CUDA installation
- CUDA Installation
This article details the installation of the Compute Unified Device Architecture (CUDA) toolkit on our servers. CUDA is a parallel computing platform and programming model developed by NVIDIA, and is essential for utilizing the GPU for accelerated computing tasks. This guide is geared towards new server engineers and assumes a basic understanding of Linux command-line operations. It covers installation on a Debian 11 (Bullseye) system, which is our standard server OS. Adjustments may be needed for other distributions. We will address download, installation, environment variable setup, and verification. See also GPU Acceleration Overview and Server Operating System Installation for related information.
Prerequisites
Before beginning the CUDA installation, ensure the following prerequisites are met:
- A supported NVIDIA GPU is installed on the server. Check GPU Hardware List for compatibility.
- The server is running Debian 11 (Bullseye) with root or sudo access.
- A stable internet connection is available for downloading the CUDA toolkit.
- Ensure the system is fully updated. Run `sudo apt update && sudo apt upgrade`.
- Verify the presence of a compatible GCC compiler (version 7.5 or higher is recommended). Check with `gcc --version`. If necessary, install a suitable GCC version using `sudo apt install gcc g++`. See Compiler Installation Guide for details.
Downloading the CUDA Toolkit
The CUDA toolkit can be downloaded from the NVIDIA developer website. A developer account is required. Navigate to the CUDA Toolkit Archive: [1](https://developer.nvidia.com/cuda-toolkit-archive). Select the appropriate version for your GPU and operating system. For this guide, we will use CUDA 12.2.
Download the Debian/Ubuntu local installer (e.g., `cuda_12.2.2_535.154.05_linux.run`). We recommend downloading directly to the `/tmp` directory.
Installation Process
1. **Navigate to the download directory:**
```bash cd /tmp ```
2. **Make the installer executable:**
```bash chmod +x cuda_12.2.2_535.154.05_linux.run ```
3. **Run the installer:**
```bash sudo ./cuda_12.2.2_535.154.05_linux.run ```
The installer will present a series of prompts. Accept the license agreement, and choose the installation directory. The default is `/usr/local/cuda-12.2`. It is recommended to keep the default. The installer will also ask if you want to install the NVIDIA drivers. If you already have the latest drivers installed (verified through `nvidia-smi`), decline this option. See NVIDIA Driver Installation for driver details.
Post-Installation Configuration
After the installation completes, several environment variables need to be set up to ensure the CUDA toolkit is properly accessible.
1. **Add CUDA paths to the `~/.bashrc` file:**
Open the `~/.bashrc` file using a text editor (e.g., `nano ~/.bashrc`) and add the following lines at the end of the file:
```bash
export PATH=/usr/local/cuda-12.2/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-12.2/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
```
These lines add the CUDA binary directory and library directory to the system's PATH and LD_LIBRARY_PATH, respectively.
2. **Source the `~/.bashrc` file:**
Run the following command to apply the changes:
```bash source ~/.bashrc ```
CUDA Toolkit Details
The CUDA toolkit includes various components. Here's a table outlining key components and their functionalities:
| Component | Description |
|---|---|
| nvcc | The NVIDIA CUDA Compiler. Used to compile CUDA C/C++ code. |
| cuda-gdb | The NVIDIA CUDA Debugger. Used for debugging CUDA applications. |
| cuda-memcheck | A memory checker for CUDA applications. |
| cuBLAS | A library for Basic Linear Algebra Subroutines (BLAS). |
| cuFFT | A library for Fast Fourier Transforms (FFT). |
Supported CUDA Versions & Hardware
The following table details the compatibility between CUDA versions and supported NVIDIA GPUs. Note that this is not exhaustive, and compatibility may vary.
| CUDA Version | Supported GPU Architecture | Example GPU |
|---|---|---|
| 11.x | Turing, Ampere | GeForce RTX 2080, GeForce RTX 3090 |
| 12.x | Ada Lovelace, Hopper | GeForce RTX 4090, NVIDIA H100 |
| 13.x (Beta) | Blackwell | NVIDIA B100 (Future) |
Verification
To verify the installation, you can run the CUDA samples. The samples are located in the `/usr/local/cuda-12.2/samples` directory.
