Overclocking And Overvolting


WizardStan said:
Jdbye said:
Someone mentioned it was designed to run at 600mhz at OPP5 earlier in this topic
That's the overdrive. TI officially rates it for 500Mhz at OOP3, and if you're feeling bold, 600Mhz at OPP5 with the understanding that this will cause damage to the CPU. Standard is to run it at 500Mhz in OPP3.

Jdbye said:
Though, is it safe to run it at 1ghz at OPP5? I'm tempted to try it out
Define safe. Texas Instruments says that anything over 500Mhz at OPP3 is out of spec and can therefore be considered "unsafe", but I don't think there's anyone who hasn't hit at least 700Mhz without a problem. Increasing the voltage to OPP5 will definitely reduce the life of the CPU drastically (depending on how you measure it) but that's a problem for years down the road, and shouldn't (grain of salt, IANAL and all that) cause it to die just by trying it. The typical worst case scenario is that it just crashes, you have to restart, and it reboots at standard voltage with standard clock speed, and you remember not to go that high again.
The real worst case is that volting to OPP5 and increasing the clockspeed increases the heat being given off enough that some connection somewhere gets just that little bit too hot and the whole thing dies. This is probably very unlikely, but the possibility of it is the reason OPT went off about overvolting voiding the warranty.
I don't think I would consider a drastic reduction of CPU life safe, I would like my Pandora to last at least until the next similar device with twice as powerful hardware comes out, though I don't know how long that will take since the Pandora is rather unique. If the lifetime is reduced to say, less than 5 years from me running it at 1ghz every time I want to play N64 games, I'm not sure if it's worth it. If it lasts 6-7 years I wouldn't mind. As long as it lasts until I get a new gadget :)

EDIT: Also, looking at the source patch, I think it's unneccessary to go to OPP4 so early, since I haven't heard about any Pandoras that can't do a stable overclock to at least 800mhz at OPP3, and it would just use more battery and cause unneccessary heat. OPP4 at >800 mhz and OPP5 at >900 or so seems like a better idea (not sure about OPP5 at >900 but OPP4 should be fine at that speed). If someone compiled a kernel with those voltages I'd be more interested :) (mine is stable at 800mhz even in PSX4Pandora, unsure about 850mhz)
 
Last edited by a moderator:
Yannick said:
so we are still talking about years here, nice.

I just saw this http://www.nokiaphones.net/nokia-n900-overclocked-up-to-1700-mhz/

alright, i'll wait some more for my pandora to come in.
*facepalm*

It was a joke ;)
 
Last edited by a moderator:
mali said:
Yannick said:
so we are still talking about years here, nice.

I just saw this http://www.nokiaphones.net/nokia-n900-overclocked-up-to-1700-mhz/

alright, i'll wait some more for my pandora to come in.
*facepalm*

It was a joke ;)

lolz

but i realy liked the intel coolingsystem,
seems exactly what i would use to cool a low power chip and/or it's power regulators
it feels like *lets prey it's stable enough to benchmark + screen shot"

i benched some athlons 10 years ago on aircooled systems that looked way more mac guyver then that
 
Last edited by a moderator:
^ Hehe, I agree, the pic is quite well made. I was lurking in that big thread on TMO when it was posted back then :)
Unfortunately some bloggers just copy and paste stuff without engaging their brain :D
 
From: TI OMAP3530 Specification Document
To avoid significant device degradation for commercial temperature OMAP3530/OMAP3525 devices (0°C
£ Tj £ 90°C), the device power-on hours (POH) must be limited to one of the following:
· 100K total POH when operating across all OPPs and keeping the time spent at OPP5-OPP6 to less
than 23K POH.
· 50K total POH when operating at OPP5 - OPP6.
· 44K total POH with no restrictions to the proportion of these POH at operating points OPP1 - OPP6.
To avoid significant device degradation for extended temperature OMAP3530A/OMAP3525A devices
(-40°C £ Tj £ 105°C), the following restrictions apply:
· OPP5 and OPP6 are not supported.*
· The total device POH must be limited to less than 50K.*

Even when running at OPP5/6 the POH is still quite high (on the order of years of contiguous use)...
 
Yannick said:
please define: definitely reduce the life of the CPU drastically
One day, years from now, you'll go to turn on your Pandora and it just won't start. There may be a few days/weeks leading up to that where it starts to act funny, possibly randomly crash for no apparent reason. This is the transistors in the CPU dying in a non-glorious way. You aren't likely to get smoke and death. It'll just stop working one day, as most devices tend to do. However...

