Clockability

[zevdawg]

I was just reading the unofficial blog, in a recent post it stated the Pandora running at 500Mhz.
Elsewhere I had read of an emulator running with the Panda clocked in at 900Mhz,

What is the most that this has been underclocked to? (able to stably boot into OS; or at least run programs from command-line-boot)

900Mhz = safest OC (for warranty reasons)? Can it 'do' more (even with cooling)?

Anyone willing to push the boundaries (with proof of it working) of either OC/UC?

 

[borgqueenx]

1ghz can be made without damaging pandora. Underclocking knows no limits for music player for example. It all depends on the software creator.

 

[fischju2000]

Don't expect 1ghz.

Without citations because it's too late, this is what I remember:

The Pandora has different power states for CPU voltage, stepping higher to increase the clock speed reduces the life of the chip. The highest state however still allows for years of use.

I believe ED stated there is a minimum CPU speed in a thread about extreme power saving when closing the lid instead of suspend to RAM. It's very low, 10mhz?

900mhz is achievable on some boards, but not with Neon being used. 850mhz or less with? I believe that's what Pickle has been using lately when overclocking and testing power hungry apps. And these aren't mass production boards.

 

[NLS]

I wouldn't bet my pandora on 1ghz
800Mhz has been shown in some vids though

 

[kingoddball]

1ghz can be made without damaging pandora. Underclocking knows no limits for music player for example. It all depends on the software creator.

Where did you ever see 1ghz?
I've seen 900Mhz, but that has not even really happened.
900Mhz is the theoretical safe maximum, but as stated above 850mhz is about the highest that will be happening.
I would love to see it clocked at 1ghz!

 

[borgqueenx]

1ghz can be made without damaging pandora. Underclocking knows no limits for music player for example. It all depends on the software creator.

Where did you ever see 1ghz?
I've seen 900Mhz, but that has not even really happened.
900Mhz is the theoretical safe maximum, but as stated above 850mhz is about the highest that will be happening.
I would love to see it clocked at 1ghz!

Ow sorry i really thought i heard 1ghz somewhere. Cant quote it though.

But 900mhz is nice as well

 

[WizardStan]

Some quick basics so we're all up to speed. Overclocking is just running the processor at a higher speed. Underclocking is running it at a lower speed. There's no magic required to get either, you just tell the processor to go faster or slower. There are settings for this in the OMAP itself that you can adjust real time, while it is running.
The caveat being that slower clock rates extend the life of the processor and save power, while higher clock rates wear it down, use more power, and can become unstable if you clock too high.
The processors in the Pandora are guaranteed to be stable at 600 mhz. TI has done some further testing and found that these processors are stable at 720 mhz as well. What we're not sure about is whether this 720mhz applies to existing chips or if it's a change in their process that only applies to future chips. Until we know for sure, assume that 600mhz is the maximum; if you can clock to anything above that, that is overclocking and not guaranteed to be stable.
That being said, all test boards seem to be able to reach 800 mhz, with some being stable at 900 mhz with adjustments to the volts applied. These speeds should be assumed to be overclocked, however, and you shouldn't expect them. "Expect" them, in that it's reasonable to say they should work, but if you can only clock yours to 600 mhz, don't complain.
One other thing: the OMAP actually has a smart clock system as well. When idle, it can be set to automatically ramp itself way, way, waaaaay down slow, theoretically saving huge amounts of power. As soon as it's no longer idle, it returns to full speed, whatever that may be.

 

[ldesnogu]

The processors in the Pandora are guaranteed to be stable at 600 mhz. TI has done some further testing and found that these processors are stable at 720 mhz as well. What we're not sure about is whether this 720mhz applies to existing chips or if it's a change in their process that only applies to future chips.

I think you got it wrong.

Chances are high that the 720MHz chips are out from the exact same fab lines as the 600MHz ones. Out of the line some tests are done to qualify chips at one frequency or the other (binning).

