Change In Mhz Ratings..

[rv6502]

We could also consider that TI know they all work at 720mhz, and are just extracting more money out of them.

yes, they probably were waiting to make sure they had sufficient yields at 720Mhz before making a real 720Mhz part, and the old 600Mhz part number is likely in order to keep suppling old orders even tho they might all be 720Mhz parts.

if your job is to order "600Mhz TI OMAP3 parts" and all of the sudden there's only 720Mhz parts, you gotta go to your manager and ask permission to change the order, who ask his manager (who's in vacation) to ask his manager (who's in a convention) who ask back "can't you order 600s from someone else?" and back down the line, and then goes back up "no",
and back and forth,
and you got a simple problem that takes 1 to 2 months to resolve. :rolleyes:


as to why they don't have a 900Mhz part yet:
if the yields are say

10% defect
10% max 600
50% max 720
30% max 900

you can sell 90% of your production as 600Mhz, 80% as 720Mhz, but only 30% as 900Mhz

if all of the sudden everyone orders 900Mhz parts, you only get 30% yields.

if the 900 part "does not exist", you can sell 80% to 90% of your production which might yield more profits than a very high price for a 900Mhz part that no one would buy, or a low price that you cannot supply enough parts to fulfill orders.

 

[Monk]

Whatever it is, we are still just guessing and look at the amount of posts...

Well, yeah. If we had all relevant facts what would there be to speculate about? That's a sure way to kill a thread

If you want a thread to thrive, the BEST way is to leave plenty to gossip about!

 

[Enverex]

Oh, and upon further inspection, the 720MHz specified chips have to be ordered as a separate part number. They have the suffix 72 at the end of the part and they cost a lot more! I'm asking my favorite TI hardware engineer what this will mean for us Pandora folks in the future since we've had no trouble with these speeds already.

Did you hear back from your friend yet?

I'm hoping it's a case of TI realising all chips could do these levels without any changes and that all the Pandora's (at least from the first batch) will be able to as well. When overclocking the main ARM CPU, does that automatically overclock the C64x? If not, has anyone tried overclocking the C64x itself yet? (as apparently that does higher speeds too).

 

[sold]

@Squidge: Dude, you may have missed my point a tiny bit. I said for dosbox, I am relying on overcloking to get the desired results. And due to the fact that I grew up on dos games, it is my favorite emulator so I would love to be able to cheat and overclock to get slightly better results.

I am not regretting pre-ordering as the Pandora is still the only portable device that has all the features I want. My point was that there have been numerous statements made regarding overclocking and they have mostly been to the tune of "should overclock to 900mhz easily". The news of TI binning their chips makes those claims shaky to say the least. Particularly the claim of "600+mhz" now.

Anyone who may have read my previous posts should not take them as criticism against the Pandora or its devs. Frustration is to blame and I apologize for any rudeness on my part.

 

[Squidge]

if all of the sudden everyone orders 900Mhz parts, you only get 30% yields.

Or you just sell them the 36x part which is available as a 900Mhz part

 

[Tensuke]

@Squidge: Dude, you may have missed my point a tiny bit. I said for dosbox, I am relying on overcloking to get the desired results. And due to the fact that I grew up on dos games, it is my favorite emulator so I would love to be able to cheat and overclock to get slightly better results.

I am not regretting pre-ordering as the Pandora is still the only portable device that has all the features I want. My point was that there have been numerous statements made regarding overclocking and they have mostly been to the tune of "should overclock to 900mhz easily". The news of TI binning their chips makes those claims shaky to say the least. Particularly the claim of "600+mhz" now.

Anyone who may have read my previous posts should not take them as criticism against the Pandora or its devs. Frustration is to blame and I apologize for any rudeness on my part.

It's likely that a good bit of the first batch Pandoras will be able to OC to at least 720MHz. It sounds like TI is just now doing tests at 720MHz and selling the ones that don't pass as the standard 600MHz. Since OP already ordered the chips, the ones they have weren't tested at 720 so there's no reason for a good number of them not to make at least 720MHz--after all, why else would TI market chips with a higher clock speed if there wasn't a significant amount of chips that could do it?
As had been said however, don't expect anything more than the specs on openpandora.org, but I'm willing to wager that there will be a lot of OC'able Pandoras. I dunno, maybe all this waiting has made me hopeful.

 

[Kramy]

I'm guessing TI tests 10000 SoCs from time to time to find out what voltages they require to run 720mhz.

When they finally hit 100% of them able to reach it on the default voltage, they created a new SKU and bumped the price up.

I'm not really surprised. I bought a cheap Athlon II (2.9ghz), but it runs 3.55ghz on the default voltage. For years Intel's low end CPUs(sub-$200) have been overclockable past their fastest ones.

