项目默认只能使用一张显卡，能否支持多GPU来生成

lzw504504 commented 1 year ago

单GPU显存和速度都有瓶颈，能否支持多GPU加速

lzhchina commented 11 months ago

同问

munabo commented 10 months ago

单GPU显存和速度都有瓶颈，能否支持多GPU加速

你好，我想问一下你的多卡是否实现

lzw504504 commented 10 months ago

单GPU显存和速度都有瓶颈，能否支持多GPU加速

你好，我想问一下你的多卡是否实现

没有

TzyTman commented 7 months ago

python inference.py --driven_audio /home/mk/SadTalker-Video-Lip-Sync-master/input_audio/xiaobao.MP3 --source_video /home/mk/SadTalker-Video-Lip-Sync-master/input_video/xiaobao.mp4 --enhancer lip --time_step 0.5 GPU执行--------------------------------------- ./checkpoints/epoch_20.pth ./checkpoints/auido2pose_00140-model.pth ./checkpoints/auido2exp_00300-model.pth ./checkpoints/facevid2vid_00189-model.pth.tar ./checkpoints/mapping_00109-model.pth.tar 3DMM Extraction for source image landmark Det:: 0%| | 0/223 [00:00<?, ?it/s]Warning: out of memory, sleep for 1s Warning: out of memory, sleep for 1s Warning: out of memory, sleep for 1s Warning: out of memory, sleep for 1s Warning: out of memory, sleep for 1s Warning: out of memory, sleep for 1s

TzyTman commented 7 months ago

CUDA_VISIBLE_DEVICES=0,1 python inference.py --driven_audio /home/mk/SadTalker-Video-Lip-Sync-master/input_audio/xiaobao.MP3 --source_video /home/mk/SadTalker-Video-Lip-Sync-master/input_video/xiaobao.mp4 --enhancer lip --time_step 0.5 添加CUDA 无效好像还是跑的一个GPU

Zz-ww / SadTalker-Video-Lip-Sync

项目默认只能使用一张显卡，能否支持多GPU来生成 #53