Spike in credits for
5 Aug 2022 19:42:07 UTC
Topic 227930
(moderation:
I've been running Einstein@Home for over a decade now. Starting around July 1st, I've noticed a massive increase in credits I've been getting, all due to Gamma-ray pulsar binary search #1 on GPUs v1.22 tasks. These tasks take just a few minutes to run on my machine and generate 3465 credits each. I don't mind the credits but wondering what's up and why this has changed recently. Computer specs below:
CPU type: AuthenticAMD AMD Ryzen 7 3800X 8-Core Processor [Family 23 Model 113 Stepping 0]
Number of processors: 16
Coprocessors: AMD AMD Radeon RX 5600 XT (6128MB)
Operating system: Microsoft Windows 11 Professional x64 Edition, (10.00.22000.00)
If you are speaking of the
)
If you are speaking of the mini-graph on your Account page, there was a big glitch a couple of weeks ago where everyone saw a big spike in credits because the stats were not being displayed or tallied correctly for several weeks.
The big spike was for a "catch up". You can ignore it.
These tasks have always
)
These tasks have always received this amount of credit.
the change was that the gravitational wave tasks have ceased. So instead of getting mostly gravitational wave (which take longer, and award only 1000cr) you are getting exclusively gamma ray tasks, which pay a lot more credit per unit time.
not sure if more gravitational wave tasks with be coming in the future. But a new BRP7 campaign should start up in the near future.
for now, the high paying gamma ray work is all that is available for AMD/Nvidia GPUs.
_________________________________________________________________________
Ian&Steve C. skrev:for now,
)
And I don't for how many weeks, the status page says there's estimated 8+ days left.
huns0004 wrote:I've been
)
I looked at when your computer first started crunching. This page says July 5th 2020. At first, I read the year as '2022' so started trying to guess what might have happened a month ago, rather than 2 years ago.
Having realised the mistake, the only thing I can think of that might have happened one month ago is either a change in GPU or a change in driver that has allowed the full potential of the current GPU to be realised. You should have been seeing the current performance all along. Has this sudden improvement resulted from some sort of update/change you made a month ago?
As well, I had a look through your tasks list. You are running multiple searches (both CPU and GPU) and some of the Arecibo large CPU tasks (with only a 7 day deadline) are failing because they are not even starting before the deadline is reached. You have far too many tasks on board and you really need to reduce the size of your work cache - to less than a day for example - until you work out what is comfortable for the machine to handle. If you want to run multiple different searches that can impact on each other, you do need to keep the work cache size reasonably small, particularly when one search has such a short deadline.
Cheers,
Gary.
Yeah. It was dropping fast.
)
Yeah. It was dropping fast. But there seems to be a problem with either the actual remaining units or the calculated rate in which they are being done.
the status page claims ~190,000+ units completed every day based on a 5-day average, and has been at this level for weeks. But the work remaining value has not been reducing by ~190k every day. So I think there’s something wrong with one or more of the values that are going into that calculated “8 day” value. Certainly seems that the work remaining value has stagnated, driving the days remaining to stagnate.
I know Bernd has commented that more gamma ray would be added. And maybe it has, but just not reflected on the status page since we’re still getting work. They’ve also mentioned that fixing the status page (for the incorrect BRP4 info) is low on the priority list of other things they need to do.
_________________________________________________________________________
Thanks for all the feedback
)
Thanks for all the feedback everybody. No changes to my computer other than installing AMD driver updates for my GPU. It's certainly possible that one of the recent updates had an effect on processing.
Regarding the number of tasks I have, I use BAM for management and I already have it set to the allowed minimum of keeping 1 day of work. So I'm not sure what else I can do in order to reduce the number of tasks I have.
huns0004 wrote:Thanks for all
)
The dramatic increase in output is visible in the basic graph that shows on your account profile page. Ignore the spike at the very start of the graph and look at the 'Himalayan Escarpment' that arises around early July. I would guess that is where OpenCL was enabled and your GPU started contributing for the very first time.
Your best option (while you are running gamma-ray pulsar GPU tasks) is probably to find a way to get your work cache size to be lower than it currently is.
You have an 8C/16T processor. Depending on how many threads are being reserved for GPU support, BOINC will fetch work for all the remaining threads (if you allow it to).
You currently have 228 Arecibo large CPU tasks 'in progress'. These seem to be taking close to 12 hours to crunch so even if you were using all 16 threads, 228 tasks is equivalent to around 7 full days of work.
On top of that, you have a further 73 gamma-ray pulsar CPU tasks. These take even longer to crunch but do have a 14 day deadline.
The most likely reason you have that much with a 1 day setting is that the very fast GPU tasks you are now crunching have reduced the duration correction factor (DCF) to such an extent that BOINC has adjusted the estimated crunch time for CPU tasks to much, much less than 12 hours. This should be very easy to see if you look at the tasks list in BOINC Manager and see what is listed there. This is probably the main factor behind all the deadline misses you are now getting.
You can't stop BOINC from creating the low estimates for CPU task crunch time. The workaround is either to reduce the work cache size (eg 0.2 days total would give you plenty of margin) or reduce the number of CPU cores that BOINC is allowed to use. I know nothing about BAM but find it surprising that you can't reduce the work cache size below 1 day. That seems a little crazy. Do you really need to use an account manager?
If you are stuck with a 1 day cache size, you have a couple of options. You could reduce the number of cores (a setting in BOINC Manager) that BOINC is allowed to use - say to 25%. If you do that, you would need to abort some CPU tasks that have no chance of starting before the deadline. There would be lots but it's better to abort now rather than expire later on. Don't start aborting until BOINC has adjusted to the new limit on CPU cores being used.
The other option is to consider not running CPU tasks at all, or at least run just one search rather than both. When you consider that the GPU search is so much more efficient (much larger tasks in a fraction of the time), it's something that is worth considering. This may also be the case when the new BRP7 GPU search arrives and replaces the FGRPB1G search that you are currently running. As with all major changes like this, the initial period is likely to be 'bumpy' so it would pay to wait until the kinks are ironed out. It could be quite a while - weeks to months - and they will probably keep FGRPB1G running until the new search is well out of beta.
Cheers,
Gary.