As I assumed, there was something off with the NVIDIA DLSS 3 implementation in Forza Horizon 5. It turns out that a driver-level bug with the MFAA setting from NVIDIA's Control Panel caused the issues described in the original story when DLSS Frame Generation is enabled in the game.
NVIDIA was able to reproduce the problems and notified me of the bug. As soon as I turned off MFAA, DLSS 3 once again behaved as expected, and its fluidity returned to normal.
I took the opportunity to run the benchmark tool a few more times with different settings. To begin with, just turning off MFAA yielded a 4 FPS increase from the previous test.
My next step was testing DLSS Frame Generation on its own without DLSS Super Resolution. The results were surprising, as the benchmark ran only 9 FPS slower, proving that the former technology is much more impactful in Forza Horizon 5. This was also the smoothest test run in terms of stutters counted by the benchmark (zero).
Given the performance headroom available, I tried enabling DLAA to improve image quality. Once again, the benchmark result only lost two frames, although three stutters were counted by the built-in tool.
For the last test, I used DLDSR (Deep Learning Downsampling Resolution) to clean up the visuals further. Using the highest scaling factor of 2.25x, the rendering resolution jumped to 5760x3240, which in turn dropped the average FPS to 98 even with DLSS Frame Generation enabled.
Ultimately, I feel that the best balance when using NVIDIA DLSS 3 in Forza Horizon 5 is achieved with DLAA enabled, as DLDSR doesn't provide enough image quality benefits over DLAA to make the FPS loss worth it. The DLDSR benchmark also shows a sizable increase in latency over the previous results.
Unfortunat
Read more on wccftech.com