Open makintosz opened 5 years ago
me too, very slow in GPU 980TI
Any progress on this? I'm also experiencing this and it just seems like the GPU isn't being utilized at all:
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.100 Driver Version: 440.100 CUDA Version: 10.2 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce GTX 980 Ti Off | 00000000:02:00.0 On | N/A |
| 0% 56C P8 16W / 275W | 610MiB / 6080MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 888 G /usr/lib/xorg/Xorg 30MiB |
| 0 2628 G /usr/lib/xorg/Xorg 110MiB |
| 0 2868 G /usr/bin/gnome-shell 113MiB |
| 0 40642 G ...AAAAAAAAAAAACAAAAAAAAAA= --shared-files 240MiB |
| 0 67185 C python3.7 90MiB |
+-----------------------------------------------------------------------------+
System information
Describe the problem
Hi, my problem is that when using pretrained coco model from the tutorial delivered with object detection API, the detection on exemplary pictures takes too much time. I checked the GPU usage and it does not move at all, instead CPU is used. My tensorflow-GPU installation should be OK, since I use it with different scripts and it works well. I tried installing it on Ubuntu 16.04 and it was the same. I tried different CUDA Toolkit, CUDNN versions and TF versions (mentioned in the System Information section) and still nothing (CuDNN 7.05 didn't work due to another error problem, fixed by using newer versions). My project however requires faster detection time.