Change In Mhz Ratings..

[craigix]

They were delivered ages ago, and they (so far) seem quite happy to clock to >720mhz. As said though, it is still nice to optimise code.

 

[sold]

Is it likely that TI had to test their boards to guarantee a higher clock speed in order to advertise it at higher specs, but it is the same as before in terms of the die? If not, why would they produce two versions of what is basically the same thing?

I'd have thought it would be detrimental to their (TI's) reputation and sales if they start pulling an intel, like "we disabled something or other so we can sell our full featured cpu at a premium later on, only to rip you a new one again in 3-6 months after that when our next thing that gives you a raging nerd boner comes out. Thanks for the cash."

That would suck. People already get upset as soon as they hear about the latest yet to be released tech as it seems to take the shine off what they have just bought.

 

[WizardStan]

Is it likely that TI had to test their boards to guarantee a higher clock speed in order to advertise it at higher specs, but it is the same as before in terms of the die? If not, why would they produce two versions of what is basically the same thing?

I'd have thought it would be detrimental to their (TI's) reputation and sales if they start pulling an intel, like "we disabled something or other so we can sell our full featured cpu at a premium later on, only to rip you a new one again in 3-6 months after that when our next thing that gives you a raging nerd boner comes out. Thanks for the cash."

That would suck. People already get upset as soon as they hear about the latest yet to be released tech as it seems to take the shine off what they have just bought.

It probably is the same die, but very minor imperfections in the casting of each chip might lead to some being stable at 720, and others only at 600. So they perform this extra test, as MWeston speculates, put the ones that clock to 720 in one bucket, and the ones that are only stable at 600 in another, and sell them separately.
There's a few schools of thought on this subject:
1) If TI takes out all the ones that test successful to 720, then those that order 600s will only get 600s. ie, no more overclocking on future Pandoras.
2) Even if that's the case, the 4000 were supposedly ordered and delivered a while ago, before they started cherry picking, so at least the first batch should still be overclockable.
My suggestion to the OP dev team: ship out the first batch as is. When the time comes for the second batch, consider upgrading to the 720s. If the 720s seem like a good deal, release a program that tests existing Pandora's for overclockability. If they cannot be overclocked beyond 720, offer a free exchange: ship back the 600 mhz only board, replace the soc, ship it home. If it can overclock to 720 (as most (if not all) of them should) then you need do nothing.
Alternatively, they can perform the test before shipping, and keep any that don't pass back for spare parts and demo models.

 

[pelrun]

IC manufacturers usually designates only a set number of chips to be stocked with the higher rating/price; once that number of parts have been reached the rest go into the lower bin regardless of their rating. That means that the lower specced parts may still be capable of the higher rating.

The entire PC overclocking community wouldn't exist if this wasn't the case.

 

[MWeston]

2) Even if that's the case, the 4000 were supposedly ordered and delivered a while ago, before they started cherry picking, so at least the first batch should still be overclockable.
My suggestion to the OP dev team: ship out the first batch as is. When the time comes for the second batch, consider upgrading to the 720s. If the 720s seem like a good deal, release a program that tests existing Pandora's for overclockability. If they cannot be overclocked beyond 720, offer a free exchange: ship back the 600 mhz only board, replace the soc, ship it home. If it can overclock to 720 (as most (if not all) of them should) then you need do nothing.
Alternatively, they can perform the test before shipping, and keep any that don't pass back for spare parts and demo models.

What, are you kidding me? TI tries to make an extra buck by binning the faster parts and suddenly we're supposed to take a huge loss if you can't overclock the chip anymore?

 

[darionapole]

i was thinking, why not make a program to test the maximum overclock of each pandora, and and then advertise them with the lesser value ?

 

[WizardStan]

What, are you kidding me? TI tries to make an extra buck by binning the faster parts and suddenly we're supposed to take a huge loss if you can't overclock the chip anymore?

