In the era of AI, the demand for GPUs such as the H100 has risen tremendously, making it difficult for an average consumer to acquire one. However, a Reddit user has created a unique solution, converting the AMD Zen 2-based Ryzen 5 4600G "Renoir" APU into a 16 GB GPU and then using it for AI workloads on Linux. You know, as they say, modern problems require modern solutions.
Your Old AMD Ryzen APUs Provide Decent Performance in AI Workloads After Conversion Into A GPU
Before we go into how the APU was converted, let's take a quick recap of the AMD Ryzen 5 4600G. The Ryzen 5 4600G was known to be one of the best APUs in the market after eventually getting replaced by its Cezanne counterpart. It featured a 6C/12T configuration with the Radeon Vega iGPU with seven CUs (Compute Units). To explain how the 16 GB VRAM mark was achieved, it is essential to note that APUs support "Shared Memory", in which you can allocate 50% of your RAM's capacity to your APU. In this case, the Reddit user had 32 GB DDR4 memory onboard, giving half of it to the processor.
The next big hurdle is actually running AI workloads on a Ryzen APU. If you had a desktop GPU, you could utilize AMD's ROCm (Radeon Open Compute) platform to run AI applications on Linux. However, in the case of an iGPU, third-party packages allow ROCm to run on APUs, which was also used here. Using ROCm solves most of your problems since now you can run every kind of AI application ranging from Tensorflow to PyTorch.
In a detailed video, the Reddit user showcased his interesting experiment, claiming that the Ryzen 5 4600G could handle all sorts of AI workloads. However, he only showed testing of Stable Diffusion, and to our surprise, the APU successfully generated a 512x512 image in around one minute and 50 seconds. This is a decent milestone for the APU, and we believe it could prove to be a breakthrough if utilized correctly.
While you can't compare the Ryzen 5 4600G to an AI-meant GPU, such as the H100, this experiment shows the untapped potential in Ryzen APUs. Imagine if instead of the Ryzen 5 4600G, the latest Phoenix APUs were being used here, which are much superior in performance. However, giving out such AI performance for $95 is undoubtedly something to appreciate, and we do hope that it provides the "average consumer" a way to fulfill his AI needs.