What limits RPi4 OC to 1750 MHz?

[pik33]

I tried to OC the ARM in RPi4 which can go stable @1750 MHz even without overvoltage, so it seems the oc headroom is huge.
Having an active cooling case I keep the CPU far from 80C.

Then I tried arm_freq=1800 and got something like 1000 MHz. All settings above 1750 set the CPU speed lower.

What causes this limit?

(1) a bug?
(2) a firmware limit set?
(3) a PLL reached its max frequency?
(4) something other?

[jahboater]

I tried to OC the ARM in RPi4 which can go stable @1750 MHz even without overvoltage, so it seems the oc headroom is huge.

What did you run to confirm its stability at 1750Mhz?

Mine fails to boot at 1700MHz but is completely stable at 1650Mhz, so I run it at 1600Mhz to give a margin of safety.
No over-voltage and no extra cooling - just mounted on edge in free air.

I doubt if it matters, but 1600Mhz is exactly half the SDRAM speed.

[gtechn]

You are not alone with this.

https://www.tomshardware.com/reviews/ra ... ,6188.html

The second setting, arm_freq=1750, sets the clock speed of the SoC’s four Arm Cortex-A72 processor cores to 1,750MHz or 1.75GHz, an increase of 250MHz over the stock 1.5GHz setting. At the time of writing, that was as high an increase as possible: Anything over 1,750MHz actively harms performance. Post-launch, firmware updates may increase this limit as far as 2GHz for those with adequate cooling.

According to Tom's Hardware, it is a firmware problem, and that anything above 1750MHz actively harms performance. I don't know what causes it, but it is interesting for sure.

Also, Tom's Hardware is probably in the know because they received a pre-launch unit.

[Nifnat]

The latest firmware from rpi-update lets you OC past 1750mhz. I've hit a wall at 2Ghz where is refuses to boot at anything higher, but that could just be the limits of my pi.

[gtechn]

The latest firmware from rpi-update lets you OC past 1750mhz. I've hit a wall at 2Ghz where is refuses to boot at anything higher, but that could just be the limits of my pi.

How stable is 2GHz? What are you using for cooling? Is any overvolting used?

Update: Obligatory WARNING for anyone reading this and thinking "I should just run rpi-update!", a message from the rpi-update README:

Even on Raspbian you should only use this with a good reason.

This gets you the latest bleeding edge kernel/firmware. There is always the possibility of regressions.

Bug fixes and improvements will eventually make their way into new Raspbian releases and apt-get when they are considered sufficiently well tested.

A good reason for using this would be if you like to help with the testing effort, and are happy to risk breakages and submit bug reports. These testers are welcome.

Also if you are suffering from a bug in current firmware (perhaps as one of the reporters of the bug on github or forum) and a fix has been pushed out for testing, then using rpi-update is the right way to get the fix until it makes its way into new Raspbian images and apt-get.

[Nifnat]

How stable is 2GHz? What are you using for cooling? Is any overvolting used?

I'm using over_voltage=4. Hasn't showed any signs of instability so far, left it running cpuburn for about 10 hrs and it was fine, Ran glxgears on it for a bit but don't really have anything to stress the GPU right now. I'm using an old 40x40x20 fan from a server and a small heat-sink. Temps top out at 54c. Haven't been able to test it with force turbo on though as it causes it to not boot (even on a fresh non overclocked image).

Tested it with hwbot prime and was definitely getting more performance versus it running at 1750 MHz.

[pik33]

Yes, it works as described

- after rpi-update RPi4 can go to 2 GHz
- after rpi-update force_turbo=1 disables boot. It worked before rpi-update

Edit: It seems there is a bug in the newest firmware which causes RPi not even try to start after parameters are set over the limit.
Starting with shift key pressed should override OC settings and boot in standard freqs to allow reediting the config.txt - it doesn't. I needed to remove the SD and edit the file in an external PC

[gtechn]

A Warning to anyone following this from the Overclocking the Raspberry Pi website:

NOTE: Setting any overclocking parameters to values other than those used by raspi-config may set a permanent bit within the SoC, making it possible to detect that your Pi has been overclocked. The specific circumstances where the overclock bit is set are if force_turbo is set to 1 and any of the over_voltage_* options are set to a value > 0. See the blog post on Turbo Mode for more information.

https://www.raspberrypi.org/documentati ... locking.md

Please be aware that if this processor bit is set, it cannot be reset and your warranty is void. If you are OK with that, by all means proceed!