What this means is that if one buys a 600MHz part now, there are reduced chances to OC it as high as previous parts. OTOH if TI can't sell enough 720MHz parts, they might badge them as 600MHz parts.

My understanding is that OP had already bought the chips before TI started doing that. But the next batch might very well suffer from less overclockability.

 

[EvilDragon]

Never had problems on my board with 720MHz

I think it can go down to something like 10 or 12 MHz.

 

[timofonic]

Never had problems on my board with 720MHz

I think it can go down to something like 10 or 12 MHz.

I'm more interested in extreme unverclocking. Thanks for the info.

It's actually possible to play audio and video files with the C64x DSP (or taking advantage of it)? It's possible to use ffmpeg on on it?

When you underclock the processor, you underclock the rest of the clock too or each component of the SoC can be underclocked/overclocked individually?

Does the SGX have video decoding capabilities of some kind?

What would be the minimal mhz required for audio playing?

How can battery be saved by underclocking? Some tests about it?

 

[mali]

Chances are high that the 720MHz chips are out from the exact same fab lines as the 600MHz ones. Out of the line some tests are done to qualify chips at one frequency or the other (binning).

What this means is that if one buys a 600MHz part now, there are reduced chances to OC it as high as previous parts. OTOH if TI can't sell enough 720MHz parts, they might badge them as 600MHz parts.

My understanding is that OP had already bought the chips before TI started doing that. But the next batch might very well suffer from less overclockability.

What I've read on these boards(so it might not be accurate), is that there are no 600MHz chips any more, the 3530 has a 720MHz specification, now. So no binning takes place to sort 600MHz and 720MHz versions as 600MHz versions don't exist. The next higher version is the 3540(800MHz).

Edit:
Regarding the original post, I would really like to see a mutiplier/divider table that shows all the available configurations of Cortex, DSP and SGX clock speeds. You have to follow certain rules when setting the multipliers/dividers and it might be possible that in some situations using a lower CPU clock allows a higher SGX clock, which might be better in graphics intensive games.
I would really appreciate if an expert could shed some light on this, or point me to an older discussion that I've missed

 

[ldesnogu]

What I've read on these boards(so it might not be accurate), is that there are no 600MHz chips any more, the 3530 has a 720MHz specification, now. So no binning takes place to sort 600MHz and 720MHz versions as 600MHz versions don't exist.

Wrong

http://focus.ti.com/docs/prod/folders/print/omap3530.html#pricingpackaging

Note the parts that end in 72 that are the 720MHz ones (for ref about naming conventions read http://www.ti.com/lit/gpn/omap3530).
All of the parts are marked as ACTIVE.

Also note how much more expensive the 720MHz parts are.

 

[mali]

Thanks, Laurent So it seems I've misunderstood something.
I'll parse some threads and try to find some quotes with OC stuff.

The clocking on OMAP3530 is complicated and I keep having to go over it.. but I'll try to grasp it a little better here.

There are five PLLs, DPLL1 through DPLL5, that generate clocks internally from SYS_CLK. DPLL3 sources SYS_CLK, and generates CORE_CLK, which goes to L3_CLK, the level 3 memory subsystem (which the SDRAM interfaces on). There's a 1 or 2 divider from DPLL3 to L3_CLK. SGX FCLK (the rate it operates at internally) is sourced from either CORE_CLK / { 3, 4, 6 } or from CM_96M_FCLK which comes from DPLL4 and as the name suggests is a fixed 96MHz.

So if I understand correctly, we're looking at a DPLL3 input of 333MHz, generating CORE_CLK at 333MHz, L3_CLK at 333MHz or 166MHz, and SGX FCLK at either 55.5MHz, 83.25MHz, or 111MHz.

Obviously the DDR won't run at 333MHz, so the question becomes one of whether or not we can change DPLL3/CORE_CLK. Although not tied to the clock speeds of the CPU or DSP, this still regulates a ton of things. I'm going to guess that since some of them have MHz ratings in their names, that it really has to be 333MHz. For instance, there are 96MHz and 48MHz outputs, I imagine these go to the SD card and USB amongst other things. I don't think you'd want to try overclocking these things.

