Why is it running so slow, so stuck?

[Logic-TARS]

I tested it twice, and almost every time I had to wait nearly 15 minutes to get a normal frame rate. It's the same whether I use gs.cuda or gs.cpu. Is anyone else experiencing this? this is hello_genesis example.

[zhouxian]

This seems pretty weird. What platform and hardware are you on?

[zhouxian]

Also, if you are using a light terminal background, you can change the theme to 'light' in gs.init()

[Logic-TARS]

This seems pretty weird. What platform and hardware are you on?

H800 and Ubuntu 22.04

[Logic-TARS]

This seems pretty weird. What platform and hardware are you on?

Does this have anything to do with my use of a remote server? I use VNC for remote desktop connections，the server itself does not have a display

[Logic-TARS]

Also, if you are using a light terminal background, you can change the theme to 'light' in gs.init()另外，如果您使用浅色终端背景，您可以在 gs.init() 中将主题更改为“浅色”

You can see that the GPU is basically unresponsive.

[zhouxian]

This is indeed a bit weird and I don't have a clue yet... it's probably caused by the way you transfer the rendered image back

[sudhir2016]

I was trying Hello_Genesis on Google Colab. The building visualizer step takes about 20 minutes regardless of whether I use CPU or GPU.

[ziqiyang107]

I got the exactly same issue, FPS is pretty low. I'm also in a remote 4090 server docker, with ubuntu 22.04, python 3.9, cuda 12.4

[renyu2016]

same issue too

[ofirbartal100]

same

[yic03685]

I have the same issue but I think what @zhouxian mentioned is valid. This only happens when I enabled camera.render. It seems like it forces the rendered image to back on cpu so it slows the simulation dramatically. I have no figured out how to avoid this.

[soooocold]

I have the same issue but I think what @zhouxian mentioned is valid. This only happens when I enabled camera.render. It seems like it forces the rendered image to back on cpu so it slows the simulation dramatically. I have no figured out how to avoid this. same error，when run cam.render() ，I noticed that the CPU utilization is 100% but the GPU utilization is 0，and very slowly

[soooocold]

I sloved this issue by upgrade the CUDA version to 12.6 NVIDIA-SMI 560.35.05 Driver Version: 560.35.05 CUDA Version: 12.6

[Taka-Hashimoto]

I encountered the same issue on Google Colab, but I was able to resolve it by adding the following code before importing genesis:

import os
NVIDIA_ICD_CONFIG_PATH = '/usr/share/glvnd/egl_vendor.d/10_nvidia.json'
ICD_CONFIG_CONTENT = """{
    "file_format_version" : "1.0.0",
    "ICD" : {
        "library_path" : "libEGL_nvidia.so.0"
    }
}
"""
with open(NVIDIA_ICD_CONFIG_PATH, 'w') as f:
    f.write(ICD_CONFIG_CONTENT)

You can find the complete code here: Gist link

[ZimengYuan]

I encountered the same issue on Google Colab, but I was able to resolve it by adding the following code before importing genesis:我在 Google Colab 上遇到了同样的问题，但我能够通过在导入 genesis 之前添加以下代码来解决它：

import os
NVIDIA_ICD_CONFIG_PATH = '/usr/share/glvnd/egl_vendor.d/10_nvidia.json'
ICD_CONFIG_CONTENT = """{
    "file_format_version" : "1.0.0",
    "ICD" : {
        "library_path" : "libEGL_nvidia.so.0"
    }
}
"""
with open(NVIDIA_ICD_CONFIG_PATH, 'w') as f:
    f.write(ICD_CONFIG_CONTENT)

You can find the complete code here: Gist link您可以在此处找到完整的代码： Gist 链接

I tried your method on my cloud server, but it doesn't seem to work.