TECH NEWS – To achieve maximum performance, Nvidia’s Blackwell-architecture GPU requires full PCIe bandwidth.

Several factors determine a GPU’s performance under various workloads, and one of the most important is the available PCIe bandwidth. With PCIe 5.0, insufficient lane allocation can significantly degrade GPU performance, especially during intense workloads like video rendering and content creation. Puget Systems conducted extensive tests on the RTX 5090 to examine how performance is affected by reduced PCIe bandwidth. The results showed a substantial difference.

If another device is installed in the primary PCIe slot besides the GPU, the lanes are essentially divided evenly, affecting the GPU’s bandwidth. Many motherboards offer a single PCIe 5.0 x16 slot, which allows the RTX 5090 to operate at full power. However, if additional cards like PCIe sound or network cards are inserted, GPU performance may suffer. Puget Systems tested the GPU in rendering and AI workloads and found that the card only achieved optimal performance when full PCIe bandwidth was available.

In After Effects, the RTX 5090 showed a significant performance drop when downgraded from PCIe 5.0 x16 to PCIe 3.0 x4, with more than a 10% loss. In DaVinci Resolve, performance fell by over 20% under PCIe 3.0 x4. Even within the same generation, cutting lane count from x16 to x4 noticeably impacted performance. Running multiple expansion cards on the motherboard can drastically affect GPU performance.

In game development benchmarks, especially with Unreal Engine, there was little change in performance across different PCIe configurations. The same applied to AI workloads such as the Llama.cpp benchmark, where GPU performance remained stable even with limited PCIe bandwidth. This is likely because these applications rely more heavily on GPU VRAM.

While everyday consumers don’t need to worry too much about PCIe bandwidth, professionals—especially content creators—should pay close attention to it.

Source: WCCFTech, Puget Systems