Jdbye said:
I don't think I would consider a drastic reduction of CPU life safe, I would like my Pandora to last at least until the next similar device with twice as powerful hardware comes out, though I don't know how long that will take since the Pandora is rather unique. If the lifetime is reduced to say, less than 5 years from me running it at 1ghz every time I want to play N64 games, I'm not sure if it's worth it. If it lasts 6-7 years I wouldn't mind. As long as it lasts until I get a new gadget :)
As mentioned above, the estimates (and they're just estimates because obviously no one has actually had these running for the length of time required to prove them) are that running non-stop, 24 hours a day, 7 days a week, ever week for a year, you will get about 20+ years in OPP3 but only 4-5 years in OPP5. So while 1/5(ish) the lifespan can be considered a drastic reduction, you should still be able to get a reasonable amount of usage out of it before the next big thing comes along.
Of course, this was all before overclocking was added to the mix. Texas Instruments made the estimates based on OPP5@600Mhz. What happens when you kick it up to 1Ghz+? Even if it's stable, it might only be stable at that for a couple years, then you need to bring it down to 900Mhz. And 6 months after that you can only reach 800, then 700... Or maybe it'll keep running at the full 1Ghz+ for 4 or 5 years until it just one day gives up completely, and you decide it's time for a new one.
Simplest solution: don't run it that high, that fast, unless you actually have to, and then turn it back down again as soon as you're done. 'Cause you never know.

edit: what? There's an OPP6 now? I don't remember that being in the spec sheet I read. Am I crazy or something?
 
Last edited by a moderator:
My opinion: Use the device, overclock it, have fun! Don't worry too much about frying it if you overclock inside a reasonable voltage/frequency window.

WizardStan said:
edit: what? There's an OPP6 now? I don't remember that being in the spec sheet I read. Am I crazy or something?
Apparently it goes up to vsel=84 or so. Out of spec overclocking for hardcore tweakers :D
 
Last edited by a moderator:
I think people are getting a little too worried about the overclocking reducing the life of the Pandora, kinda like the limited writes on a NAND/Memory card, yes it will reduce the life but you will most likely be bored/finished with it by then. Some how I doubt we'll all be using our Pandoras in 5 years time (and thats assuming its on 24/7, which is very unlikely), We'll all have Pandora 3's by then.
 
The cpu should last longer with power management. We will be able to undervolt at the normal 500mhz and there will be an idling 0mhz mode with far lower voltage. That should make up for the lost life time when the chip is operating at 1ghz.
 
Can someone please explain Vsel to me? It does not seem to be linear with the voltage_core selections.
 
mali said:
My opinion: Use the device, overclock it, have fun! Don't worry too much about frying it if you overclock inside a reasonable voltage/frequency window.

WizardStan said:
edit: what? There's an OPP6 now? I don't remember that being in the spec sheet I read. Am I crazy or something?
Apparently it goes up to vsel=84 or so. Out of spec overclocking for hardcore tweakers :D

i am just thinking, this all sounds like a current issues, you draw more, you do more damage/second (euh, day)
difference between us and most other devices is:

we can get the instant reaction on the heart attack the designer of our hardware will get when we tell him how much current we want to pull from his circuits.

that has to count for something

if you buy a racecar you want to know it's limits right?
 
Last edited by a moderator:
You people are freaking out like the ones that says that a journalling filesystem will absolutely and instantly obliterate a flash based memory. Flash memory will have its life reduce by an increased use, but how much... you are not constantly writing millions and millions of files with of gigabytes 24/7, you just write very small changes every so often and bigger ones, much more separated. Most of the time you just read which makes no difference whether it is journalled or not.

Same thing goes for the cpu overvolting. Sure it is going to reduce its life, but how much are you going to use it? A couple of hours a day with mame and mupen, not that much more, all the other moments, you just run at it's normal operating mode.
What's more, such voltages should be tweakable so you could choose the lowest possible at a given speed. For example, as said, most first batch pandoras can do at least 800 at opp3, my first pandora did 880, my replacement does 830.
Also chances are that at 500 mhz you could use opp2 and thus offsetting the time you use the pandora at 1 ghz, hopefully at opp4, not 5.

Finally, as an anecdote my computer with a P4 HT at 2,6 has been overclocked to 3,1 Ghz all the time and after 6 years it is going strong still.

Obviously, I'm not preaching to do 2 ghz at opp9 :D But the fact that you push the thing to its limits for a limited period of time is not going to hurt.
I'm waiting when the patient and valiant kernel compilers releases a tar with the modified drivers so we could install it with a simple copy, and thank them for their efforts.
 
The panda's cpu automatically scales it's speed, right? So overclocking to 1ghz wouldn't mean it's running at 1ghz non-stop, but only when it's needed, so the lifespan shouldn't decrease as much as people expect it to unless you use your pandora as a server or somethin'. I do wonder about the other components on the cpu chip, if they are affected by overclocking and overvolting. There's only one way to find that out for sure.

