Unreal Engine 5.4: Epic’s latest revision delivers big performance and feature wins

Unreal Engine 5 has reached a new milestone release, version 5.4. Alongside technology that sets the groundwork for future iterations of the engine, developer Epic Games is promising substantial performance improvements with better CPU utilisation, potentially solving a common bugbear we’ve seen on PC and consoles this generation. Of course, the new release also includes visual improvements, including to the Nanite geometry system and Temporal Super Resolution (TSR) upscaling. We’ve tested this latest engine revision to see how everything works and to get a sense of the kind of performance improvements that might be possible in the new version.

As stunning as it was, the first Matrix Awakens Unreal Engine 5.0 demo back in 2021 already exhibited the core performance issues that Epic is looking to address with 5.4. Moving through the Matrix demo city at speed on PlayStation 5 was enough to cause precipitous performance drops, as the engine grapples with loading in new areas. This is just a demo, of course, not a shipping game, but the sub-30fps readouts and frame-time stutters coupled with dynamic resolution scaling suggested a severe CPU limitation on console hardware. This was confirmed with the City Sample demo on PC in 2022, where simple CPU and GPU utilisation metrics show severe CPU bottlenecking even on powerful modern CPUs like the Ryzen 7 7800X3D.

Logically, this makes sense. A lot of the most prominent UE5 features have significant CPU components, like MetaHuman, open world streaming, hardware ray tracing and Nanite, and therefore place a heavy demand on CPU performance. More critically, you tend to be limited by single-thread performance, so higher core count CPUs aren’t able to meaningfully scale by spreading the load across multiple cores and threads. Given that both budget PCs and consoles tend to have core counts in the six to eight range and an increasing number of Unreal Engine 5 games are coming to market, that’s a lot of untapped performance potential.

Read more

Source

About Author