Anybody home :)

soon :)

Finally another GPU project!

Arrived...
(me, not task)

Would it not make sense to open the Team creation for now, so the Team leader can create Team accounts already. Team creation is blocked "You must complete a task to create a team"

...which is not quite accurate. the requirement is that you must have a validated task i.e. have credit. poor wording from upstream

it's one of the anti-spam measures. I'll leave it in place for now. if you don't have credit yet to create your team, the project will be fully live soon and you can create when you have credit.

Maybe post those system requirements somewhere on this site?

sure. I'll write one up soon.

In reply to bluestang's message of 4 Nov 2026:

Maybe post those system requirements somewhere on this site?

I made a new FAQs forum where I can post information to common questions that might come up

It's not clear how concurrent cpu only or gpu threads can be configured. And if it's to be via app_info.xml or app_config.xml files, samples files would be helpful. Perhaps something could be added to the FAQ's.

In reply to mountkidd's message of 4 Nov 2026:

It's not clear how concurrent cpu only or gpu threads can be configured. And if it's to be via app_info.xml or app_config.xml files, samples files would be helpful. Perhaps something could be added to the FAQ's.

concurrent CPU tasks are not really a thing. the CPU tasks are single threaded and will run and use the CPU in the normal way BOINC does for all CPU apps, across all projects. trying to run multiple tasks "per core" or "per thread" outside of normal HT/SMT technologies built into the CPU usually does not work well or provide any benefit. BOINC will run however many CPU tasks that you allow via the normal "use X% of CPU" settings that are set in the local or web compute settings, whichever you use.

edit: just typed one up
app_info.xml is not used for this, that is only used to define Anonymous Platform applications

app_config.xml for changing GPU concurrency is the same across any project for the most part. you can take an example from einstein and just put the "kessler" app name and put it in the kessler project folder and it will act the same as it does on Einstein or any project. I can put a short note about doing that for GPU in the FAQs.

Since I can't post in the FAQ thread, how do we know we are getting the 12 or 24HR fast return bonus?

In reply to bluestang's message of 5 Nov 2026:

Since I can't post in the FAQ thread, how do we know we are getting the 12 or 24HR fast return bonus?

It's not really possible to know just by looking at it (as a volunteer), since the credit reward itself can vary a lot. but I have checked and it is working as intended. almost every task in the first batches got the 12hr reward.

two additional comments;

some of these tasks ran much longer than I had anticipated, slower CPUs crunching what ended up being extremely difficult tasks with a large number of objects. I extended the 12hr/24hr bonus periods to 24/48hr now. and will probably keep it that way. I also corrected the credit for any task that didn't make the 12hr cut yesterday and bumped them to the 50% bonus for 24hrs instead. I'll update the FAQ to reflect the new bonus periods.

I will look into maybe adding some kind of flag or marker in the results page that might be able to reflect which bonus you got.

In reply to Ian&Steve C.'s message of 5 Nov 2026:


some of these tasks ran much longer than I had anticipated, slower CPUs crunching what ended up being extremely difficult tasks with a large number of objects. I extended the 12hr/24hr bonus periods to 24/48hr now. and will probably keep it that way. I also corrected the credit for any task that didn't make the 12hr cut yesterday and bumped them to the 50% bonus for 24hrs instead. I'll update the FAQ to reflect the new bonus periods.

[/quote]

Is there a way for the Project to only send the shorter tasks to pc's with slower cpu's, kinda like some projects restrict gpu tasks to gpu's with 4gb or more?

Is there a way for the Project to only send the shorter tasks to pc's with slower cpu's, kinda like some projects restrict gpu tasks to gpu's with 4gb or more?

there's not really a way to know ahead of time what runs will be long or short. it's up to the random collisions and interactions that happen in the sim.

runtime is a function of many factors

examples:
- in general, more objects make each "step" of the simulation take longer.
- more collisions cause more fragments, which propagate and create more overall objects. different runs (seeds) of the same scenario end up with different probabilities that collisions happen (other variables are influenced by the seed's randomness also)
- some situations blow up to many objects quickly and self-abort when it reaches 2million
- some situations don't generate too many objects and run the full 200 years
- some situations generate a large number of objects, but under the 2 million cap, and still run the full 200 years. these are the ones that tend to take the longest.

but I've tried to create a credit system that still rewards credits fairly so the long runs get more credit and you don't feel like you are missing anything from the long runs.

Thanks Steve (or is it Ian?).
One, possibly unintended, consequence of the great variability in run time, and "quick finish" bonus may be to encourage people to have much smaller caches, and so fewer tasks available for a rainy day. These combined may encourage others to join the project because there is a greater chance of everyone getting a small number of tasks instead of a few grabbing a large proportion of those available.

It's Ian :)

encouraging people to return quickly will naturally encourage people to have smaller caches so they don't sit on the host waiting a long time. It's really an effort to say like "hey I really value getting these results back quickly, and I'll pay you for it". I extended the bonus periods to ensure that it's possible for just about any device to make at least one of the bonus periods and the vast majority of devices should be able to make the first 50% bonus period.

right now there are limits in the scheduler that limit the tasks in progress for devices. 10x tasks per CPU core, or 16x tasks per GPU. maybe the CPU setting is too high and would send too much work to slow CPUs, i'll reduce it to 4x. but I will probably add a limiter function that some projects have (Milkyway/Primegrid) to limit the number of tasks a host gets if you want less than the standard settings.

I'm also open to relaxing these limits for the purposes of competitions :)

Yeah, we have badges!

yeah i made them (with claude) and added them last night for those of you who like them :)

everyone that helped out during the beta period gets a commemorative "beta" badge, and then there are normal "levels" badges for:

1 million
5 million
10 million
25 million
50 million
100 million
500 million
1 billion