DLSS 5 запущен на Apple M5 Pro – результат в 10 раз медленнее RTX 5050
Enthusiasts have managed to run neural rendering NVIDIA DLSS 5 on computers with Apple Silicon. The experimental implementation does alter the image, but on the M5 Pro at 1440p, the speed dropped to 2.32 FPS, and processing a single frame took about 240 ms.
How DLSS 5 was launched on Mac
Developer iamwavecut ported DLSS 5 neural network models to Apple Silicon using MLX and Metal. The project they created, MLX-DLSS, can run neural rendering and frame generation without an NVIDIA graphics card. However, the project does not distribute ready-made models or NVIDIA libraries: weights are extracted from the user’s own DLSS libraries.
Another developer, Mappsnet7, connected a Metal backend to Windows games via ReShade and Game Porting Toolkit 4.0b2. This means it is not official DLSS 5 support on macOS, but rather an experimental chain of several software layers.
Reddit user AnyPomelo3352 tested it on a MacBook Pro with an M5 Pro chip and a 16-core GPU. The game launched via Game Porting Toolkit 4.0b2 and Metal 4 at a resolution of 2560×1600 pixels. After enabling neural rendering, changes in surface lighting, shading, and scene depth became more noticeable.
The image is better, but it’s unplayable
The main issue turned out to be performance. At 1440p, the latency of neural rendering reached approximately 240 ms, and the frame rate dropped to 2.32 FPS. For comparison, TechSpot estimated DLSS 5 processing time on a desktop GeForce RTX 5050 at 1440p to be around 24 ms. This results in a tenfold difference just during the DLSS 5 stage.
One reason is tied to FP8 computations, which DLSS 5 relies on. The M5 Pro lacks a dedicated hardware path for FP8, so the experimental implementation performs these operations via FP16. According to the tester’s estimate, the 16-core GPU M5 Pro delivers around 26 TOPS in this scenario, compared to approximately 105 TOPS for the RTX 5050. Additional performance losses come from ReShade and graphics conversion through Game Porting Toolkit.
On the more powerful M5 Max, results are better, but still far from comfortable gaming. In an early test of The Witcher 3 at maximum settings via GPTK4, the system delivered approximately 10 FPS FPS at 1728×1080, 15 FPS FPS at 1280×800, and 22 FPS FPS at 960×600. The developer noted that optimization of Metal cores and frame transfer between the game and the neural network is still ongoing.
DLSS 5 differs from previous DLSS generations by not merely reconstructing an image from a lower-resolution frame. The neural network generates the final frame appearance using data from standard rendering, motion vectors, and temporal information. NVIDIA’s official implementation currently operates as part of the gaming pipeline on GeForce RTX 50 graphics cards.
Thus, the experiment confirms that DLSS 5 models can technically be ported beyond the NVIDIA ecosystem. However, the current results on M5 Pro and M5 Max remain more of a demonstration of launch capability than a practical way to use DLSS 5 in games on Mac.
