Better To Wait For The Omap 3630?

[Butterman]

QUOTE
Firstly, the CPU is a 900 MHz Cortex A8 which has actually been underclocked to 600 MHz.

WRONG!

It's a 600MHz CPU that seems to do well at up to 900MHz under the circumstances it has found itself in the hands of Pandora devs. I wish this particular inaccurate fact could be remedied once and for all. For all we know TI could change the silicon three months from now and suddenly all chips fail at 650MHz. I don't see that happening but it is worth noting.



Keep repeating this to yourselves gang. Pandora's going to allow people to do overclocking, yes. You're damned nuts if you're writing code that ends up needing it.

Keep solidly in mind of what we're seeing in the way of performance on the Pandora with the GP2X straight migrations- I'm curious just precisely what an overclock would allow to happen that's only half again faster that a code clean up would be unlikely to accomplish the same result...

You can only clean up code so much.

Also, overclocking the OMAP won't decrease its lifetime. Unless it gets really hot, which it probably won't. Considering we're not increasing its CPUv at all.

 

[Svartalf]

Now that you've mentioned it, I don't actually know where the 900 Mhz figure actually came from. The CortexA8 is rated from 600 to "greater than 1GHz" from the ARM white pages. On the other hand, the OMAP3530 data sheet only lists MPU clock speeds of 125, 500, 550, and 600 as being possible. On the other other hand, Notaz, MWeston, and Squidge have all posted that they had great success running at 900 MHz (but not beyond). On the other other other hand, the BeagleBoard, using the same SOC, can only reach 600 Mhz, according to it's spec sheet, and I've not seen anyone suggest otherwise.
So now I'm rightly confused. I've got one too many hands.

Heh... The CPU Core is capable of up to 1GHz. With the right parts choices, board layout, etc. you can overclock the SoC we're discussing to possibly run at least the Core stably at 900MHz, but it's a difficult thing to accomplish and it's not 100% assured to be stable for all cases. We apparently got a slightly better layout that allows some things to be ran hotter. But it's silly for anyone to expect someone to run in this mode.

It's ill advised to run overclock or to have a title expect or need it overclocked- there's too much variabilty operating outside of official stated parameters for a device. Most developers of games will bluntly tell you to take your machine out of overclock before reporting issues with a title- and won't give you the time of day until you do take it out of an overclocked state. This includes anything that doesn't come stock "overclocked" like the top-end display adapters which have been qualified to run in that configuration.

You can only clean up code so much.

Also, overclocking the OMAP won't decrease its lifetime. Unless it gets really hot, which it probably won't. Considering we're not increasing its CPUv at all.

Sigh... When your hardware engineer designing the thing corrects everyone on a periodic basis, you should listen to the man.

It's not lifetime that we're talking about. It's variability. It will not be consistent from unit to unit- or between production runs. You can, in theory, have a Pandora that won't overclock period because it's not operating within TI's stated parameters. Expecting code to run in that mode is bogus.

Here's a hint for you: Unless you're attempting something like a Dreamcast emu or something along the lines of Halo2 on this handheld, you're going to be unlikely to need more than the stock configuration. And the odds are good that you can actually wedge those in there as well.

 

[Butterman]

Eh? What are you talking about?

I've overclocked my C2D by 1GHz (2.6 to 3.75), I never have any stability problems. It's nothing to do with the fact that you're overclocking. Its because most people have no idea how to overclock correctly. They just change the CPU multiplier or FSB and if it boots fine leave it. Wrong. A proper overclock takes about a few hours to properly configure (getting the lowest possible vCore for your overclock). Constant rebooting and 1 hour stability tests. Then, you leave it over night running a stability test. After 12 hours 100% load, it's safe to say you're stable.

That's why, when you look for support, you don't mention that it's overclocked, or you get some ignorant idiot telling you to stop doing what you're very experienced at.

EDIT: I understand what you're getting at, not making programs that only run at ~900Mhz. Because they won't run on all models and I agree.

 

[Svartalf]

QUOTE
EDIT: I understand what you're getting at, not making programs that only run at ~900Mhz. Because they won't run on all models and I agree.


[post trimmed... Seems the edit got in as soon as I responded... ]

 

[mali]

Just a question, what is the topmost chip on the SoC? Is it the CPU or is it the PoP RAM?
Ever since I saw the PCB I thought I could build a heat-spreader out of copper plate and connect it to the SD-slots, if needed/wanted.

 

[joseluisjazz]

...
I've overclocked my C2D .....

Excuse me if I say something untruthful, but I had the idea that the Arm & x86 processors are very different regarding overclocking, so your experience with x86 wont be of much help in this case. Arm processors are very efficient and I heard the quickly become very unstable when you try to overclock them, long before they start becomming warm, they crash.

