Jump to content
Sam Marginson

simultaneous GPU & CPU runs - run time impacts

Recommended Posts

Hi Sam,


There are two hardware factors that could be impacting your run times:

1) When you start a GPU run, TUFLOW compiles the model on the CPU, as per normal, before passing it to the GPU. The model then stays on the GPU, except at output times, where it passes information back to the CPU to get written. Thus, when you are running a GPU model run, you take up one GPU and a part of the CPU. Depending on the output frequency and size of the model, how much of the CPU you take up can be substantial. When I run a GPU model, I like to keep a CPU core free for this purpose.

2) TUFLOW GPU runs need both CPU RAM and GPU RAM for the process I've described above. Each run needs about 5 times more CPU RAM than GPU RAM, this has been optimised in the 2016-03-AD release, so check which version you're running.  If you exceed the amount of GPU RAM available on the card, the model wont run. But if you exceed the amount of CPU RAM available on your computer, you will move to virtual memory, which is very slow for models! Check in your tlf or in the task manager how much memory your models need each.


So, if you've got a 4 CPU core machine with 64 GB of CPU RAM and 2 GPU cards with 2GB of GPU RAM each, you'll be able to run 4 models, 2 of which could be GPU. The total amount of RAM required for the models should not exceed 64GB and each GPU model could take up 2GB of GPU RAM each.

I hope that makes sense. Others who are hardware minded feel free to weigh in!

Share this post

Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

  • Create New...