timing question

[Alfonso Yamamoto]

Hi, 

I am running a simulation with timstep = 0.1, time scale factor = 20

for with runtime (milliseconds according to the source code) = 10. I would expect the total runtime to be 0.2 second. But on the energy report I get 2 seconds. 

The report I generate with write_energy_report=True. 

What am I missing ? 

Alfonso Yamamoto 

[Donal Fellows]

​

	What am I missing ?

​

The energy report uses the real time used by the simulator, not the simulated timesteps, as might be expected given that energy consumption is an inherently real-time linked concept. The simulator's real-time basic timestep is always 1ms; the assumption of that is baked in very strongly into the time partitioning (e.g., in the assumption of how many machine instructions can be run in that time) so it's not something that's able to be configured.

Instead, you configure how fast the virtual clock ticks and for how much virtual time things run. I'm guessing that the calculation is like this: 10 virtual milliseconds with a combined timestep of 0.1ms and a time scaling factor gives (10 * (1 / 0.1) * 20) 2000 simulator basic timesteps, which is 2 real seconds. I'm not sure that I recommend combining timestep and time scale factor, simply because that can cause confusion (as we see from this conversation!).

The energy report doesn't describe how much energy would be used by the neurons being simulated; we've never had that information or simulated it. If it did, it would indeed report it in the virtual time scale (i.e., it would be for the 10 milliseconds you asked for) but it doesn't as it's a technical measure more than a scientific measure.

Donal.

Donal Fellows

Senior Research Software Engineer

University of Manchester, UK

---

From: spinnak...@googlegroups.com <spinnak...@googlegroups.com> on behalf of Alfonso Yamamoto <alfyama...@gmail.com>
Sent: Friday, October 17, 2025 16:01
To: SpiNNaker Users Group <spinnak...@googlegroups.com>
Subject: [SpiNNaker Mailing List] timing question

 

ZjQcmQRYFpfptBannerEnd

[Christian Brenninkmeijer]

Hi Alfonso,

I completely disagree with Donal here.

The actual runtime was as Alfonso suggested

Using: 

sim.setup(timestep=0.1,  time_scale_factor= 20)

sim.run(10)

This will run for 10 ms

There will be 100 timesteps (runtime / timestep) (10 / 0.1)

Each timestep takes 2 ms  (timestep *  time_scale_factor)  (0.1 * 20)

So the runtime is 200 ms (timestep * timesteps)  (100 * 2)

So 0.2 seconds

Verification.

Ran a simple script with the above parameters

INFO: Application runner took 0:00:01.108488 

(This includes overhead to start and stop)

Looking at the length of V there where 100 values recorded.

Running for 1000 ms

Application runner took 0:00:20.687639 

Here the overhead is less of a factor

So why does the Energy report say 2s

Because it was wrong.

The energy report is a lot simpler than it should be and was never intended to be run with timestep=0.1

Updating the energy report is currently a plan but a low priority

I think a quick fix is

https://github.com/SpiNNakerManchester/SpiNNFrontEndCommon/pull/1312

Christian

---

From: spinnak...@googlegroups.com <spinnak...@googlegroups.com> on behalf of Donal Fellows <donal.k...@manchester.ac.uk>
Sent: 17 October 2025 16:19
To: Alfonso Yamamoto <alfyama...@gmail.com>; SpiNNaker Users Group <spinnak...@googlegroups.com>
Subject: Re: [SpiNNaker Mailing List] timing question

 

--
You received this message because you are subscribed to the Google Groups "SpiNNaker Users Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to spinnakeruser...@googlegroups.com.
To view this discussion, visit https://groups.google.com/d/msgid/spinnakerusers/LO6P265MB696972CD5DA68590B49FF2AA91F6A%40LO6P265MB6969.GBRP265.PROD.OUTLOOK.COM [groups.google.com].

[Alfonso Yamamoto]

Hi, 

Thanks for the quick response.

Just one question.

Are the values for energy consumption affected by this issue ? 

Alfonso