Upgrading to $95 AMD CPU with 16GB GPU for AI

As one of the best CPUs for gaming, the Ryzen 5 5600G (Cezanne) has supplanted the Ryzen 5 4600G (Renoir). To run AI applications on Linux, a hack has given the Ryzen 5 4600G new life by turning the low-cost Zen 2 APU into a 16GB graphics card.

AMD CPU with 16GB GPU

Not everyone has the financial means to rent or purchase an Nvidia H100 (Hopper) to conduct AI experiments. Even if you have the money, you might not be able to get a graphics card that focuses on AI because of the current demand. Fortunately, AI doesn’t require a pricey H100, an A100 (Ampere), or one of the top graphics cards. One Redditor showed how a Ryzen 5 4600G, which has a suggested retail price of $95, can handle various AI workloads.

The Ryzen 5 4600G, which debuted in 2020, is a hexa-core, 12-thread APU with Zen 2 cores that run at 3.7 GHz base frequency and 4.2 GHz boost frequency, respectively. A Radeon Vega iGPU with seven compute units running at up to 1.9 GHz is also equipped with the 65W chip. Keep in mind that APUs share system memory rather than having dedicated memory.

In the BIOS of the motherboard, you can find out how much memory is installed. In this instance, the Redditor gave the Ryzen 5 4600G 16GB of his 32GB of DDR4 memory. The most memory you can typically dedicate to the iGPU is 16GB. However, according to some user reviews, some ASRock AMD motherboards support higher memory allocation, possibly up to 64GB.

The hack turns the Ryzen 5 4600G into a 16GB “graphics card,” boasting more memory than some of the most recent GeForce RTX 40-series SKUs from Nvidia, like the GeForce RTX 4070 or GeForce RTX 4070 Ti, which are only capable of 12GB. Since 16GB is more than enough for light workloads, the APU logically cannot perform as well as a high-end graphics card, but at least it won’t run out of memory when handling AI workloads.

It is not officially supported by AMD’s Radeon Open Compute platform (ROCm) to use Ryzen APUs. Third-party businesses, like BruhnBruhn Holding, provide experimental ROCm packages compatible with APUs. This opens the door to the majority of AI software because APUs can work with the PyTorch and TensorFlow frameworks. We are curious as to how well AMD‘s newest mobile Ryzen chips perform, such as Phoenix, which uses DDR5 memory.

The Redditor shared a YouTube video that purported to show the Ryzen 5 4600G running a variety of AI programs, including Whisper, LLM, Alpaca-LoRA, FastChat, and LLaMA. Sadly, he only offered demos for Stable Diffusion, an artificial intelligence image generator that uses text input. He doesn’t explain how he made his Linux system’s Ryzen 5 4600G work with the AI software. The YouTuber promised to publish an in-depth video of the setup procedure.

Performance-wise, the Ryzen 5 4600G generated a 512 x 512-pixel image in just one minute and 50 seconds using the default setting of 50 steps. It is a fantastic outcome for a $95 APU and competes with some high-end processors. Despite not providing the specifications, the author claimed to have used DDR4 memory. Although many samples can reach DDR4-4000, the Ryzen 5 4600G only supports DDR4-3200 natively. It would be fascinating to see how AI performance scales with faster memory.

For those who have a Ryzen 5 4600G or Ryzen 5 5600G and want to experiment with AI, the experiment is fantastic. Investing $500 in an APU build doesn’t make much sense for those who don’t when you can probably get a discrete graphics card with better performance. Examples include AMD’s Radeon 16GB graphics cards, which start at $499, and the recently released GeForce RTX 4060 Ti 16GB from Nvidia, which has a comparable starting price.

Source(s)