Vulkan Memory Management
summary
Vulkan pushes memory decisions to the app, so use 256 MB blocks for suballocation (drivers often cap allocations around 4,000), because individual allocations are slow and expensive. Use the transfer queue for async texture streaming and background defragmentation, but avoid GPU-to-GPU copies on that queue. Prefer optimal tiled images, avoid setting every usage bit, and keep about 20–33% free of heaps. Aliasing render targets can save up to 300 MB. The open‑source Vulkan Memory Allocator handles type selection, suballocations and dumps JSON for analysis.
AI-written summary · about