Nicholai/jan
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #1405 from janhq/docs/improve-gpu-not-used

Browse Source 
docs: improve gpu not used guide
This commit is contained in:
Hieu 2024-01-06 14:19:57 +07:00 committed by GitHub
commit a1e9e0e5f1
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 151 additions and 79 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

124
docs/docs/guides/07-troubleshooting/03-gpu-not-used.mdx
View File

@ -11,16 +11,26 @@ keywords: [
no-subscription fee,
large language model,
troubleshooting,
using GPU,
]
---
## Requirements for running Jan App in GPU mode on Windows and Linux
- You must have an NVIDIA driver that supports CUDA 11.4 or higher. Refer [here](https://docs.nvidia.com/deploy/cuda-compatibility/index.html#binary-compatibility__table-toolkit-driver).
To check if the NVIDIA driver is installed, open PowerShell or Terminal and enter the following command:
This guide provides steps to troubleshoot and resolve issues when Jan app does not utilize the GPU on Windows and Linux systems.
## Requirements for Running Jan in GPU Mode on Windows and Linux
### NVIDIA Driver
Ensure that you have installed the NVIDIA driver that supports CUDA 11.4 or higher. For a detailed of CUDA compatibility, please refer [here](https://docs.nvidia.com/deploy/cuda-compatibility/index.html#binary-compatibility__table-toolkit-driver).
To verify, open PowerShell or Terminal and enter the following command:
```bash
nvidia-smi
```
If you see a result similar to the following, you have successfully installed the NVIDIA driver:
```bash
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.57.02 Driver Version: 470.57.02 CUDA Version: 11.4 |
@ -35,12 +45,18 @@ keywords: [
+-------------------------------+----------------------+----------------------+
```
- You must have CUDA 11.4 or higher (refer [here](https://developer.nvidia.com/cuda-toolkit-archive)).
To check if CUDA is installed, open PowerShell or Terminal and enter the following command:
### CUDA Toolkit
Ensure that you have installed the CUDA toolkit that is compatible with your NVIDIA driver. For a detailed of CUDA compatibility, please refer [here](https://docs.nvidia.com/deploy/cuda-compatibility/index.html#binary-compatibility__table-toolkit-driver).
To verify, open PowerShell or Terminal and enter the following command:
```bash
nvcc --version
```
If you see a result similar to the following, you have successfully installed CUDA:
```bash
nvcc: NVIDIA (R) Cuda compiler driver
@ -48,8 +64,10 @@ keywords: [
Build cuda_11.4.r11.4/compiler.30033411_0
```
- Specifically for Linux:
- you must have `gcc-11`, `g++-11`, `cpp-11` or higher, refer [here](https://gcc.gnu.org/projects/cxx-status.html#cxx17). For Ubuntu, you can install g++ 11 by following the instructions [here](https://linuxconfig.org/how-to-switch-between-multiple-gcc-and-g-compiler-versions-on-ubuntu-20-04-lts-focal-fossa).
### Specific Requirements for Linux
**GCC and G++ Version**: Ensure that you have installed `gcc-11`, `g++-11`, `cpp-11` or higher, refer [here](https://gcc.gnu.org/projects/cxx-status.html#cxx17). For Ubuntu, you can install g++ 11 by following the instructions [here](https://linuxconfig.org/how-to-switch-between-multiple-gcc-and-g-compiler-versions-on-ubuntu-20-04-lts-focal-fossa).
```bash
# Example for ubuntu
# Add the following PPA repository
@ -69,7 +87,9 @@ keywords: [
# Check the default g++ version
g++ --version
```
- You must add the `.so` libraries of CUDA to the `LD_LIBRARY_PATH` environment variable, refer [here](https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#post-installation-actions).
**Post-Installation Actions**: You must add the `.so` libraries of CUDA to the `LD_LIBRARY_PATH` environment variable by following the [Post-installation Actions instruction](https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#post-installation-actions).
```bash
# Example for ubuntu with CUDA 11.4
sudo nano /etc/environment
@ -80,38 +100,90 @@ keywords: [
# Save and exit
# Restart your computer or log out and log in again, the changes will take effect
```
## How to switch mode CPU/GPU Jan app
By default, Jan app will run in CPU mode. When starting Jan app, the program will automatically check if your computer meets the requirements to run in GPU mode. If it does, we will automatically enable GPU mode and pick the GPU has highest VGRAM for you (feature allowing users to select one or more GPU devices for use - currently in planning). You can check whether you are using CPU mode or GPU mode in the settings/advance section of Jan app. (see image below). ![](../../../static/img/usage/jan-gpu-enable-setting.png)
## Switching Between CPU/GPU Modes in Jan
If you have GPU mode but it is not enabled by default, the following possibilities may exist, you can follow the next steps to fix the error:
By default, Jan runs in CPU mode. Upon start, Jan checks if your system is capable of running in GPU mode. If compatible, GPU mode is enabled automatically, and the GPU with the highest VRAM is selected. This setting can be verified in the `Settings` > `Advanced` section.