Not at all. That's why I said to see if it's a good deal, first. If, by the time the second batch rolls around, the 720s are the same cost as the 600s are now (or possibly even cheaper) then you lose nothing in that batches upgrade. There's even the possibility they may scrap the 600s entirely, discovering that almost all the chips coming off the line are capable of 720 mhz, with the few that don't not reaching high enough quantities to sell in bulk. If that happens, are you just going to say "sorry, you lose" to the first batchers?
To keep things fair, you then have to assess whether it's feasible to upgrade the first batch. The existing ones should all clock to 720 already, if I'm understanding correctly, unless you've been very, very lucky. Supposing that as many as 5% are not able to reach the 720, that's about 200, which is approximately the excess number you're producing anyway, correct? I'm sure something can be arranged for those 200 so that they aren't a total loss. That was my "spare parts and demo models" comment, which was what my last impression of those 200 excess were intended to be anyway. I'm sorry if I misunderstood that much.
I'm hoping the final number of unoverclockable Pandoras will be significantly less than 5%, however. There's also a chance that a percentage of that percentage won't care whether they can overclock or not, bringing the number that would have to be exchanged down even further.
Again, these are just suggestions based on limited knowledge, all prefaced by "if it's feasible" (or even "necessary") which we won't know until the time comes along.
If I were you and you were me, I would put overclocking-to-800-mhz as part of the setup and testing process though, but I assume it's an easy thing to automate which, again, might be totally wrong.

 

[MWeston]

Not at all. That's why I said to see if it's a good deal, first. If, by the time the second batch rolls around, the 720s are the same cost as the 600s are now (or possibly even cheaper) then you lose nothing in that batches upgrade. There's even the possibility they may scrap the 600s entirely, discovering that almost all the chips coming off the line are capable of 720 mhz, with the few that don't not reaching high enough quantities to sell in bulk.

If they are separating the parts, then there will always be a price difference.

If that happens, are you just going to say "sorry, you lose" to the first batchers?

The first batchers are likely to get the fastest chips out of everyone since they were never separated and we have proven how fast they run already. Of course, don't forget that 600MHz is the only official speed TI or OpenPandora Ltd. has to guarantee. We don't make the chips. We only use them.

To keep things fair, you then have to assess whether it's feasible to upgrade the first batch. The existing ones should all clock to 720 already, if I'm understanding correctly, unless you've been very, very lucky. Supposing that as many as 5% are not able to reach the 720, that's about 200, which is approximately the excess number you're producing anyway, correct? I'm sure something can be arranged for those 200 so that they aren't a total loss.

Again, the parts are only rated at 600MHz. They will all do that with a Coke and a smile.

That was my "spare parts and demo models" comment, which was what my last impression of those 200 excess were intended to be anyway. I'm sorry if I misunderstood that much.
I'm hoping the final number of unoverclockable Pandoras will be significantly less than 5%, however. There's also a chance that a percentage of that percentage won't care whether they can overclock or not, bringing the number that would have to be exchanged down even further.

I hope they all overclock as well as they do today! We'll just to wait and see. It's no different than the GP2X or the Wiz where people try their luck and see how far it goes.

Again, these are just suggestions based on limited knowledge, all prefaced by "if it's feasible" (or even "necessary") which we won't know until the time comes along.
If I were you and you were me, I would put overclocking-to-800-mhz as part of the setup and testing process though, but I assume it's an easy thing to automate which, again, might be totally wrong.

I've read posts from you over the months and you have always come across as a highly educated person who gets it right all of the time and probably works in the industry. I'm just a bit floored by this comment though. Putting an out of spec, optional, never guaranteed qualification into the required functional testing makes zero sense.

It could be put into the functional test for sure. It's all software controlled, but failing a board because it won't do something considered a lucky bonus is absurd.

