NVIDIA Reportedly Developing Technology to Use NVMe SSDs as Extra VRAM for GeForce Cards

NVIDIA is developing a new technology that would allow future GeForce graphics cards to use high-speed NVMe SSDs as additional video memory. When a card's on-board VRAM is insufficient, the GPU can directly access data stored on the SSD. The feature is based on NVIDIA's RTX IO technology and Microsoft's DirectStorage API. RTX IO provides a direct data channel from the NVMe SSD to the GPU, allowing the GPU to bypass the CPU and system memory, read compressed data directly from the SSD, and decompress it on the GPU.

The technology aims to reduce the burden on the CPU. Traditional storage access paths require CPU intervention, whereas RTX IO and DirectStorage let the GPU handle storage I/O directly, thereby lowering CPU load. The technology also accelerates the loading of game assets, enabling faster entry into games and reducing stuttering. When a GPU's on-board VRAM is exhausted, the new technology allows the GPU to fall back on slower but higher-capacity NVMe SSDs as back-up video memory, making it suitable for games or AI applications that require large amounts of VRAM.

SSD speeds remain far lower than those of physical VRAM. Related project data shows that theoretical bandwidth over a PCIe 4.0 x16 connection is roughly 32 GB per second, far below that of GDDR memory. The technology has not yet been officially confirmed, and its implementation will depend on the maturity of the underlying ecosystem, including RTX IO and DirectStorage.