[pik33]

So I have this bit already set. I have this RPi4 to learn all about RPi4 including OC. and bleeding edge firmwares

And then only one thing that force_turbo does is disabling boot at all.

I am writing this post now using this overclocked RPi 4 which is now working at 1998 on ARM and 666 on core (1:3), over_voltage=4, current temperature is 55C, still stable. The difference between this and standard settings are visible. Now maybe over_voltage can be set lower.

[Leeloo]

Please be aware that if this processor bit is set, it cannot be reset and your warranty is void. If you are OK with that, by all means proceed!

but is it legal ? you can't really give a user the option to overclock via software and void their warranty if they do so.... can you?

I mean in the US it was found those "warranty void if broken" stickers weren't legal eve though they have been used for decades and continue to be used.

[jdb]

The latest firmware from rpi-update lets you OC past 1750mhz. I've hit a wall at 2Ghz where is refuses to boot at anything higher, but that could just be the limits of my pi.

The ARM clock frequency is derived from a PLL VCO divided by 2, and previously on Pi4 we limited all VCOs to 3GHz. The Raspbian/NOOBS release firmwares have this limit so any requested VCO will be clamped to 3GHz and because of various other interactions in the firmware you will get some larger divisor (>2) for the turbo-mode ARM frequency.

4GHz VCO is the current limit on rpi-update firmware. If we're feeling brave, we add a config knob that drops the minimum divisor for the ARM cores to 1 so the frequency cap is 4GHz - but it may be that you get more instability when overclocking with a /1 than a /2 because PLL VCOs don't deliver perfect 50% duty cycle clocks.

[Nifnat]

4GHz VCO is the current limit on rpi-update firmware. If we're feeling brave, we add a config knob that drops the minimum divisor for the ARM cores to 1 so the frequency cap is 4GHz - but it may be that you get more instability when overclocking with a /1 than a /2 because PLL VCOs don't deliver perfect 50% duty cycle clocks.

I'd definitely be interested in that config setting if it ever becomes available, would be interesting to see how far past 2GHz it can be pushed.

[pik33]

1998/666 doesn't work at over_voltage=3, but the problem may be too high core frequency as it hangs up while starting x.

[bunklung]

How stable is 2GHz? What are you using for cooling? Is any overvolting used?

I'm using over_voltage=4. Hasn't showed any signs of instability so far, left it running cpuburn for about 10 hrs and it was fine, Ran glxgears on it for a bit but don't really have anything to stress the GPU right now. I'm using an old 40x40x20 fan from a server and a small heat-sink. Temps top out at 54c. Haven't been able to test it with force turbo on though as it causes it to not boot (even on a fresh non overclocked image).

Tested it with hwbot prime and was definitely getting more performance versus it running at 1750 MHz.

I am curious how high you can get with the GPU? Any results to post on this?

[Nifnat]

I am curious how high you can get with the GPU? Any results to post on this?

Haven't looked into it much. I've heard you can't push it above 600 but haven't tested it myself. I'll probably wait for further firmware updates before I bother with it.

[Imperf3kt]

Please be aware that if this processor bit is set, it cannot be reset and your warranty is void. If you are OK with that, by all means proceed!

but is it legal ? you can't really give a user the option to overclock via software and void their warranty if they do so.... can you?

I mean in the US it was found those "warranty void if broken" stickers weren't legal eve though they have been used for decades and continue to be used.

It's a warranty... The Pi was warranted to work at the settings provided to you, but it can be adjusted. Those adjusted settings, do not fall under the guarantee and are thus not covered by warranty.

'warranty void if broken' stickers exist as proof of tampering. Tampering voids warranties. How can they be not legal? It's proof you've done something to the device that isn't covered by the warranty.

[DavidS]

Please be aware that if this processor bit is set, it cannot be reset and your warranty is void. If you are OK with that, by all means proceed!

but is it legal ? you can't really give a user the option to overclock via software and void their warranty if they do so.... can you?

I mean in the US it was found those "warranty void if broken" stickers weren't legal eve though they have been used for decades and continue to be used.

It's a warranty... The Pi was warranted to work at the settings provided to you, but it can be adjusted. Those adjusted settings, do not fall under the guarantee and are thus not covered by warranty.

'warranty void if broken' stickers exist as proof of tampering. Tampering voids warranties. How can they be not legal? It's proof you've done something to the device that isn't covered by the warranty.