It is possible that nothing will really change for us anyway. These functional tests done by TI on the chips marked 72 are designed to run at 720MHz across the full temperature range but how often are you going to be playing games in 120F+ temperatures? The TI engineer I spoke with thinks it probably won't make a huge difference to us but the faster parts just offer a better guarantee to go faster in any condition. It's a numbers test. If certain tested values exceed certain thresholds, the chips get dumped in the 72 bin. I don't think they power it up at 720MHz and check twitter and then decide this one is a 72.

 

[gaterooze]

When the GP2X came out, it was rated at 200mhz. In the first batches, 99% of units could hit 250+ mhz, most commonly 260 mhz. Lucky ones, perhaps 20%, could hit 280+ mhz. Eventually, due to demand from potential buyers, craigix started selling GP2Xs tested at 280+ mhz at a premium price. Later batches had reduced "overclockability" and it seemed only 50% of units could get above something like 230 mhz, and hardly any above 260 mhz.

IMHO the Pandora should just be rated at 600mhz and anything above that is a bonus.

 

[BackAssward]

I agree with MWeston. Overclocking was never in the specs or a guaranteed feature. If it were, it wouldn't be called 'overclocking', it would be called the spec. The list speed is 600MHz, they just said 'oh and btw we were able to get higher speeds onj our test boards, isn't that cool' (paraphrase). It has already been mentioned that the ones from the factory might or might not overclock as well as the hand built ones, simply because the differences in the methods used to assemble them. That was a known factor already.

To expect a company to guarantee anything outside the list specs, is absurd. For example, my GP2X doesn't overclock as ell as others, it never crossed my mind that anyone would let me return it. My old desktop also doesn't overclock as well as some others of the same spec, I would be laughed at if I had tried to return it for that reason. Especially after having used it for months.

 

[Squidge]

It probably is the same die, but very minor imperfections in the casting of each chip might lead to some being stable at 720, and others only at 600. So they perform this extra test, as MWeston speculates, put the ones that clock to 720 in one bucket, and the ones that are only stable at 600 in another, and sell them separately.

I suppose it's depends on the yield and time to test. Having to test chips at both 720 and 600 is obviously going to cost more. Usually what happens is that the 720Mhz test is simply additional and you pay for that extra test in the price. The cheaper chips simply don't get that test.

I'd be very suprised if the chips that don't clock @ 720Mhz would be sufficient for TI's customers purchase projections. However, TI are obviously not going to tell anyone the percentage that fail. If it's low, you might find that TI simply throw them away rather than retesting them at another frequency.

At what OPP value are the chips testing at 720Mhz? I'm thinking the chips that fail may still be workable by increasing the OPP value.

 

[Trip]

Don't have a problem with this personally.
I paid for a Pandora that can run at 600Mhz and as long as they can do that speed on delivery, anything else is just gravy

 

[r00tw00t]

From the offical website [openpandora.org]:

ARM® Cortex™-A8 600Mhz+ CPU running Linux

Maybe you should change that to 600MHz without the "+" so people like me don't get confused and upset. Because that looks to me it can OFFICALLY be OC'd!

 

[BackAssward]

From the offical website [openpandora.org]:

ARM® Cortex™-A8 600Mhz+ CPU running Linux

Maybe you should change that to 600MHz without the "+" so people like me don't get confused and upset. Because that looks to me it can OFFICALLY be OC'd!

I'm willing to bet that they would guarantee 600.001MHz (or 601MHz if it can't do fractional incrementation). And it can 'officially' be overclocked, they wrote the official software to allow it. They are just not guaranteeing by how much, or what power level you need to use for it to be stable at a given clock rate.

Now if they said "ARM® Cortex™-A8 up to 720Mhz at the standard voltage" you might have a point.

