The validate errors will account for some of your 40% reduction in RAC but there is also a disparity in the old unit resulting in 227 points and the new at 230 which considering the work unit takes longer than the 3 point extra awarded there is a defecit on every work unit. This has resulted in a shortfall of about 250K per day in my situation without the failed units.

The time difference in my case is similarly small for regular WUs. With computer otherwise idle, used to take almost exactly one hour, give or take mere seconds. Now it's 1:01-1:02. But there are indeed those (so far few, but existing) WUs taking about 1:47, yet those reward 230 credits just as well.

Either way, doing the math, at almost exactly 1h/WU and 2 cores, if computer would have done nothing else at all, my maximum RAC before this latest run should have been up to 10900, and it was usually a little over 10000, with peaks around 10500 if I'd hardly use the computer for a few days and dipping below if I'd use it more, such as playing a game. So, very little room for invalids before. And I also occasionally checked anyway.
Now, with the new values, my maximum should be up to 10700, a bit less if 1-2 of those longer ones would come per day. The difference to the current value of less than 7000 comes from invalids.

And if there are those tiny but sufficiently significant differences between my results and those of wingmen that produce those invalids, I wonder whether the new run does something or uses some instructions that older or weaker CPUs don't deal with so well... Or may it have something to do with the 32-bit OS?

While I'm here, more long ones:

https://milkyway.cs.rpi.edu/milkyway/result.php?resultid=260469551
https://milkyway.cs.rpi.edu/milkyway/result.php?resultid=260929684
https://milkyway.cs.rpi.edu/milkyway/result.php?resultid=262187343

The validate errors will account for some of your 40% reduction in RAC but there is also a disparity in the old unit resulting in 227 points and the new at 230 which considering the work unit takes longer than the 3 point extra awarded there is a defecit on every work unit. This has resulted in a shortfall of about 250K per day in my situation without the failed units.

The time difference in my case is similarly small for regular WUs. With computer otherwise idle, used to take almost exactly one hour, give or take mere seconds. Now it's 1:01-1:02. But there are indeed those (so far few, but existing) WUs taking about 1:47, yet those reward 230 credits just as well.

Either way, doing the math, at almost exactly 1h/WU and 2 cores, if computer would have done nothing else at all, my maximum RAC before this latest run should have been up to 10900, and it was usually a little over 10000, with peaks around 10500 if I'd hardly use the computer for a few days and dipping below if I'd use it more, such as playing a game. So, very little room for invalids before. And I also occasionally checked anyway.
Now, with the new values, my maximum should be up to 10700, a bit less if 1-2 of those longer ones would come per day. The difference to the current value of less than 7000 comes from invalids.

And if there are those tiny but sufficiently significant differences between my results and those of wingmen that produce those invalids, I wonder whether the new run does something or uses some instructions that older or weaker CPUs don't deal with so well... Or may it have something to do with the 32-bit OS?

Cavalary you are running a fairly old version of BOINC 7.6.22 it was released in Dec of 2015. Maybe you could consider upgrading to 7.14.2 (32 bit) or higher if you have a 64 bit system. Also has your PC or mother board manufacturer released any updates for your hardware that might help you ?

Thanks
Bill F

I've seen many more validation errors since this change. This workunit, Workunit 134898225, seems to show a problem Apple systems which has been mentioned before.
https://milkyway.cs.rpi.edu/milkyway/workunit.php?wuid=134898225

Having not done MW GPU for a while, I thought I would jump in and see.

It is working OK for the moment on my RX 570 under Ubuntu 20.04.2 (latest 21.20 drivers) on a Ryzen 3600.
https://milkyway.cs.rpi.edu/milkyway/results.php?hostid=887513&offset=0&show_names=0&state=4&appid=

Cavalary you are running a fairly old version of BOINC 7.6.22 it was released in Dec of 2015. Maybe you could consider upgrading to 7.14.2 (32 bit) or higher if you have a 64 bit system. Also has your PC or mother board manufacturer released any updates for your hardware that might help you ?

Go with if it ain't broke, don't fix it, and at least so far BOINC didn't cause issues, and saw some things about changes to scheduling in newer versions.
No BIOS or similar updates that I can see (well, there is a beta BIOS update from 2018, but no corresponding final version, so not going there).

Oh, and one more long one https://milkyway.cs.rpi.edu/milkyway/result.php?resultid=262617599

Cavalary you are running a fairly old version of BOINC 7.6.22 it was released in Dec of 2015. Maybe you could consider upgrading to 7.14.2 (32 bit) or higher if you have a 64 bit system. Also has your PC or mother board manufacturer released any updates for your hardware that might help you ?

Go with if it ain't broke, don't fix it, and at least so far BOINC didn't cause issues, and saw some things about changes to scheduling in newer versions.

At some point all versions of Boinc below 7.10.XX will not be allowed because they don't have the capability to still work on all the different Projects, there was a big fuss awhile back about them all going to https and then some also banned version below 7.10.xx because of a possible security thing I think as well. Going to a version of Boinc above 7.14.xx also gives you the 'pre-fetch' that alot of people hate while others love it. If you have a zero resource share Project it knows you will need a new task soon so downloads it before the task you are working on is completed so there is no more lag between tasks. It only partially works on MilkyWay because they have a 10 minute wait period between asking for gpu tasks but not for the cpu tasks so it will continue to run cpu tasks even with a zero resource share.

Cavalary you are running a fairly old version of BOINC 7.6.22 it was released in Dec of 2015. Maybe you could consider upgrading to 7.14.2 (32 bit) or higher if you have a 64 bit system. Also has your PC or mother board manufacturer released any updates for your hardware that might help you ?

Go with if it ain't broke, don't fix it, and at least so far BOINC didn't cause issues, and saw some things about changes to scheduling in newer versions.

At some point all versions of Boinc below 7.10.XX will not be allowed because they don't have the capability to still work on all the different Projects, there was a big fuss awhile back about them all going to https and then some also banned version below 7.10.xx because of a possible security thing I think as well. Going to a version of Boinc above 7.14.xx also gives you the 'pre-fetch' that alot of people hate while others love it. If you have a zero resource share Project it knows you will need a new task soon so downloads it before the task you are working on is completed so there is no more lag between tasks. It only partially works on MilkyWay because they have a 10 minute wait period between asking for gpu tasks but not for the cpu tasks so it will continue to run cpu tasks even with a zero resource share.

There were also library version updates in 7.6.33 and 7.8.2 and a few bug fixes between then and now.

Bill F

---

In October of 1969 I took an oath to support and defend the Constitution of the United States against all enemies, foreign and domestic;
There was no expiration date.

Going to a version of Boinc above 7.14.xx also gives you the 'pre-fetch' that alot of people hate while others love it. If you have a zero resource share Project it knows you will need a new task soon so downloads it before the task you are working on is completed so there is no more lag between tasks.

Shouldn't this be an option then, add the capability but let the user decide?

And invalid task: https://milkyway.cs.rpi.edu/milkyway/workunit.php?wuid=137777181 Instant computation error, and see wingman result being the same.

Going to a version of Boinc above 7.14.xx also gives you the 'pre-fetch' that alot of people hate while others love it. If you have a zero resource share Project it knows you will need a new task soon so downloads it before the task you are working on is completed so there is no more lag between tasks.

Shouldn't this be an option then, add the capability but let the user decide?

And invalid task: https://milkyway.cs.rpi.edu/milkyway/workunit.php?wuid=137777181 Instant computation error, and see wingman result being the same.

That you have you to bring up with the Developers, it is what it is.