search

limbo018 / DREAMPlace

Deep learning toolkit-enabled VLSI placement
BSD 3-Clause "New" or "Revised" License
714 stars 205 forks source link

run without gpu #153

Closed mavoyang closed 1 year ago

mavoyang commented 1 year ago

How can I configure CMAKE to make Dreamplacer run without GPU???

To work with other projects, I need to build version 2.2.0 of Dreamplacer. I first built the Docker container, and I can find Cuda 10.0 in it. Then, if I run the Docker container with GPU, I can successfully 'make'. But running any benchmark with 'placer.py' gets stuck at '[INFO] DREAMPlace - move cells to the center of the layout with random noise.' 截屏2023-10-17 10 40 06

This may be due to the incompatibility between the old version of Dreamplacer and the current CUDA. So, I tried running the Docker container without GPU, but running any benchmark with 'placer.py' results in an error 'Found no NVIDIA driver on your system' after 'move cells to the center of the layout with random noise.' 截屏2023-10-17 11 17 34

How should I configure CMake to build a Dreamplacer that doesn't use CUDA at all?" Or how can I fix it to run with Cuda?

One more thing, for the newest Dreamplacer, it runs with GPU well when I use the provided docker.

limbo018 commented 1 year ago

You can modify the gpu parameter in the JSON input file to 0 for CPU.

mavoyang commented 1 year ago

Ha, it's really a rookie question. Thanks!