On the other hand, maybe you could alter DPLL4 to change the SGX clock speed..

I've seen the SGX core clocked at 166MHz based on CORE_CLK/3: 500/3

I think the core wasn't doing much at the time but you can physically set it to 166 or even 200MHz. I'm not saying it won't go nuts but it will be tested by someone sometime soon enough I'm sure.

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.

Edit:
What got me confused in the first place is this:

It's possible TI will increase the speed of the OMAP3, but the chances are the older chips will reach that speed anyway.

It looks like they might start saying they can all go to 750mhz. So you may well get a 'free' upgrade.

It seems so. The new beagleboards are using 720MHz chips since 600MHz parts are unavailable.

According to Wkipedia, the OMAP 3530 is officially at 720 MHz now.

 

[ldesnogu]

Beware of what people say, many can be wrong, and this includes me of course :lol:

 

[Elanzer]

One thing to be aware with about overclocking/underclocking is that sometimes underclocking can actually hurt battery life. CPUs are most power efficient when they get their work done fast and drop to idle state, the lower you have the CPU clocked, the longer it must stay in the higher powered state - which often isn't most efficient as the clock speed decrease and power usage decrease is not directly relative.

A good example is media encoding: typically when underclocking a desktop CPU by 25% a video encode generally will be ~20-25% slower but power savings will be less than 25%, the underclock results in worse power efficiency over an equal period of time. Then for actual computer use, you have to take into consideration for how much longer it takes you to actually use the device out of standby to get things done - if you had to spend more time waiting for it to process it's workload before putting it back into "standby", did it really save you the battery drain?

It was different with older CPUs which did not have low power idle states or automatically clocked/undervolted the CPU from stock.

I'm not sure how the voltage/frequency automatically ramps down on the TI CPUs, but this is true to Intel/AMD CPUs and it sounds like the TI CPU acts similar during idle state, for regular use it may be more beneficial to have the higher clock speed. That, and it wouldn't be very beneficial to battery life anyways as at 600mhz the TI CPU uses maybe 0.6 watt load, the wifi module itself probably uses double the power of the SoC alone.

 

[mali]

Elancer, I agree. Right now I'm using an 1810t(su3500). I did some hardware mods for overclocking and undervolting and did extensive power consumption tests. Now I have disabled speed step and run at maximum multiplier all the time. Race to idle is where it's at It might not be comparable to ARM SoCs though, because of the interdependencies of all the other component's clock speeds.

 

[Yoyobuae]

Intel chips are notorious for lower power consumption at idle state. And more importantly are x86 chips which aren't precisely the best at power management (with speed step or similar).

I'd like to believe that ARM chips can do much, much better job at this, their strong point: power management.

 

[Dead1nside]

Now I'm confused, I thought the Pandora uses an OMAP 3430 not a 3530?

The 3530 has a 720Mhz clock speed and the DSP has a 90Mhz faster clock speed (520Mhz) compared to the 3430, according to Wikipedia.

Moreover the DSP on openpandora.org is listed as having a speed of 430Mhz, which would position it as a 3430.

 

[mali]

The wikipedia article is confusing, as Laurent pointed out. The 600MHz version is still available as it seems. The 3430 can only be ordered in large quantities, afaik.

 

[Dead1nside]

The wikipedia article is confusing, as Laurent pointed out. The 600MHz version is still available as it seems. The 3430 can only be ordered in large quantities, afaik.

Thank you but which SOC are we using, I thought we've always been using the 3430 @ 600Mhz despite what speed-binning has occurred later. If we are indeed using the 3530, which I don't think is possible given the quoted specs on the official site and the availability of other 3430 SOC devices, then we'd have extra horsepower on the DSP too and a higher standard clockspeed.