I’m getting confused in my old age :-(

I’ve just rebuilt my FX8250 as a Ryzen 5 3600 and commissioned it as a new machine alongside my Ryzen 5 2600.

After a couple of day’s processing the estimates have sorted themselves out from < 3 hours to the expected 8 hours but :-

The 2600 is happily pulling down all x86-64 work units and processing to the 8 hour setting but the 3600 is pulling down all i686 work units and about a third of them are running for 18 hours with several not producing a single decoy and being allocated 20 credits.

From memory I’ve set no_alt_platform to 1 in cc_config.xml in an attempt to correct this but it has had no effect.

The only other differences are that the 2600 is running Ubuntu 18.04 and the 3600 is running Debian 10 (I wanted to see if the lighter os would run Boinc more efficiently) which means that instead of Boinc v7.16.6 it’s running v7.14.2.

So, Boinc version, OS or configuration?

all i686

Are you sure you installed an amd64 build of Debian?

all i686

Are you sure you installed an amd64 build of Debian?

Dumbklutz!

I built the Debian stick specifically to load onto an old thinkpad because Ubuntu has dropped 32 bit support.

I’m going senile - thank you for asking the obvious.

OK, no new tasks overnight and load Ubuntu in the morning.

OK, 90 minutes work yesterday morning and it’s now running Ubuntu 20.04 and that’s definitely 64 bit.

Started with a 23 hour estimate (Debian started with 2.3) but that will sort itself out soon enough. What does worry me is the credit allocation, although it’s not a driver it is there and should be right.

Most of the WUs on the new machine https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=5177210 (e.g. https://boinc.bakerlab.org/rosetta/result.php?resultid=1242230180) are achieving about 60 credits instead of the 400 I would expect. It doesn’t appear the matter whether they do 120 decoys or 1020 the credits stick around 66-68.

Any suggestions?

Any suggestions?

Measured floating point speed 1000 million ops/sec
Measured integer speed        1000 million ops/sec

Re-run the benchmarks.
Even then, it will still take a while for the Credit per Task to settle down.

---

Grant
Darwin NT

Any suggestions?

Measured floating point speed 1000 million ops/sec
Measured integer speed        1000 million ops/sec

Re-run the benchmarks.
Even then, it will still take a while for the Credit per Task to settle down.

This is where I totally fail to understand the credit system - if a computer generates 1000 decoys in 8 hours from a protein of complexity x then why are then not awarded 8*3600*1000*x*f credits where f is a simple function that converts the measure of complexity into a measure of gigaflops with no need of the benchmarks?

However, it is what it is and thanks for giving the solution. Looking at my overnight results there are 6 alternating between 200s and high 400s, after 6 results yesterday in the 66-67 credit range so it must have reset and it’s on its way again.

As has been said so often in these pages, time is the answer - time to settle down without interference.

Once again, many thanks for your help :-)

if a computer generates 1000 decoys in 8 hours from a protein of complexity x then why are then not awarded 8*3600*1000*x*f credits where f is a simple function that converts the measure of complexity into a measure of gigaflops with no need of the benchmarks?

Because Tasks that produce only 1 Decoy may require even more actual FLOPs than a Task that produces thousands, even though they run for the same period of time.

However, it is what it is and thanks for giving the solution. Looking at my overnight results there are 6 alternating between 200s and high 400s, after 6 results yesterday in the 66-67 credit range so it must have reset and it’s on its way again.

It generally does end up around the correct values, however when you get a different batch of work, it'll tend to start out with the excessively low credits again before settling down to the more usual values.
So it's worth just giving the benchmarks a run to avoid that initial excessive underpayment.

---

Grant
Darwin NT

if a computer generates 1000 decoys in 8 hours from a protein of complexity x then why are then not awarded 8*3600*1000*x*f credits where f is a simple function that converts the measure of complexity into a measure of gigaflops with no need of the benchmarks?

Because Tasks that produce only 1 Decoy may require even more actual FLOPs than a Task that produces thousands, even though they run for the same period of time.

But that's due to the complexity of the protein involved which should be predictable?????

However, it is what it is and thanks for giving the solution. Looking at my overnight results there are 6 alternating between 200s and high 400s, after 6 results yesterday in the 66-67 credit range so it must have reset and it’s on its way again.

It generally does end up around the correct values, however when you get a different batch of work, it'll tend to start out with the excessively low credits again before settling down to the more usual values.

So it's worth just giving the benchmarks a run to avoid that initial excessive underpayment.

Shall do :-)

Does Boinc periodically run benchmarks on its own or is this something I should do myself every so often?
Thanks.

Does Boinc periodically run benchmarks on its own or is this something I should do myself every so often?
Thanks.

I'm not sure.
I think it might depend on the version of the Manager- i do recall that sometimes after a re-boot (so basically restarting BOINC), BOINC would run the benchmarks before starting computation work. However i don't recall seeing that behaviour for some time now (my systems never power down, and generally only get a reboot after major updates, so every months or so).

A search came up with a post at the BOINC forums

BOINC runs those benchmarks every fifth day.

but that was over 10 years ago, and quite a few BOINC Manager revisions. And disabling the Benchmarks is a cc_config.xml option, so some Manager versions may have it enabled by default whereas others don't.

---

Grant
Darwin NT

Does Boinc periodically run benchmarks on its own or is this something I should do myself every so often?
Thanks.

Yes it does, but I think its something like a week before it will run them on its own. It will also run benchmarks on a BOINC version change. It only seems to check when BOINC starts up, so if you leave BOINC running for weeks on end (like some of mine) you'll have to give it a prod to get it to run the benchmarks.

---

BOINC blog

Benchmarks are run every 30 days maximum. Or whenever the client version changes as mentioned. Or when significant hardware changes are detected.

#define BENCHMARK_PERIOD (SECONDS_PER_DAY*30)
// rerun CPU benchmarks this often (hardware may have been upgraded)