Question-and-Answer Resource for the Building Energy Modeling Community
Get s tarted with the Help page
Ask Your Question
1

GPU specs to speed up PixelCounting

asked 2021-09-21 08:27:51 -0500

rafael.alonso gravatar image

updated 2021-09-21 15:03:42 -0500

Hi all,

I recently realised that the lack of dedicated GPU in some of our machines is slowing down our E+ simulations by 50% or more. We normally rely on Pixel Counting to overcome the limitations with solar distribution in concave zones (which is great). I think Pixel Counting runs on the GPU and slows down the simulation, so we are looking into installing dedicated graphic cards in our machines.

As with everything, the range of graphic cards and their cost is considerable, so I was wondering if anyone had some tips on which one is good enough, or at which specs it is not worth to keep investing.

For reference, we run simulations with about 50-100 zones or so which often take 0.5-3h, on Intel i7/i9 about 1-2 years old.

Thanks, Rafael

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted
1

answered 2021-09-28 16:36:48 -0500

The GPU accelerated PixelCounting method is most useful for models with large number of exterior solar shading surfaces (see slide 12 here). There is a difference in performance with a good dedicated GPU, however, an integrated GPU still does much better than the CPU-based method for large number of solar shading surfaces (seen in the previously referenced slide, 12 min vs 50 min for dedicated vs integrated respectively). There is a threshold (about 128-1024 shading surfaces depending on GPU) where the traditional CPU-based method is still slightly faster than the GPU accelerated PixelCounting approach.

Are you sure the lack of a dedicated GPU is slowing down the E+ simulations by 50% or more? A dedicated GPU will improve simulation time over an integrated GPU only if you have a large number (1000+) of exterior solar shading surfaces OR a large number (100s) of interior surfaces within a zone. Otherwise, it is likely there is some other issue slowing down your simulations unrelated to PixelCounting and the GPU.

If it is the lack of a dedicated GPU slowing down your simulations then I would find the most recent gaming or workstation GPU available to you in your price range. I have not done a large parametric study to see how different GPUs affect simulation time and what is the most sensitive parameter on the GPU (number of CUDA cores, memory bandwidth, PCIe connection, GB of memory, etc.). But most dedicated gaming or workstation GPUs will be better than an integrated GPU.

edit flag offensive delete link more

Comments

Thanks for the details Mark! I'll make some tests with a GPU I can borrow and report back.

rafael.alonso gravatar image rafael.alonso  ( 2021-09-29 11:11:01 -0500 )edit

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

 

Question Tools

1 follower

Stats

Asked: 2021-09-21 08:27:51 -0500

Seen: 83 times

Last updated: Sep 28