1. **Navigate to the samples directory:**
```bash cd /usr/local/cuda-12.2/samples ```
2. **Compile the deviceQuery sample:**
```bash make ```
3. **Run the deviceQuery executable:**
```bash ./deviceQuery ```
If the installation was successful, the `deviceQuery` program will display information about your NVIDIA GPU, including its name, compute capability, and memory size. See Troubleshooting CUDA Installation if issues arise.
Additional Resources
- CUDA C++ Programming Guide
- NVIDIA Documentation
- GPU Resource Monitoring
- Parallel Processing Concepts
System Specifications
Here's a table showing our standard server specifications for CUDA development:
| Component | Specification |
|---|---|
| CPU | Intel Xeon Gold 6248R |
| RAM | 128 GB DDR4 ECC |
| GPU | NVIDIA GeForce RTX 3090 |
| Storage | 2 TB NVMe SSD |
| Operating System | Debian 11 (Bullseye) |
Intel-Based Server Configurations
| Configuration | Specifications | Benchmark |
|---|---|---|
| Core i7-6700K/7700 Server | 64 GB DDR4, NVMe SSD 2 x 512 GB | CPU Benchmark: 8046 |
| Core i7-8700 Server | 64 GB DDR4, NVMe SSD 2x1 TB | CPU Benchmark: 13124 |
| Core i9-9900K Server | 128 GB DDR4, NVMe SSD 2 x 1 TB | CPU Benchmark: 49969 |
| Core i9-13900 Server (64GB) | 64 GB RAM, 2x2 TB NVMe SSD | |
| Core i9-13900 Server (128GB) | 128 GB RAM, 2x2 TB NVMe SSD | |
| Core i5-13500 Server (64GB) | 64 GB RAM, 2x500 GB NVMe SSD | |
| Core i5-13500 Server (128GB) | 128 GB RAM, 2x500 GB NVMe SSD | |
| Core i5-13500 Workstation | 64 GB DDR5 RAM, 2 NVMe SSD, NVIDIA RTX 4000 |
AMD-Based Server Configurations
| Configuration | Specifications | Benchmark |
|---|---|---|
| Ryzen 5 3600 Server | 64 GB RAM, 2x480 GB NVMe | CPU Benchmark: 17849 |
| Ryzen 7 7700 Server | 64 GB DDR5 RAM, 2x1 TB NVMe | CPU Benchmark: 35224 |
| Ryzen 9 5950X Server | 128 GB RAM, 2x4 TB NVMe | CPU Benchmark: 46045 |
| Ryzen 9 7950X Server | 128 GB DDR5 ECC, 2x2 TB NVMe | CPU Benchmark: 63561 |
| EPYC 7502P Server (128GB/1TB) | 128 GB RAM, 1 TB NVMe | CPU Benchmark: 48021 |
| EPYC 7502P Server (128GB/2TB) | 128 GB RAM, 2 TB NVMe | CPU Benchmark: 48021 |
| EPYC 7502P Server (128GB/4TB) | 128 GB RAM, 2x2 TB NVMe | CPU Benchmark: 48021 |
| EPYC 7502P Server (256GB/1TB) | 256 GB RAM, 1 TB NVMe | CPU Benchmark: 48021 |
| EPYC 7502P Server (256GB/4TB) | 256 GB RAM, 2x2 TB NVMe | CPU Benchmark: 48021 |
| EPYC 9454P Server | 256 GB RAM, 2x2 TB NVMe |
Order Your Dedicated Server
Configure and order your ideal server configuration
Need Assistance?
- Telegram: @powervps Servers at a discounted price
⚠️ *Note: All benchmark scores are approximate and may vary based on configuration. Server availability subject to stock.* ⚠️