I've seen "Milkyway@home N-Body Simulation with Orbit Fitting 1.87 (mt)" tasks complete in about 3 minutes (16 cores) or less sometimes. Now that I have all 32 threads running (AMD Ryzen 9 7950X3D), I have two 16 CPU tasks taking over an hour going at the same time. Is this a problem? Or do I just have new tasks that take longer normally?

I just switched back to using 50% CPU in settings to allow one task of 16 CPUs going, not two.

I'll watch and see how long tasks take at 50%.

What's the normal amount of time these tasks take, or do they vary?

Thanks,
Rich

Edit: It looks like I just switched at a time when processes take 50+ minutes for me to run. I've got one M@H N-Body Simulation with Orbit Fitting 1.87 (mt) [16 CPUs] going at 54% and 28 min ran, 23 minutes to go. I'll continue to watch, and switch back if it seems like they all run this way.

What's the normal amount of time these tasks take, or do they vary?

They vary quite a lot here, I have tasks, which take less than an hour of CPU-time while others take over a day (usually those with "data__9" in the name). For highest production I recommend running single thread tasks, i.e. set "Max # of threads for each MilkyWay@home task" to 1. Each task might run quite long than, but MT tasks are not very efficient, so you'll get more work done that way. For my Ryzen 5700G this is the difference I found between ST and MT tasks, you should get similar results.

---

Oh wow, that's counter-intuitive to me. nice hack!

For highest production I recommend running single thread tasks, i.e. set "Max # of threads for each MilkyWay@home task" to 1. Each task might run quite long than, but MT tasks are not very efficient, so you'll get more work done that way. For my Ryzen 5700G this is the difference I found between ST and MT tasks, you should get similar results.

Thank you very much for this information !
I was just about to ask this question on this forum.
I noticed the same thing : you can calculate more tasks in a day by setting Max # of threads for each MilkyWay@home task[/url]" to 1.
So I don't see what the point is of being able to calculate a task on multiple threads.
Why, then, does the project offer this possibility ?
I also noticed that when I chose one thread per task, the project didn't award more points, even if the amount of work done was far superior.
Is that normal ?

So I don't see what the point is of being able to calculate a task on multiple threads.
Why, then, does the project offer this possibility ?

It was introduced 15 or so years ago when computers were a lot slower than today and one WU needed several CPU days. Today it's obsolete IMHO, in particular since we don't have the "fits in cache" benefit here like on PrimeGrid.

I also noticed that when I chose one thread per task, the project didn't award more points, even if the amount of work done was far superior.
Is that normal ?

Well, that's CreditNew AKA Random Credit Generator, it never worked properly and likely never will, but it's unfortunately pretty much the only thing we have for WUs with random runtimes. It starts to calculate the performance for each application from scratch, the outcome is different every time. For example I was getting a significantly lower credit for v1.90 than for v1.93.

Also don't expect your RAC to adjust instantly, it needs about 6 weeks. The difference for your Threadripper with 128 threads might also not be as big as for the average computer, you were down to 6 threads per task already.

---

Thank you for your reply, which confirms my findings.
It is important for me to be sure that the project is getting the most out of my CPU and that there are no malfunctions on my machine.

The difference for your Threadripper with 128 threads might also not be as big as for the average computer, you were down to 6 threads per task already.

Yes, I have abandoned more than a thousand tasks twice.
I don't like doing that, I don't think it's “clean.”
But I didn't want to waste several days calculating more slowly than my machine can do.
I'm going to make up for it now by letting it run for several weeks at optimal speed for my machine, just the way I like it !

Reading an article in a French astronomy magazine reminded me how important physical modeling and simulations are for establishing a good correlation between theory and observations !