VRAM Allocation Strategies and DirectStorage GPU Asset Decompression: Fundamental Mechanics and Deep Mathematical Modeling

[Mason Collins]

I’ve been digging into how DirectStorage actually moves data during live gameplay, and I want to hear from people who’ve benchmarked this on real hardware rather than just reading whitepapers. When GDeflate decompression runs on the GPU’s compute units, how are you all handling the tradeoff between decompression throughput and the VRAM you have left for textures? On my 4090 with 24GB, once I’m pushing 4K ultra assets, the moment decompression spikes seems to cause visible hitches in competitive shooters where I need consistent frame times, not just peak FPS.

For those tuning PCIe Gen 4 vs Gen 5 NVMe setups, have you measured where the throughput ceiling actually bites during sustained asset streaming? I’m curious about the math behind how decompression parallelism scales across compute cores versus the lane bottleneck—is there a point where adding lanes stops helping because the shader pipes become the limiter? And does anyone have hard data on whether it’s better to reserve VRAM for the decompression scratch space or let the OS page it dynamically? Would really value real-world numbers over theory.

🎙️ Shure SM7B + Cloudlifter CL-1 + Elgato Cam Link 4K (Sony ZV-E10)

[Author]

Posts