GPU not utilized with Honeybee ASE simulation

[ebrobe...@gmail.com]

Hi Nathaniel, 

This is my first attempt at using Accelerad and it seems promising; however, I cannot seem to get the GPU running in HoneyBee. I installed everything according to your instructions into the Radiance Bin and Lib folders. When I run your demo files, it seems that everything is fine; however when running 'AcceleradRT' I do not see my GPU being used more than 2-3% or less. 

My Raypath seems fine. I am using an NVIDIA RTX 5000. 

I am clearly missing something here. Also, I attempted to add the radiance parameter '-g+' (which should be the default) just to make sure its not using the CPU. Is there any diagnostic I can run to make sure the pipeline is setup properly? Also, do I need to install cuda toolkit? There seems to be an ERR with my GPU when I look at it in the CLI. 

Thank you, 

Erik

[Nathaniel Jones]

Hi Erik,

There are a couple of points to address here, so I will handle them one at a time:

1. As noted in a previous post, most Accelerad jobs (for example accelerad_rpict or accelerad_rtrace) are too quick to register in the Task Manager. If you want to know whether the GPU is being used, look for the line "OptiX x.x.x found display driver xxx.xx, CUDA driver x.x.x, and x GPU device(s)" in the command line output. Any simulation that displays this line is running on the GPU.

2. AcceleradRT (the GUI program included with Accelerad 0.7) only runs on the GPU. It does not matter if you provide -g+ because there is no CPU equivalent or fallback. The GPU utilization will be proportional to the model size and number of ray relays and ambient bounces. The demo file model included with the installation package is tiny, so it should not result in noticeable GPU usage. AcceleradRT's GPU use will show up in the Task Manager because it places a constant load on the GPU.

3. The Grasshopper screenshot you shared shows a Daysim simulation using rtrace_dc. As noted in this post, rtrace_dc is not supported by Accelerad. If you wish to run annual simulations with Accelerad, use the newer 3-phase or 5-phase methods, which are supported by Honeybee+.

I hope this answers your questions.

Nathaniel

[ebrobe...@gmail.com]

Thanks for the quick reply Nathaniel, 

Your responses make sense and I apologize for not thoroughly checking the forum as I am new to accelerad. I will take a deeper look. 

For HB Legacy, the batch file is now recognizing my GPU; however, when I run the ANN sim in HB+ I am not getting much of a speed increase even though I have 1200 grid points. I suppose my simulation is possibly too small or not complex enough.  

I appreciate the insight. 

Erik

[Nathaniel Jones]

Hi Erik,

For the annual simulation in Honeybee+, it may be that most of the time is going into dctimestep. Taking a look at the command line output could reveal how much time goes into the ray tracing itself.

Nathaniel

[Mostapha Sadeghipour]

Hi Erik and Nathaniel,

1. I do not recommend to use 3 or 5 phase unless you really know what is going on in those simulations. The annual recipe (AKA enhanced 2-Phase) is the safest option.
2. For comparison you should only time the run of the commands. The component time includes time for writing the model and time for loading the results. Specifically the second one turned out to be a bottleneck in Grasshopper. This will be fixed in the upcoming release but I just wanted you to know.
3. The commands can be enhanced by piping the results between them instead of writing them to text files which Honeybee[+] does right now. See here: (https://discourse.radiance-online.org/t/is-there-an-rcalc-for-dctimestep/4603/2) This will also be addressed in the upcoming release.
4. Better support for Accelerad from Honeybee has been on our list for some time now. It will not happen in this release but should happen in the future.
   

PS: This will be up to Nathaniel to make the final decision but I think it will be easier for more people to get involved and answer to question if we use the Accelerad category on the Radiance forum: https://discourse.radiance-online.org/c/accelerad/13