Use your Nvidia GPU's VRAM as swap space on Linux
github.com
June 2, 2026
5 min read
🔥🔥🔥🔥🔥
67/100
Summary
c0deJedi's nbd-vram enables the use of NVIDIA GPU VRAM as swap space on Linux, particularly for laptops with soldered memory. This solution allows high-priority swap allocation, significantly increasing total addressable memory by utilizing unused VRAM.
Key Takeaways
- The nbd-vram project allows users to utilize NVIDIA GPU VRAM as swap space on Linux, particularly for laptops with soldered memory and no upgrade path.
- The implementation uses a small daemon that allocates VRAM via the CUDA driver API and exposes it as a block device using the Network Block Device (NBD) protocol.
- The system can achieve a total addressable memory of approximately 46 GB by combining RAM, VRAM, zram, and SSD swap.
- The service automatically starts on boot and can be configured to manage power-aware settings based on AC power status.
Community Sentiment
Positives
- Utilizing VRAM as swap space can optimize resource usage, particularly for systems with limited RAM, enhancing performance for specific applications.
- The concept of GPU-accelerated databases is gaining traction, with major companies investing in this technology, indicating a strong future for data processing capabilities.
- This approach could allow users with high-end GPUs to leverage their idle VRAM for other tasks, potentially increasing overall system efficiency.
Concerns
- The reported sequential throughput of ~1.3 GB/s on an RTX 3070 is significantly underwhelming compared to its potential, raising concerns about performance limitations.
- Swapping to NVMe drives may be faster but introduces higher latency, which could negate the benefits of using VRAM as swap space.
- There are worries about the implementation of this system, as it may not be optimized for high performance, leading to inefficiencies.
