NVIDIA RTX 5090 lost more than 25% performance without full PCIE bandwidth, with real losses in giving workloads

Embidia Blackwell GPU Nvidia apparently lost most of the performance if not operated under the complete PCIE bandwidth, especially with video editing applications.

Using other add-in cards apart from RTX 5090 NVIDIA can spend many differences in performance

Many factors determine the performance of GPUs in various workloads, and one important is PCIE bandwidth that can be accessed by it. With the latest generation of PCIe 5.0, inadequate PCIe paths can greatly damage the performance of GPUs, especially in all intensive workloads such as video rendering and content making. Pugen Systems have conducted extensive testing on Nvidia’s GeForce RTX 5090 to determine the impact of performance with lower PCIE bandwidth, and based on the results, it is proven that the difference is quite significant.

Interestingly, PCIE bandwidth allocated for GPUs is affected if other devices are put into PCIE slots separate from the main, because they basically distribute equivalent paths. Many motherboard manufacturers have a single 5.0 x 16 PCIE slot, which allows RTX 5090 to operate with full performance, but if other add-in cards, such as PCIE network cards, are connected, it does affect the performance of the GPU. Pugen Systems test the superior Nvidia Blackwell GPU in rendering AI applications and workloads, and it seems that the GPU runs perfectly only under the consumption of full PCIE bandwidth.

Starting with the after effect, RTX 5090 sees a significant performance hit when it comes down from PCIE 5.0 x 16 to PCIE 3.0 x 4, marking more than a 10%difference. Likewise, with Davinci Resolve, hit more than 20% when the GPU runs under PCIE 3.0 x 4, and there is a hit that is also seen when the number of paths drops from X16 to X4 in the same generation. This shows that operating several add-in cards on your motherboard can affect the performance of the GPU on a large scale.

In the game Dev Benchmarks, especially in Unreal Engine, almost all PCIE configurations show a slight change in performance. This also applies to AI workload, such as benchmark llammark.CPP, where performance is not affected when the lower PCIE bandwidth is assigned to the GPU onboard. The reason why the decline is not too significant here is that this application depends more on the VRAM GPU. While the average consumer should not be too worried about PCIE bandwidth, professional, especially content makers, must remember it.