking.. yet, plus with access to the JTAG a person who knows what they are doing can just reset the switch.. eFuse isn't a permanent fuse.

To use the JTAG you have to solder wires on the pads, therefore losing the warranty

 

[TrashyMG]

Quote from another thread:

You could use eFuse's to tell whether the cpu had been overclocked or not.I doubt the OMAP3530 has that though.

I was reading somewhere that the OMAP chips do have efuse, but not sure if it can be programmed in anyway to check for overclocking.. yet, plus with access to the JTAG a person who knows what they are doing can just reset the switch.. eFuse isn't a permanent fuse.

To use the JTAG you have to solder wires on the pads, therefore losing the warranty

Why would you need to solder wires? get some Thin Wire-wrap (30 AWG) wires and some tape would do the trick, not the best wiring solution but nice and temporary.

By no means I'm saying I'd do this..

 

[atomicthumbs]

To use the JTAG you have to solder wires on the pads, therefore losing the warranty 

With some pins and a clamp you could turn it into a little land-grid array thing.

 

[kingoddball]

Why not just say warranty aint void if you overclock? Seems simple and everyone will be happy.

NOTHING to do with Voltages. Just a simple software overclock sanctioned by Open Pandora Ltd - Maybe add a little footnote of cannot over clock higher than 800mhz?

 

[WizardStan]

Why not just say warranty aint void if you overclock? Seems simple and everyone will be happy.

NOTHING to do with Voltages. Just a simple software overclock sanctioned by Open Pandora Ltd - Maybe add a little footnote of cannot over clock higher than 800mhz?

Because they haven't decided that yet. We shouldn't tell OpenPandora what they should or should not accept for RMA.

 

[nikkopt]

To use the JTAG you have to solder wires on the pads, therefore losing the warranty 

With some pins and a clamp you could turn it into a little land-grid array thing.

That crossed my mind too but it would leave scratches on the copper. And you can always put a sticker on top of the pads saying "Warranty void if removed", the ones that you cannot take off without ruin them.
Stickers that change color with water are also a nice idea, in case of water damage..

 

[Monk]

By no means I'm saying I'd do this..

Based on the "who will rid me of this troublesome priest?" ruse, that would seem to be "What's your order number?" material. I'd tread carefully if I were you... "I wouldn't reocmmend doing it but you COULD con OpenPandroa into a free 'out-of-warranty' repair/replacement by..." doesn't seem to go down too well.

 

[⬡]

It's a bad thing to point out how, if a given method were used to detect tampering, it could be defeated?

 

[Dimacus]

There is one thing that has bothered me.
It has been said that you won't be able to change OPP by hand, that probably means that it will ship set at OPP3.
This means that to reach the advertised 600Mhz you would need to overclock. (And this is outside TI's spec, so there is no guarantee it will reach that high, even if it is higly likely.)
But what really bothers me is that there is no way to change to OPP1 or OPP2, in order to save battery life.