It's just a ploy to extract more money out of customers. It doesn't mean anything about the quality of the SoCs.

 

[S-r-ex]

But isn't this regular doing in the processor business? They see how well the chip performs, sorts them out by how well they do, label the series that clocks the highest whilst being stable as the "flagship", waits a few months, and release a new series clocked even higher, branding them as the "new flagship" with a higher pricetag. It appears to be happening in both the Intel and AMD camps.

 

[tsh]

They are choosing to push their product a little harder and are guaranteeing a slightly higher speed, but it is sounding more and more like they haven't changed a thing. This "grading" thing stinks. We are not buying a freaking steak here. This isn't something that is grown naturally where differences in quality are to be expected due to variables (environment/genetics/etc.) where it is justifiable that product can be graded and sold accordingly. This is a high-tech precision part manufactured in an environment that is designed to eliminate variables which create undesired effects and TI should offer better quality control than what it sounds like they are offering currently imho.

It is basically quality control that they are charging for.
Fuck that. They should be testing ALL the chips to verify that they work properly.

You could read up a bit on semiconductor manufacturing and process variation a bit before getting too excited. Imagine you're trying to bake cakes so they rise to 4cm +/-1mm. that would be much easier... Each stage in the processing of a wafer makes parts of it faster or slower, and the variation is factored into the design pretty much by trial and error. Too fast or too slow in critical places can make a difference to max speed, sometimes two errors can add in opposition and cancel out. Ever heard of growing silicon ingots? the imperfections start there!

 

[rv6502]

if all of the sudden everyone orders 900Mhz parts, you only get 30% yields.

Or you just sell them the 36x part which is available as a 900Mhz part

lol, right

my example numbers were off but it still explains the marketing strategy behind not selling your low-percentage highest-speed yield.
Intel and Amd can sell rarer "extreme editions" since the cpu on the motherboard is user-replaceable, its a bit harder to do on embedded systems

 

[Kramy]

But isn't this regular doing in the processor business? They see how well the chip performs, sorts them out by how well they do, label the series that clocks the highest whilst being stable as the "flagship", waits a few months, and release a new series clocked even higher, branding them as the "new flagship" with a higher pricetag. It appears to be happening in both the Intel and AMD camps.

Not really. It used to be that way, but not so much anymore. They find out how fast their processors can go, and find the highest pricepoint they can sell them at. Then they scale back performance and price to allow market saturation at different pricepoints.

When Intel released their low-end sub-2ghz Core 2 Duos, there was no reason to believe any of them would do less than 2.6ghz. Most easily managed 3.2+ ghz, in the hands of a semi skilled overclocker. These are basically high end CPUs sold as low end ones, because otherwise Intel wouldn't have buyers.

When all the CPUs in a generation can outperform the fastest model with a bit of overclocking, regardless of what speeds they are sold at, you know it's just the company trying to saturate the market at different pricepoints.

 

[sold]

You could read up a bit on semiconductor manufacturing and process variation a bit before getting too excited. Imagine you're trying to bake cakes so they rise to 4cm +/-1mm. that would be much easier... Each stage in the processing of a wafer makes parts of it faster or slower, and the variation is factored into the design pretty much by trial and error. Too fast or too slow in critical places can make a difference to max speed, sometimes two errors can add in opposition and cancel out. Ever heard of growing silicon ingots? the imperfections start there!

Fair enough, but what I was getting at without going into the technical aspects of it is they are using a proven manufacturing process. That should mean a certain amount of consistency in the finished product. If a part doesn't work as intended because it has faults that exceed the fault tolerance for the part, do they still use it but sell it as a lower clocked unit? And wouldn't determining if finished parts are within tolerance be part of quality control in the first place?

 

[Monk]

If a part doesn't work as intended because it has faults that exceed the fault tolerance for the part, do they still use it but sell it as a lower clocked unit? And wouldn't determining if finished parts are within tolerance be part of quality control in the first place?

The SOCs OpenPandora bought were intended to be operated at 600MHz and "possibly faster with overclocking". They ARE all working "as intended" if they work at 600MHz but not at 720 MHz. 720MHz isn't "as intended" for the Pandora SOCs - it's better than the spec. Unless you get (and pay for) the newly-badged 72 version - which seems to be the same SOC from the same process, but tested to work at 720MHZ (and may, of course, work at higher clock rates).

 

[Pleng]

How many times have words and phrases such as 'probably', 'possibly', 'may', 'I guess', 'should', and 'could' appeared in this thread? :huh:

 

[Monk]

How many times have words and phrases such as 'probably', 'possibly', 'may', 'I guess', 'should', and 'could' appeared in this thread?

