Overclocked Raspberry Pi 4 performance

Thomas Kristensen
Thomas Kristensen
Joined: 17 Mar 05
Posts: 4
Credit: 2,666,793
RAC: 6,471
Topic 224065

Hi!

I run einstein@home on three Raspberry Pi 4 2 Gb. I've overclocked them all to 2 Ghz, expecting faster processing of einstein@home workunits. But my average credit remains the same as when the Pi's weren't overclocked. How can that be?

Thanks in advance!

/Thomas Kristensen

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,354
Credit: 48,526,107,107
RAC: 57,668,125

Wouldn't it be far easier to

Wouldn't it be far easier to just check using the before/after crunch times?

Credit is a poor guide since it depends on when your completed work actually gets validated.  Maybe you have some slow quorum partners?  Have you checked that overclocking isn't producing some invalid results?

Cheers,
Gary.

George
George
Joined: 8 Jan 18
Posts: 168
Credit: 102,757,924
RAC: 259,428

Gary Roberts wrote: Wouldn't

Gary Roberts wrote:

Wouldn't it be far easier to just check using the before/after crunch times?

Pardon me for stepping in, but where might one check "using the before/after crunch times"?  Is this only for Rasberies?

George

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 991
Credit: 1,025,967,825
RAC: 2,983,316

George wrote: Gary Roberts

George wrote:

Gary Roberts wrote:

Wouldn't it be far easier to just check using the before/after crunch times?

Pardon me for stepping in, but where might one check "using the before/after crunch times"?  Is this only for Rasberies?

Just was saying to compare crunch times on the stock system and then again AFTER the intended configuration change crunch times to see if the change was beneficial or not.

 

George
George
Joined: 8 Jan 18
Posts: 168
Credit: 102,757,924
RAC: 259,428

Keith Myers wrote: George

Keith Myers wrote:

George wrote:

Gary Roberts wrote:

Wouldn't it be far easier to just check using the before/after crunch times?

Pardon me for stepping in, but where might one check "using the before/after crunch times"?  Is this only for Rasberies?

Just was saying to compare crunch times on the stock system and then again AFTER the intended configuration change crunch times to see if the change was beneficial or not.

Okay, I get it now.  Thanks.

George

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,354
Credit: 48,526,107,107
RAC: 57,668,125

George wrote:Okay, I get it

George wrote:
Okay, I get it now.  Thanks.

Sorry for causing some confusion.  I was in a hurry and was just trying to shoot off a quick response.

I also should have mentioned that there can be task to task variations that could disguise any benefit from overclocking so it's not a good idea to pick a single random task completed on stock settings and compare it with a single random task after overclocking.  The best idea is to select substantial 'before' and 'after' groups (perhaps 30 - 50 tasks each - the more the merrier) and work out the mean crunch time for each.  Looking at the individual variations within a group will guide you on how many tasks to choose for a reliable average time.

It can happen that factors other than raw frequency can limit the effect so there may not be as big a benefit as you might expect.  Care is needed so as not to cause bad results or to risk any damage to the device itself - eg. excess heat from greater power consumption.  I tend to think the downsides tend to outweigh the (usually) minor benefits from overclocking.

Cheers,
Gary.

George
George
Joined: 8 Jan 18
Posts: 168
Credit: 102,757,924
RAC: 259,428

Gary Roberts wrote: George

Gary Roberts wrote:

George wrote:
Okay, I get it now.  Thanks.

Sorry for causing some confusion.  I was in a hurry and was just trying to shoot off a quick response.

I also should have mentioned that there can be task to task variations that could disguise any benefit from overclocking so it's not a good idea to pick a single random task completed on stock settings and compare it with a single random task after overclocking.  The best idea is to select substantial 'before' and 'after' groups (perhaps 30 - 50 tasks each - the more the merrier) and work out the mean crunch time for each.  Looking at the individual variations within a group will guide you on how many tasks to choose for a reliable average time.

It can happen that factors other than raw frequency can limit the effect so there may not be as big a benefit as you might expect.  Care is needed so as not to cause bad results or to risk any damage to the device itself - eg. excess heat from greater power consumption.  I tend to think the downsides tend to outweigh the (usually) minor benefits from overclocking.

Thank you Gary for a more detailed explanation.  Again, I was just interested to know whether 'Overclocked Raspberry Pi 4 performance' had any different BOINC operational rules than a computer running Linux, Windows, or MAC operating systems.  I know nothing about Raspberry systems other than what I have read here in the forums.

