DLSS 5 Test on Apple M5 Pro Shows Huge Latency Penalty
An unofficial port of NVIDIA's DLSS 5 to Apple's M5 Pro chip delivers dramatically improved image quality but causes framerates to plummet to 2.32 FPS and
NVIDIA's DLSS 5 has been tested unofficially on Apple's M5 Pro chip, revealing massive image quality gains paired with crippling performance trade-offs. The experimental test saw framerates drop to single digits and input latency soar to ten times that of a desktop RTX 5050 GPU.
Developers @iamwavecut and @Mappsnet7 ported DLSS 5's neural rendering models to run natively on Apple Silicon using the Metal framework. They then hooked this Metal backend into Windows games using ReShade and Game Porting Toolkit 4.0b2. On the M5 Pro's 16-core GPU, this setup resulted in enhanced surface lighting, better ambient occlusion, and improved depth realism.
The visual improvements, however, came at an extreme cost. Performance on the M5 Pro collapsed, with framerates falling to just 2.32 FPS at a 1440p resolution. More critically, input latency increased to 240 milliseconds.
Technical Cause of the Performance Loss
Redditor "AnyPomelo3352" explained that the severe performance drop is due to the Apple M5 Pro's architectural design. The chip lacks dedicated FP8 (8-bit floating point) hardware units, which DLSS 5 relies on for efficient neural rendering. Without these specialized units, the M5 Pro is forced to route the neural workload through its FP16 (16-bit floating point) path, resulting in no performance benefit from the upscaling technology.
A comparison of theoretical compute power highlights the gap between the Apple silicon and a dedicated NVIDIA gaming GPU.
| Component | FP8 Compute Performance (TOPS) |
|---|---|
| Apple M5 Pro (16-core GPU) | 26 TOPS |
| NVIDIA RTX 5050 Desktop GPU | ~105 TOPS |
The RTX 5050 delivers roughly four times the FP8 performance of the M5 Pro. The report also notes that running the neural inference through the ReShade and Game Porting Toolkit layers adds its own significant performance overhead.
Implications for Apple Silicon Gaming
The test shows a fundamental challenge for gaming on Apple's in-house processors. While Apple has made efforts to boost gaming adoption for its silicon, the industry's future is increasingly tied to neural rendering techniques like DLSS. The report states that DLSS 5 has "pretty much established" that the M5 Pro and M5 Max are "several generations behind" in this specific regard.
The question of whether Apple's next-generation M6 chip will support FP8 hardware remains open. For now, the experimental test demonstrates that simply porting advanced upscaling software is insufficient without the underlying dedicated hardware to run it efficiently. The result is stunning image quality paired with practically unusable performance.