 

[ldesnogu]

Now that you've mentioned it, I don't actually know where the 900 Mhz figure actually came from. The CortexA8 is rated from 600 to "greater than 1GHz" from the ARM white pages. On the other hand, the OMAP3530 data sheet only lists MPU clock speeds of 125, 500, 550, and 600 as being possible. On the other other hand, Notaz, MWeston, and Squidge have all posted that they had great success running at 900 MHz (but not beyond). On the other other other hand, the BeagleBoard, using the same SOC, can only reach 600 Mhz, according to it's spec sheet, and I've not seen anyone suggest otherwise.
So now I'm rightly confused. I've got one too many hands.

Here's one other hand I have no doubt Pandora's people reached 900 MHz. But this was propagated as "hey the Pandora can run at 900 MHz" and now people think that all of OMAP3530 chips can reach that speed. That's simply wrong.
As an example, think about revisions OMAP3530 has seen in the last few months, 2.0 then 2.1 then 3.0 (isn't there also 3.1?). All these revisions probably fix some bugs, and these fixes could increase critical paths, hence reducing potential safe overclock.

As far as ARM saying A8 is rated from 600 to >1GHz, this depends on the process being used, and this is not only a function of size (65nm or 45nm). For instance FreeScale officially claims its Cortex-A8 based i.MX515 can reach 1 GHz.

 

[mali]

Excuse me if I say something untruthful, but I had the idea that the Arm & x86 processors are very different regarding overclocking, so your experience with x86 wont be of much help in this case. Arm processors are very efficient and I heard the quickly become very unstable when you try to overclock them, long before they start becomming warm, they crash.

You could increase Vcore so that they are stable but tend to overheat.
The 800Mhz stock speed version of Cortex A8 uses a higher Vcore IIRC

 

[baztailor]

http://www.engadget.com/2009/02/16/arm-deb...book-processor/

just bringing it to attention

 

[ldesnogu]

Also, overclocking the OMAP won't decrease its lifetime. Unless it gets really hot, which it probably won't. Considering we're not increasing its CPUv at all.

That's wrong. To increase frequency you have to play with OPP levels. Now please take the time to read page 124 of http://focus.ti.com/lit/ds/symlink/omap3530.pdf
For me it just says the contrary of what you claimed. Do you have some other information that confirms what you wrote?

 

[Butterman]

You could increase Vcore so that they are stable but tend to overheat.
The 800Mhz stock speed version of Cortex A8 uses a higher Vcore IIRC

The main problem with increasing the vCore is that you're opening yourself up to electromigration. Electromigration = dead Pandora .

 

[ldesnogu]

http://www.engadget.com/2009/02/16/arm-deb...book-processor/

just bringing it to attention

Blatant mistake : Cortex-A9 is not Sparrow, Cortex-A9 was named Falcon.

 

[cb88]

so ... how much do these new 45nm chips cost? is it even viable? I thought the SoC had already been produced...

if they are pin compatible and the only difference is size (also smaller process chips are cheaper to make in mass production)

why not?


I find it a bit strange that the 3530 isn't listed on this page but a newer 36xx is...

CODE
http://focus.ti.com/general/docs/wtbu/wtbuproductcontent.tsp?templateId=6123&navigationId=11989&contentId=4682

 

[ldesnogu]

I find it a bit strange that the 3530 isn''t listed on this page but a newer 36xx is...

CODE
http://focus.ti.com/general/docs/wtbu/wtbuproductcontent.tsp?templateId=6123&navigationId=11989&contentId=4682

This is because the page you point to is for Wireless Business Unit chips These chips are for large companies (e.g., Nokia).
OMAP3530 is called an application processor, look here:
CODE
http://focus.ti.com/paramsearch/docs/parametricsearch.tsp?family=dsp&sectionId=2&tabId=2218&familyId=1525&paramCriteria=no

 

[mali]

So, Pandora II in 2011/12 maybe?

CODE
http://www.prnewswire.com/mnr/ti/36966/

 

[benjiro]

So, Pandora II in 2011/12 maybe?

CODE
http://www.prnewswire.com/mnr/ti/36966/

I don't think that the OMAP4430 differs so much from the 3640. The biggest change is the Multi core A9, and a second camera it seem. All the rest like BT, WIFI, GPS connections etc are present in the 36xx serie.

 

[mali]

That thing is hefty

 

[MWeston]

So, Pandora II in 2011/12 maybe?

CODE
http://www.prnewswire.com/mnr/ti/36966/

I don't think that the OMAP4430 differs so much from the 3640. The biggest change is the Multi core A9, and a second camera it seem. All the rest like BT, WIFI, GPS connections etc are present in the 36xx serie.

What, are you kidding me!?

Dual core Cortex-A9, SGX540 (4 times throughput), DDR2 memory interface, built in HDMI, new power management chips with more integrated features, 45nm, 720MHz clock, IVA3, 1080p, back end audio processor, plus some other features integrated right into OMAP chip instead of needing a separate chip (ie. keypad scanner, mic)

How is that anything like OMAP3?

 

[ldesnogu]

HaaaaaaCortex-A9 at last If you like Cortex-A8, you'll love Cortex-A9

 

[mali]

Pipelined VFP anyone?
*runs*