George

Thomas Kristensen
Thomas Kristensen
Joined: 17 Mar 05
Posts: 4
Credit: 2,666,793
RAC: 6,471

Gary Roberts wrote: Wouldn't

Gary Roberts wrote:

Wouldn't it be far easier to just check using the before/after crunch times?

Credit is a poor guide since it depends on when your completed work actually gets validated.  Maybe you have some slow quorum partners?  Have you checked that overclocking isn't producing some invalid results?

I have no idea how to check "crunch times" and check for "invalid results". Could you point me in the right direction?

Cheers,

/Thomas

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 991
Credit: 1,025,967,825
RAC: 2,983,316

Thomas Kristensen

Thomas Kristensen wrote:

Gary Roberts wrote:

Wouldn't it be far easier to just check using the before/after crunch times?

Credit is a poor guide since it depends on when your completed work actually gets validated.  Maybe you have some slow quorum partners?  Have you checked that overclocking isn't producing some invalid results?

I have no idea how to check "crunch times" and check for "invalid results". Could you point me in the right direction?

Cheers,

/Thomas

Simply look at your account's computer hosts and the tasks they have done.  Run time is "crunch time"

https://einsteinathome.org/host/12845472/tasks/0/0

 

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,354
Credit: 48,526,107,107
RAC: 57,668,125

Thomas Kristensen wrote:I

Thomas Kristensen wrote:
I have no idea how to check "crunch times" and check for "invalid results".

If you go to your account page on the website, it shows you a list of your computers.  If you have more than the small number listed, you can use a 'MORE' link at the bottom which leads you to a full summary list.

In the full list, for each computer there are more links - things like 'Details', 'Tasks', as well as a 'Last contact' link, all of which can be explored to reveal useful information.  You should try it out sometime - lots of interesting stuff to find.  If you follow the Tasks link, you get to see your full set of results categorised - ie. columns for things like 'In progress', 'Pending', 'Valid', 'Invalid', 'Error'.

When you look at your tasks, just be aware that they don't last all that long in the online database (maybe a week or a bit more if your quorum partner is slow) so if you want to note details like cpu time and run time, you should do so reasonably promptly.

I tend to call the run time the "crunch" time - its the full elapsed time that crunching lasted before the task was finished.  It will always be a bit longer than the actual cpu time because the cpu core doing the crunching may well be diverted to other higher priority jobs from time to time.  If you see a big difference between cpu time and run time, it's an indication that your device is heavily loaded - doing lots of other higher priority things which require large amounts of cpu time.

If you're definitely not running other compute intensive stuff while crunching, but you see a big difference anyway, perhaps you should investigate what is using all that time.  I haven't browsed any of your results so I'm not saying it's happening to you.  It's just something I tend to check on my own results if performance is not as expected.

Cheers,
Gary.

Thomas Kristensen
Thomas Kristensen
Joined: 17 Mar 05
Posts: 4
Credit: 2,666,793
RAC: 6,471

Gary Roberts wrote: Thomas

Gary Roberts wrote:

Thomas Kristensen wrote:
I have no idea how to check "crunch times" and check for "invalid results".

If you go to your account page on the website, it shows you a list of your computers.  If you have more than the small number listed, you can use a 'MORE' link at the bottom which leads you to a full summary list.

In the full list, for each computer there are more links - things like 'Details', 'Tasks', as well as a 'Last contact' link, all of which can be explored to reveal useful information.  You should try it out sometime - lots of interesting stuff to find.  If you follow the Tasks link, you get to see your full set of results categorised - ie. columns for things like 'In progress', 'Pending', 'Valid', 'Invalid', 'Error'.

When you look at your tasks, just be aware that they don't last all that long in the online database (maybe a week or a bit more if your quorum partner is slow) so if you want to note details like cpu time and run time, you should do so reasonably promptly.

I tend to call the run time the "crunch" time - its the full elapsed time that crunching lasted before the task was finished.  It will always be a bit longer than the actual cpu time because the cpu core doing the crunching may well be diverted to other higher priority jobs from time to time.  If you see a big difference between cpu time and run time, it's an indication that your device is heavily loaded - doing lots of other higher priority things which require large amounts of cpu time.

If you're definitely not running other compute intensive stuff while crunching, but you see a big difference anyway, perhaps you should investigate what is using all that time.  I haven't browsed any of your results so I'm not saying it's happening to you.  It's just something I tend to check on my own results if performance is not as expected.

Hi Gary!

Thanks for that very thorough explanation, I really appreciate that :-)

I've checked the numbers you mention, normal clocked vs. overclocked, and I see almost no difference in crunch time. Apparently overclocking have no impact on crunch times. I, tbh, can't figure out why. I'm all ears if someone has a theory...

/Thomas

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.