Suddenly I'm only receiving single threaded tasks and not the 3 threaded MT tasks as normal according my settings. Any idea why?

I did too, but the last seem to be finishing up now. Several took more than a day. Only happened to one of my computers though.
Tuba Jim

The last 5 WUs I got today are MT tasks again, so far so good.

To avoid a mix of single thread and multi thread WUs change in your Milkyway@home preferences "Max # of threads for each MilkyWay@home task" from "no limit" to the number of cores you want to use for each task.

---

To avoid a mix of single thread and multi thread WUs change in your Milkyway@home preferences "Max # of threads for each MilkyWay@home task" from "no limit" to the number of cores you want to use for each task.

I did change it to 3 and it worked until 2-3 days ago where I then got single tread tasks. Now I'm using app_config, and it seems to work.

I usually make sure the Options > Computing Preferences (Advanced view) uses 2 cores (or > 34% and < 49% on my old 6-core PC).

On the BOINC Advanced View, my Status appends (2 CPUs) for the dual-threaded work units. If the (2 CPUs) is not present, I assume that indicates a single-threaded work unit.

I sometimes receive a few single-threaded tasks even with "Max # of threads for each MilkyWay@home task" set to 2. Not sure why!

Whether you get single- or multi-threaded work units might depend upon your Computing Preferences, such that MilkyWay@home won't assign work units with more threads than your Computing Preferences number of cores. It might also depend on what BOINC projects are active at the time the request for more tasks is made, because I often run multiple BOINC projects.

I do not have an app config XML in C:\ProgramData\BOINC\projects\milkyway.cs.rpi.edu_milkyway

One box (my main desktop, 5950X) not getting (mt) tasks now... well for maybe a month...

https://milkyway.cs.rpi.edu/milkyway/show_host_detail.php?hostid=1029692

I have another box(5900X) that has the same OS, same virtualization (none) and it's getting mt tasks.
I've looked at this till I'm blue in the face so it is likely something very obvious that I'm just overlooking. Any thoughts?

Thanx, Skip

PS: All are Linux Mint boxes.

Not getting MT tasks is not an issue, it's actually something good. MT seems very inefficient, so unless you want MT tasks for some special reason, I'd recommend to run Milkyway in single thread mode. This also makes it a lot easier to run Milkyway tasks together with other projects, which are usually ST too.

---

One of my computers doesn't receive any mt tasks and I've tried everything I can think of. Almost all the single thread tasks take an extremely long time. More than 24 hours is much too long.

Any help is appreciated.

Neal

I just noticed that a mt task has appeared on that computer.

Neal

Not getting MT tasks is not an issue, it's actually something good. MT seems very inefficient, so unless you want MT tasks for some special reason, I'd recommend to run Milkyway in single thread mode. This also makes it a lot easier to run Milkyway tasks together with other projects, which are usually ST too.

+1

---

Milkyway@home isn't my primary project, so I prefer short tasks. I noticed that Milkyway (mt) never hits 100% CPU utilization in windows task manager. However, you may get higher cache utilization from multithreading. It would be interesting if somebody ran MT and ST for a while and was able to compare the two.

Neal

It would be interesting if somebody ran MT and ST for a while and was able to compare the two.

I did and as I have written just 4 posts above yours, running the application in MT mode is complete nonsense IMHO, even if you use just two thraeds per task. With more threads per task it only gets worse.

---

The kind of efficiency I care most about is power efficiency. If MT wastes power, then that matters to me.

Have you determined the credit rate with an ST task running on every core vs the max number of MT tasks running concurrently?

Neal

The kind of efficiency I care most about is power efficiency. If MT wastes power, then that matters to me.

It always does, the synchronisation between the threads is additional work. In addition to that, cores are not loaded 100%, there's lots of waiting, you can see that in task manager.

Have you determined the credit rate with an ST task running on every core vs the max number of MT tasks running concurrently?

Yes, runtimes and credits for 300 tasks of each were recorded and the credit per day calculated based on those tasks.

---

The kind of efficiency I care most about is power efficiency. If MT wastes power, then that matters to me.

It always does, the synchronisation between the threads is additional work. In addition to that, cores are not loaded 100%, there's lots of waiting, you can see that in task manager.

There shouldn't be a lot of overhead associated with multi-threading if each thread has its own hardware thread where it can just park.

I think that the clock speed has a much bigger impact on power efficiency. High clock speeds are less efficient. So reducing the utilization could be beneficial to power efficiency.

 

Have you determined the credit rate with an ST task running on every core vs the max number of MT tasks running concurrently?

Yes, runtimes and credits for 300 tasks of each were recorded and the credit per day calculated based on those tasks.

That's a valuable contribution. I just wanted to get to 5 million credits on this project. I'm going to go back to favoring bio/medicine projects.

Neal

There shouldn't be a lot of overhead associated with multi-threading if each thread has its own hardware thread where it can just park.

Well, there obviously is some bottleneck somewhere, just check how low CPU utilisation is when running one task on all cores.

I think that the clock speed has a much bigger impact on power efficiency. High clock speeds are less efficient. So reducing the utilization could be beneficial to power efficiency.

Yes, high clock speeds are less efficient. That's why I don't let my Ryzen 5700G boost beyond 4 GHz manually via BIOS. The MT-tasks do not downclock your CPU, actually it can boost a lot more compared to other projects because it doesn't run into any limits as fast as on other projects. For maximum efficiency you want low clock and you want to use every cycle, not cores doing nothing part of the time at high clocks and voltages.

---

I think that the clock speed has a much bigger impact on power efficiency. High clock speeds are less efficient. So reducing the utilization could be beneficial to power efficiency.

Yes, high clock speeds are less efficient. That's why I don't let my Ryzen 5700G boost beyond 4 GHz manually via BIOS. The MT-tasks do not downclock your CPU, actually it can boost a lot more compared to other projects because it doesn't run into any limits as fast as on other projects. For maximum efficiency you want low clock and you want to use every cycle, not cores doing nothing part of the time at high clocks and voltages.

I've observed some odd behavior with the MT tasks. Going by the ratio of reported CPU time to wall clock time, it will often use 1 thread less than it's allowed. It might use around 12 threads on average when it's allowed 16, but if it's given 8 threads it will average 7.

Going by the ratio of reported CPU time to wall clock time, it will often use 1 thread less than it's allowed. It might use around 12 threads on average when it's allowed 16, but if it's given 8 threads it will average 7.

Yes, that are basically cores waiting for the other cores to finish their part of the job. The fewer threads you allow per task, the less waiting time you get, with single thread WUs it's zero.

---

Never mind it's back up... :-)