On the Hammer Software site, you'll find many bundles of generative-AI tools. These bundles usually include runtime environments, but many users still prefer to deploy themselves. However, they may hit runtime errors. The reason is mostly missing key runtime support.
To help solve these, we've compiled essential tools, frameworks, and plugins. This article focuses on several basic programs needed to run generative AI, with a detailed install tutorial - so newbies don't get headaches over tech hurdles.
For generative AI, these three tools are unavoidable base support:
1. CUDA (Compute Unified Device Architecture)
CUDA is a parallel-computing platform and programming model from NVIDIA, built to fully harness NVIDIA GPU computing power. It lets programs offload heavy computation from CPU to GPU, greatly boosting efficiency.
– Core role: use the GPU instead of CPU for complex operations.
– Use cases: suited for efficiently processing big data or running compute-heavy tasks.
Simply put, CUDA makes the GPU "come alive"; in the generative-AI world, processing speed jumps a notch.
2. cuDNN
Next we introduce this essential deep-learning acceleration tool and how it seamlessly integrates with CUDA.
3. FFmpeg
An A/V processing tool that solves multimedia-data needs in AI projects.
Given that many users have no programming experience, our tutorial is zero-threshold - just follow the steps. It not only makes your generative AI run smoothly but also helps you understand what these tools actually do.
1. Install the NVIDIA graphics driver
First confirm you're using an NVIDIA GPU and the driver is installed. Driver installation isn't covered here; use this link to download and install the NVIDIA driver (https://www.nvidia.cn/drivers/lookup/).

2. Determine the installable CUDA version
① After installing the NVIDIA driver, right-click an empty spot on the desktop and choose "NVIDIA Control Panel."

② In the NVIDIA Control Panel, click "System Information" at bottom-left; in the System Information window click "Components." On the third row you'll see the maximum CUDA version your GPU supports.

3. Download CUDA
① Confirm support, then download the matching CUDA version from NVIDIA's site. Run the installer and follow prompts.

② In the Windows search box at bottom-left, search for "environment variables" and choose "Edit the system environment variables."

③ In the System Properties dialog, click "Environment Variables." Under System variables you'll see CUDA variables auto-added; double-click "Path" and confirm the relevant paths are there. If OK, click OK three times, then reboot.

④ If environment variables weren't auto-added, click "New" and manually paste the CUDA install path. After setting, click OK three times and reboot.
These are:
CUDA_PATH: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6
CUDA_PATH_V12_6: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\bin
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\libnvvp

5. Verify CUDA installed successfully
① Hold "Win" and press "R"; in Run, type "cmd" and click OK to open the command prompt (a black code window).

② In the code window type "nvcc --version" and press Enter. If CUDA version info appears, CUDA installed successfully.

Detailed illustrated cuDNN install tutorial
Detailed illustrated FFmpeg install tutorial