NVIDIA has announced significant advancements in its Deep Learning Super Sampling (DLSS) technology with the release of DLSS 5. The latest optimization update, scheduled for October 16, 2026, promises substantial improvements for users.
Performance and Memory Efficiency Improvements
According to the company, the new iteration of the technology will deliver an additional twofold boost in performance. This means that workloads previously optimized using DLSS 5 will operate at twice the speed. Beyond the performance gain, NVIDIA also guarantees that when using the neural rendering technology for 4K gameplay, the utilization of VRAM will decrease by 20%. This reduction is intended to free up memory capacity, allowing game textures to load without DLSS 5 competing for or depleting system resources.
NVIDIA continues to refine DLSS by utilizing a perpetual optimization process running on an internal supercomputer. For this specific update, the company stated that achieving the 2x performance increase required combining systematic pipeline optimizations with sophisticated refinements to the neural model.
Technical Model Optimization
The continuous refinement of the DLSS model has resulted in a considerably smaller footprint. Previously, leaked details concerning the DLSS 5 file indicated a model size of approximately 148 million parameters, an estimate that has proven accurate. NVIDIA stated that the October update reduces this model size by an additional 24%, bringing the parameter count down to roughly 118 million parameters. This smaller model size allows for a near-linear reduction in memory requirements, resulting in the reported 20% cut in VRAM usage.
The rapid rate of improvement is notable. Comparing this release to the initial showcase of DLSS 5 in mid-March, when it was presented on two GeForce RTX 5090 GPUs, NVIDIA has managed to scale the system’s performance by an impressive factor of 10 within just a couple of months.
