DLSS 5 Cuts Frame Rates by Up to 75%: Why Nvidia’s New Tech Slows Down GPUs Instead of Speeding Them Up

Early independent tests of DLSS 5 in the basketball sim NBA 2K27 reveal a surprisingly heavy cost for enabling the neural-rendered image. At 1440p and 4K, activating Nvidia’s new tech drops real frame rates by 40% to 75%. Without multi-frame generation, even the near-flagship GeForce RTX 5080 falls from 185 fps to 53 fps at 4K, losing a stable 60 fps.

Not an Upscaler, but a Heavy Post-Processor

This sharp drop stems from how the system works: contrary to the established DLSS brand meaning, version 5 isn’t meant to boost performance. As Nvidia explained in its official announcement, the 3D-Guided Neural Rendering feature is a separate post-processing stage applied to an already rendered frame. The neural network takes geometry buffers and recalculates micro-shadows, lighting, skin texture, hair, and complex materials. Super Resolution upscaling and 6X Multi Frame Generation remain separate add-ons that run in parallel.

To measure the pure computational cost of the technology, TechSpot reviewers tested RTX 50-series cards with frame generation forcibly disabled. The load proved critical for most models:

• RTX 5090 at 4K lost about 60% of its original performance, dropping from 192 to 76 fps;
• RTX 5080 at 4K lost 71% of its base frame rate (falling from 185 to 53 fps), and at 1440p dropped from 221 to 102 fps;
• RTX 5070 at 4K plummeted from 124 to 33 fps;
• RTX 5060 Ti (16 GB) at 1440p managed only 50 fps versus 165 without the neural network’s intervention.

Independent measurements from German outlet Hardwareluxx confirmed the overall trend: on average across the test suite, performance losses were about 36% at 1080p, 54% at 1440p, and nearly 65% at 4K.

Why Resolution Doesn’t Save the Tensor Cores

The main engineering challenge lies in the pipeline architecture: DLSS 5 is applied to the final image. Switching upscaling profiles to Performance mode barely reduces resource usage, because the dense grid of enhancements must be calculated at the monitor’s native resolution. If the tensor core block lacks bandwidth, rendering hits a hardware bottleneck.

Nvidia’s official charts showing hundreds of frames per second come from running 6X Multi Frame Generation simultaneously. The generator literally masks the internal speed drop, filling the gap with bursts of synthetic intermediate frames.

Currently, DLSS 5 is exclusive to Blackwell-family accelerators (RTX 50). Company representatives say they are working on optimizing the algorithm, and they will consider releasing the technology on the previous RTX 40 generation only after debugging the current version.