Unified Memory Architecture#
NVIDIA RTX Spark uses a unified memory architecture (UMA). The CPU and integrated GPU (iGPU) share the same physical memory. The platform maintains coherence for memory that is accessible to more than one processor. Memory capacity, interface width, and bandwidth are listed in System Overview.
NVIDIA RTX Spark exposes the physical memory pool as three logical regions:
Dedicated GPU memory (carveout): This region is reserved for the iGPU and reported by Windows as dedicated GPU memory. The carveout is physically backed by system DRAM rather than separate VRAM, but it serves the same role as VRAM on a discrete GPU.
Shared system memory: This region is managed by Windows and accessible to the CPU and iGPU.
CPU-only system memory: The remaining system memory is accessible only to the CPU and cannot be used for GPU allocations.
After the dedicated carveout is reserved, the remaining memory is divided between shared system memory and CPU-only system memory. The nominal shared system memory size is the post-carveout capacity minus 16 GB. That value is clamped to a minimum of 50% and a maximum of 80% of the post-carveout capacity. CPU-only system memory receives the balance.