I've killed video cards in the past (easily in less than a week) but never a cpu, but i've also never been able to overclock to more than 25% of the cpu's stable speed. With that said, the mention of 1ghz pandoras got me excited, now i'm interested in seeing how fast mine will go.

Time to see just how unbrickable this thing is. 1.6ghz here I come.

On an unrelated note, I wish I could punch my boss.
 
Mofokubik said:
The panda's cpu automatically scales it's speed, right? So overclocking to 1ghz wouldn't mean it's running at 1ghz non-stop, but only when it's needed
That's how it's supposed to work in theory, and how I thought it was originally set up, but as I understand it that part of the kernel still has not been activated.
 
Last edited by a moderator:
Its not just a case of selecting an option when compiling, it needs an update to a more up-to-date kernel, or back porting of these features. As I understand, the latest kernel causes some components not to work at the moment.

Although the cpu may currently auto-scale it's frequency, the automatic voltage scaling is certainly not in.
 
Ok, I understand all CPU's are different, but here's my own take on things. When I got my pentium 166mhz (top of the range back in 1996) my friend told me I was mental cause I immediately overclocked it to >200mhz, it never went below that 200mhz mark ever. In about 2001 I got an old motherboard with a 400mhz AMDk62, and immediately overclocked it to run at 500mhz - it remained overclocked till I got my second hand p4 3ghz around 2007. I suppose the biggest difference in my thinking, with a PC and Pandora is that if I fry a PC CPU it's a piece of piss to pick up a new one and bung it in to replace the fried one - and even if the chip lasted only a year or 2 CPU prices come down fast enough to make one cheap as... well... chips (sorry - bad pun). I have 3 c64's that still run great, my Amiga 1200 is working fine too. Honestly - I have many many computers, and I've NEVER EVER had a CPU die on me, several motherboards sure, but not the CPU. I once wired up a "Laptop of DEATH" where I overclocked the 486DX and pumped most of the national grid trhough it, at one point it litterally began glowing and the acrid smoke turned me on. I cant tell you the major surprise I had when I popped the 486DX chip back into it's original mobo and it worked like nothing had ever happened to it. I still have it running as a file server for one of my c64's.

Now here's my final say, I very much believe the theory that switching a computer on and off does more damage than overclocking (within reasonable limits). Constant powering off and on of the chip causes micro-metal fatigue through the drastic temperature changes, better to run it hot and keep it hot permenantly, never swithch the machine off (the CPU should never really go off at all on the pandora anyway)! Having said that, my quad core I7 is 2.66ghz and the only thing I've ever found that maxes out the CPU is using a multithreading 64 bit parity file creation utility - so I haven't overclocked it, yet. The day will come when I need some extra oomph and I will pump up the GHZ without compunction.

Be brave, overclock. In 3 years time you'll probably be able to replace the Pandora by getting a 15th batch second hand one from ebay for £50. More likely it will just work and continue working for many years to come. How many devices have you had, how many dead CPU's have you burried?
 
Ah the memories. Overclocking CPU's to the max, 1FSB up would crash getting into windows, back in the day everything was maxed out. I remember my old 500Mhz P3, overclocked that to 1Ghz, no PCI divider so that was running at 50Mhz, it all ran perfectly fine. Never had a CPU fry and had my PC on 24/7 with some pretty high voltages too, same went for memory and GPU's.

Ha, found an old screenshot too!
p3-500e-206fsb.jpg



Anyway back to topic, if the Pandora hardware is built from quality components then it'll be fine, the voltages are within specs and that's what kills, not Mhz.
 
I did some extensive testing on my Milestone, which should react fairly similar to Pandora. The perfect setting for me is 900MHz vsel=50. 1GHz(tested vsel 56-60) produces noticably more heat and also causes graphical glitches in psx4droid. I did some experimenting and tried 1.1GHz which works well too, but the performance gain is not enough to justify the high voltage and degradation of the CPU.
IMO, the most sensible approach would be to ask OPT to up the vsel setting from 48 to 50, which is still OPP3 and include it in the official kernel. That way most if not all should reach 900MHz, which delivers the best performance at the lowest voltage, without problems and without causing damage in the long run.

edit: Some more details

1000/54 a few Quake 3 demo loops no problem, youtube app crashes.
1000/56 youtube app stable, PSX with graphical glitches.
1000/58 still graphical glitches.
1000/60 PSX ok.

800/46 no problems at all.
900/50 no problems at all.

That means you have to up the voltage 10 notches when you go from 900MHz to 1000MHz to have the same stability while you just need 4 notches when going from 800MHz to 900MHz -> not worth it, imo.

900MHz might be the sweet spot while it's still at OPP3.
 
Back
Top