1. You have not installed the NVIDIA driver, refer to the NVIDIA driver that supports CUDA 11.4 [here](https://docs.nvidia.com/deploy/cuda-compatibility/index.html#binary-compatibility__table-toolkit-driver).
![jan-gpu-enable-setting](../../../static/img/usage/jan-gpu-enable-setting.png)
2. You have not installed the CUDA toolkit or your CUDA toolkit is not compatible with the NVIDIA driver, refer to CUDA compatibility [here](https://docs.nvidia.com/deploy/cuda-compatibility/index.html#binary-compatibility__table-toolkit-driver).
If you find that GPU mode is available but not enabled by default, consider the following troubleshooting steps:
3. You have not installed a CUDA compatible driver, refer [here](https://docs.nvidia.com/deploy/cuda-compatibility/index.html#binary-compatibility__table-toolkit-driver), and you must add the `.so` libraries of CUDA and the CUDA compatible driver to the `LD_LIBRARY_PATH` environment variable, refer [here](https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#post-installation-actions). For Windows, add the `.dll` libraries of CUDA and the CUDA compatible driver to the `PATH` environment variable. Usually, when installing CUDA on Windows, this environment variable is automatically added, but if you do not see it, you can add it manually by referring [here](https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html#environment-setup).
:::tip
## To check the current GPU-related settings that Jan app has detected, you can go to the Settings/Advanced section as shown in the image below.
![](../../../static/img/usage/jan-open-home-directory.png)
![](../../../static/img/usage/jan-open-settings-1.png)
![](../../../static/img/usage/jan-open-settings-2.png)
![](../../../static/img/usage/jan-open-settings-3.png)
1. Check if you have installed the NVIDIA driver that supports CUDA 11.4 or higher. For a detailed of CUDA compatibility, please refer [here](https://docs.nvidia.com/deploy/cuda-compatibility/index.html#binary-compatibility__table-toolkit-driver).
When you have an issue with GPU mode, share the `settings.json` with us will help us to solve the problem faster.
2. Ensure that the CUDA toolkit is installed and compatible with your NVIDIA driver. For a detailed of CUDA compatibility, please refer [here](https://docs.nvidia.com/deploy/cuda-compatibility/index.html#binary-compatibility__table-toolkit-driver).
## Tested on
3. For Linux, it's crucial to add the `.so` libraries of CUDA and the CUDA compatible driver to the `LD_LIBRARY_PATH` environment variable. For Windows, users should ensure that the `.dll` libraries of CUDA and the CUDA compatible driver are included in the PATH environment variable. Usually, when installing CUDA on Windows, this environment variable is automatically added, but if you do not see it, you can add it manually by referring [here](https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html#environment-setup).
:::
## Checking GPU Settings in Jan
1. To check the current GPU settings detected by Jan, navigate to `Settings` > `Advanced` > `Open App Directory`
<br></br>
![OpenAppDirectory](../../../static/img/usage/jan-open-home-directory.png)
<br></br>
2. Open the `settings.json` file under the `settings` folder. The following is an example of the `settings.json` file:
<br></br>
```json title="~/jan/settings/settings.json"
{
"notify": true,
"run_mode": "gpu",
"nvidia_driver": {
"exist": true,
"version": "531.18"
},
"cuda": {
"exist": true,
"version": "12"
},
"gpus": [
{
"id": "0",
"vram": "12282"
},
{
"id": "1",
"vram": "6144"
},
{
"id": "2",
"vram": "6144"
}
],
"gpu_highest_vram": "0"
}
```
:::tip
Troubleshooting tips:
- Ensure the `nvidia_driver` and `cuda` fields indicate that requirements software are installed.
- If the `gpus` field is empty or does not list your GPU, verify the installation of the NVIDIA driver and CUDA toolkit.
- For further assistance, please share the `settings.json` with us.
:::
## Tested Configurations
- Windows 11 Pro 64-bit, NVIDIA GeForce RTX 4070ti GPU, CUDA 12.2, NVIDIA driver 531.18 (Bare metal)
- Ubuntu 22.04 LTS, NVIDIA GeForce RTX 4070ti GPU, CUDA 12.2, NVIDIA driver 545 (Bare metal)
- Ubuntu 18.04 LTS, NVIDIA GeForce GTX 1660ti GPU, CUDA 12.1, NVIDIA driver 535 (Proxmox VM passthrough GPU)
- Ubuntu 20.04 LTS, NVIDIA GeForce GTX 1660ti GPU, CUDA 12.1, NVIDIA driver 535 (Proxmox VM passthrough GPU)
- Ubuntu 18.04 LTS, NVIDIA GeForce GTX 1660ti GPU, CUDA 12.1, NVIDIA driver 535 (Proxmox VM passthrough GPU)
## If you are experiencing this issue
## Common Issues and Solutions
1. If the issue persists, please install the [Nightly version](/install/nightly) instead.
1. If the issue persists, ensure your (V)RAM is accessible by the application. Some folks have virtual RAM and needs additional configuration.
2. If the issue persists, ensure your (V)RAM is accessible by the application. Some folks have virtual RAM and need additional configuration.
1. Get help in [Jan Discord](https://discord.gg/mY69SZaMaC).
3. Get help in [Jan Discord](https://discord.gg/mY69SZaMaC).