Enthusiast repurposes Lenovo Yoga laptop and AMD Radeon RX 7900 XT into M.2-powered desktop for local AI chatbots but hits DRAM bottleneck

An enthusiast attempted to repurpose a Lenovo Yoga laptop into a desktop-like system for running local large language models (LLMs) by connecting an AMD Radeon RX 7900 XT via the laptop’s M.2 slot. The GPU was chosen for its 20 GB of VRAM, which is useful for LLM workloads, but the laptop’s limited DRAM became a critical bottleneck. The setup required booting from a USB SSD and initially worked for basic tasks, but the integrated display’s framebuffer consumed VRAM, reducing available memory for the models.

The project’s core issue emerged when running larger LLMs like Qwen 35B A3B and GLM-4.7 Flash with a 128K-token context window. The laptop’s DRAM was insufficient to handle the memory demands of the models, leading to excessive swapping and near-idle GPU performance. This mirrors a classic trade-off in AI workloads: balancing GPU memory, system RAM, and context window size. The experiment demonstrated that even high-VRAM GPUs cannot compensate for inadequate system memory in memory-intensive tasks.

The failure of the laptop-based setup underscores the importance of hardware balance for local AI workloads. While the M.2 slot provided a creative workaround for GPU connectivity, the laptop’s DRAM limitations made it unsuitable for sustained LLM operation. The project’s conclusion, migrating the GPU to a desktop system, highlights that consumer laptops, even when modified, are not ideal for high-memory AI tasks. Engineers evaluating similar setups should prioritize systems with sufficient RAM to avoid performance degradation.