About five per post. I guess

 

[Tensuke]

Not counting my post*:
Probably: 21 (20.8% of posts mention it)
Possibly: 4 (3.96% of posts mention it)
May: 18 (17.82% of posts mention it)
I Guess: 5 (4.95% of posts mention it)
Should: 36 (35.64% of posts mention it)
Could: 33 (32.67% of posts mention it)

*NOTE: These numbers include multiple uses per post and in quotes, and therefore does not distinguish between them. Unless someone wanted a more statistical breakdown.

By the looks of it, I guess, the first batch Pandoras could possibly overclock a bit, while some may not, and the new '72' chips should probably not affect future Pandoras.

 

[Username]

So I'm confused, does this mean that all boards are verified to go to 720 Mhz or are the ones that don't make it that high up sold as only 600 Mhz? That would be a real shame.

 

[Squidge]

So I'm confused, does this mean that all boards are verified to go to 720 Mhz or are the ones that don't make it that high up sold as only 600 Mhz? That would be a real shame.

We don't know yet. All we can do is guess. You'll just have to wait and see. Theres a possibility for both cases for the first batch.

For the second batch, probably the later.

 

[tsh]

Fair enough, but what I was getting at without going into the technical aspects of it is they are using a proven manufacturing process. That should mean a certain amount of consistency in the finished product. If a part doesn't work as intended because it has faults that exceed the fault tolerance for the part, do they still use it but sell it as a lower clocked unit? And wouldn't determining if finished parts are within tolerance be part of quality control in the first place?

The process may be proven, but I guess it's far from mature. I don't know if they plan for yields improving over the next couple of years, but I'm sure they have people tasked with achieving this. Developing newer processes may help older process tuning.
The parts arn't out of spec, and arn't faulty, they do check (somehow) that the 600MHz quoted speed is achievable. View it the same as buying a set of balanced pistons to build into a race engine. The absolute weight may not matter too much, but people will pay a premium for a set which are all within a very tight tolerance of mass. There should be no suggestion that a part which will only run at 600MHz is any way inferior in terms of quality than one which is stable at 900MHz (assuming you compare them at 600!). Maximum operating frequency probably won't be a good way to determine which chips are likely to fail in the 1st 50 hours of use (that being a very small, but non-zero number).

 

[Monk]

So I'm confused, does this mean that all boards are verified to go to 720 Mhz or are the ones that don't make it that high up sold as only 600 Mhz? That would be a real shame.

1.All batches should be verified to run at 600MHz. It's what the Pandora spec page specifies*

2.Many, possibly all, of the first batch of Pandoras may be able to reach 720MHz or more but are not guranteed to do so.

3.Second and later batches would use SOCs that fall under ONE of the below sets:
A. Have not been tested for speed (these would therefore be the same as SOCs from 2008)
B. Have failed the 720 test
or
C. Have passed the 720 test (if OpenPandora pay more for the speed tested chips)

For #3, it is unclear HOW TI are doing their testing. If they sell SOCs that fail the 720 test as 600MHz units, then we get poorer SOCs in later batches unless OpenPandora pay more for the chips. If they are not doing that - if they are just CONFIRMING that the SOCs run at 720MHz in order to charge more for some of them, then paying more for that test to be run would IMHO be silly. IIRC the cost difference is something like $8? that'd be around $32,000 extra cost to OpenPandora for another batch of 4000 just for TI to confirm that the things operate as well as they already believe they do. That's a lot of beer and pizza just for some markings on the chip.


Ta. I hope that's pretty clear. For some water muddying, see the below.


*Some users have been cautioning devs for some time not to rely on overclocking above 600MHz, and to aim to get everything working/demoed at the advertised speed. I'm one of those irritating people who have been saying that we cannot always guarantee that our Pandoras will be able to do 720MHz or more, and been told to hush. Well, nerr. TI have changed the rules under our feet and have produced the POSSIBILITY that future Pandoras may indeed not be able to overclock significantly.

Anti-doom-sayer bit. Having said the above, it's not clear EXACTLY what TI are doing. Sure, they are running tests, but we do not KNOW how the tests are run. We don't know, in short, if the tests will affect the capabilities of SOCs not sold as 720MHz checked. So there is still plenty of room to suspect fresh, new SOCs will still overclock to 720MHz and indeed possibly higher - we just don't know. My gloating is just a "See? See? the world does not stand still". I'm slightly surprised at the direction this is going in - I would have expected the chips to get faster rather than slower, overall, as the manufacturing process becomes old hat. My point all along has been that we should be able to depend on the guaranteed specs - but depending/relying on anything else is a recipe for dissapointment.