More number of thread = slower #860
-
|
I have installed opengate through
I notice that if I set
What I notice is that it seems more thread lead to the simulation being repeated as if the source Code is quite simple just like this: sim = gate.Simulation()
# Basic simulation configuration
add_materials(sim)
sim.check_volumes_overlap = True
sim.world.size = [WORLD_RADIUS * 2, WORLD_RADIUS * 2, WORLD_DEPTH]
sim.world.material = "Air"
sim.world.color = [1, 0, 1, 1] # invisible
... # adding 8 detector discretion panels with the help of get_grid_repetition and then get_circular_repetition from opengate.geometry.utility
# Simulation parameters
sim.number_of_threads = num_thread
sim.random_seed = 123456
# Visualization (disable for production runs)
sim.visu = visu
# Run parameters (for actual simulation, not just visualization)
# Uncomment and adjust for production:
sim.run_timing_intervals = [(0, sim_time * sec)] # 1 second acquisition
# sim.number_of_events = 1e4 # Number of primary particles
print("\nStarting simulation...")
print("Close visualization window to end.")
sim.run()
print("Simulation complete!")I am using v10.0.2 as I can't build the core for 10.0.3 |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments 5 replies
-
|
yes you should divide the activity or the number of events by number_of_threads ;) |
Beta Was this translation helpful? Give feedback.
-
|
Thanks! That was quick! I was planning to do so and is reassured that's the correct way. |
Beta Was this translation helpful? Give feedback.
-
|
exact, I thought is was somewhere in the doc, but indeed, I cannot find where ;) The doc is also open to contribution: do not hesitate to propose a PR to improve it. Thanks |
Beta Was this translation helpful? Give feedback.
-
|
Hello @alabamagan To complete, if your computer has only 12 cores, when your run 48 threads, the processor need to wait the 12 firsts threads finished to run the following. It explains the growth of run-time in your question. Another question, you install opengate using pip but I do not understand why you "can't build the core for 10.0.3" |
Beta Was this translation helpful? Give feedback.
-
|
Hi @tbaudier, I do have a workstation with 64 cores and an M2 SSD for data, so have ruled out hardware issues. For the versioning, when I do I tried to clone the repo and build it in the |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for the report. pip install opengate==10.0.3 does not work in your case because I can see you use Linux x64 with python 3.11 and gcc 9.4.0. Because we use Qt6, we need to have a quite recent version of gcc, gcc 14 in our case. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks @tbaudier, Yes I've got VTIK and ITK toolchain installed, but I failed to do the same Geant4, probably the same reason with the gcc version, which is why I've been using the docker geant and gate v9.0 before I moved to gate 10. There was some tweaks needed including:
Curious to know if opengate 10 can use gean4 in the docker as its core? |
Beta Was this translation helpful? Give feedback.
-
|
For the docker with Gate9 installed, I have no idea, I did not try. But if your are familiar with Docker, you can use this one https://hub.docker.com/r/tbaudier/opengate_core/tags This is the docker image used to compile the wheel in the Github CI. v11.3.2 is the latest. |
Beta Was this translation helpful? Give feedback.
yes you should divide the activity or the number of events by number_of_threads ;)
Note that the speedup due to MT is not always linear, it depends on the actors. Moreover, it is not very efficient when multiple runs are used (which does not seems the case here).