They were ruled as illegal for devices where they prevent addition of expansion when the device is designed to accept expansion, or when they are in the way of regular maintenance on a device or repair of said device. Put simply the Warranty Void If Broken stickers are illegal because invariably they need to be broken to effect repairs, which must always be allowed for the owner and user of a device to attempt without risking there warranty (at least here in the United States of America).

The issue first came to the court system in regards to the expandability of personal computers that had internal expansion slots, then the issue of user repair/maintenance was added to the cases, and later cases covered the even more.

Also it was ruled in the same vain that a warranty can not be voided by pushing a device beyond its design limits, so long as no physical modification is needed to push the device beyond its design limits. So in the USA you can not void the warranty by setting the warranty bit.

Remember that these laws were implemented to protect the consumer from companies finding invalid excuses to refuse to honor a warranty, which is still a difficulty in the USA.

[Imperf3kt]

You have two choices with overclocking (regular PC, or Pi, it's irrelevant)
Option one: you are given the ability to do it with the possibility to destroy the device
Or option two: you are locked from doing so.

Option one comes with the clause "these are the limits, any further and its on your own head".

Let's use an example.
A car tachometer has a "red line", you can always push the car beyond that, but you are told not to. If you do, and the engine blows up, is that the fault of the manufacturer for not forcefully limiting you? You were told not to do it after all.

[jahboater]

You have two choices with overclocking (regular PC, or Pi, it's irrelevant)
Option one: you are given the ability to do it with the possibility to destroy the device
Or option two: you are locked from doing so.

Intels chips (the K models anyway) are still explicitly covered by warranty even when overclocked.
I presume this is because the thermal and electrical protection is so good nowadays that Intel believe there is no risk.

[rpdom]

You have two choices with overclocking (regular PC, or Pi, it's irrelevant)
Option one: you are given the ability to do it with the possibility to destroy the device
Or option two: you are locked from doing so.

Option one comes with the clause "these are the limits, any further and its on your own head".

Let's use an example.
A car tachometer has a "red line", you can always push the car beyond that, but you are told not to. If you do, and the engine blows up, is that the fault of the manufacturer for not forcefully limiting you? You were told not to do it after all.

Most cars have rev limiters that will prevent you going past the red line, so yes, the manufacturer does forcefully limit you.
I suppose you can get hacked software for the car that will allow you to go over the line and blow up the engine. Actually, I know someone who did that. Twice.

[webbsmurfen]

How about that warranty then? Is it ok to OC or not? (not that i intend to but just curious)

This may seem as stupid question But..

It’s stated in Overclocking options

over_voltage
CPU/GPU core voltage adjustment. [-16,8] equates to [0.8V,1.4V] with 0.025V steps. In other words, specifying -16 will give 0.8V as the GPU/core voltage, and specifying 8 will give 1.4V. For defaults see table below. Values above 6 are only allowed when force_turbo is specified: this sets the warranty bit if over_voltage_* is also set.

So Over voltage=1-6 is okay and does apply to warrenty? or doest it mean that force_turbo=1 in combination with Over_voltage more than a value of 6 avoids waranty

As i get it, you’re allowed to OC as long as you don’t enter force_turbo=1 or/and Over_voltage _* with a greater value than 0 OR Over_voltage higher than=6 in combination with force_turbo=1

Is that correct?

That would make
arm_freq=1750
over_voltage=2
Perfectly okay to use

[jamesh]

How about that warranty then? Is it ok to OC or not? (not that i intend to but just curious)

This may seem as stupid question But..

It’s stated in Overclocking options

over_voltage
CPU/GPU core voltage adjustment. [-16,8] equates to [0.8V,1.4V] with 0.025V steps. In other words, specifying -16 will give 0.8V as the GPU/core voltage, and specifying 8 will give 1.4V. For defaults see table below. Values above 6 are only allowed when force_turbo is specified: this sets the warranty bit if over_voltage_* is also set.

So Over voltage=1-6 is okay and does apply to warrenty? or doest it mean that force_turbo=1 in combination with Over_voltage more than a value of 6 avoids waranty

As i get it, you’re allowed to OC as long as you don’t enter force_turbo=1 or/and Over_voltage _* with a greater value than 0 OR Over_voltage higher than=6 in combination with force_turbo=1

Is that correct?

That would make
arm_freq=1750
over_voltage=2
Perfectly okay to use

That seems about right.

[drgeoff]

Also it was ruled in the same vain that a warranty can not be voided by pushing a device beyond its design limits, so long as no physical modification is needed to push the device beyond its design limits. So in the USA you can not void the warranty by setting the warranty bit.

Are you claiming that setting that bit is not a physical modification?