 

[r00tw00t]

Ah, so it was some kind of semi-bluff to make it look good. Not saying it might only be overclockable by .001%.
That's business I guess and I should have looked it up myself to know about it. I don't think it's very honest though

edit: I also found this interview with Craig Rothwell: http://www.pocketgamer.co.uk/r/Various/Pandora/news.asp?c=8667 and I quote:
It is, without doubt, the most powerful portable games system ever and at the time of writing the most powerful mini computer, too. It has openGL 2.0 3D hardware, a 600-900Mhz CPU, media coprocessors, an 800x480 LCD and 128MB of RAM. The killer app with the hardware is that it is using an ARM Cortex A8 CPU, meaning astonishing battery life – over ten hours – which blows away similar mini PCs based on the old x86 chipsets. You will be able to play games like Quake III with ease and at the cutting edge even Doom III (if the engine goes open source soon, of course).

 

[peca]

I am sure, that at least Craig stated "up to 900 MHz" few times. I not to go search for it, as I early spot official "600+" parameter and accept "up to 900" as "only when lucky".

I think that for other series of Pandora will be better to use 720 MHz SoCs. This ensure that they are as good overclockable as 1st serie. But this is on OP team decision and responsibility.

 

[BackAssward]

...a 600-900Mhz CPU.

Correct, it's maximum Mhz is somewhere between 600 and 900. If he was giving the actual range of the CPU, he would have said a 1-900 Mhz CPU, since it can be underclocked too. Regardless of what you may think, it was never guaranteed to overclock to a specific clock rate. Just that the ability to overclock will be available. 600Mhz+ means 600Mhz base frequency with the ability to overclock.

The 'official' site has the official clock rate, period. Then there's (on both official and unofficial sites), I don't see 900 mentioned anywhere on the spec list:

- openpandora.ca: Texas Instruments OMAP3530 processor at 600MHz (officially).
- gbax.com: ARM® Cortex-A8 600Mhz+ CPU running Linux
- openpandora.de: ARM® Cortex™-A8 600Mhz+ CPU mit Linux
- openpandora.es: Procesador Texas Instruments OMAP3530 a 600MHz, con núcleo ARM® Cortex™-A8
- openpandora.eu: ARM® CortexTM -A8 600Mhz+ CPU
- pandoratr.com: ARM® CortexTM -A8 600Mhz+ CPU
- openpandora.jp: テキサスインスツルメント社OMAP3530プロセッサ600MHz

 

[WizardStan]

It could be put into the functional test for sure. It's all software controlled, but failing a board because it won't do something considered a lucky bonus is absurd.

Ack, sorry, I misspoke. I didn't mean that if they couldn't overclock they'd flat out fail, but rather that you can just set them aside so that, hopefully, only overclockable ones go out. If there's only five or ten from the first batch that can't get over 600 mHz, that seems an acceptable "loss" (air quotes because you just find a different use for them) to be able to proudly announce that they can all overclock (at users own risk) to 720 or 800mhz. Client enablement and marketing hype and all that. Even if as many as 200 can't overclock, you may be able to pull something off. Maybe, I dunno. That's a lot of demo models.
On the other hand, now that I've had some sleep, it occurs to me that you won't know that there's only 5 or 10 until the end (or can't make an accurate guess until reasonably well into the shipping) and it may turn out that fully a quarter or more aren't overclockable, and by setting aside those, it creates an unfair balance between those that ordered early vs late from the first batch; it creates the situation I was hoping to avoid in a different place. Quite a conundrum.

 

[tsh]

I don't see any way that a functional test can qualify a specific part for overclocking, without the resources that TI have probably put into developing a test which is suitably sensitive to the way in which these chips fail to function at speed. The tested speed isn't the only factor which affects if you will see a problem, temperature, power supply and maybe even precise function all have an influence.
It's a shame that the spec change is tied to a new part number. It should give us a better feeling about overclocking to know that the same chip (presumably) can test faster, but there doesn't seem much point getting too excited until we've seen what 600MHz feels like, and how the first batch react to being overclocked. Decisions for batch 2 can be made after the first 100 or 1000 have shipped, but I think OpenPandora have always been open about the risks of overclocking and setting expectations for what software should rely on.

 

[S-r-ex]

I myself will only find the OC possibility as a bonus. Maybe I'll do it once I've gotten the hang of it on my desktop.