Impact of PCIe 5.0 Bandwidth on GPU Content Creation and LLM Performance (pugetsystems.com)

Fascinating and (to me) unintuitive result that these cards don't seem bandwidth constrained at PCI 4x16; the update to PCI 5x16 doesn't seem like it has any measurable impact on performance, allowing a PCI 5x8 configuration that "saves" some of your lanes at seemingly no penalty.

I wonder if we will have to wait another generation of cards (and apps) to make full use of a PCI 5x16 connection?

i can not see graphs

reason: "DataTables warning: table id=table_5 - Ajax error. For more information about this error, please see http://datatables.net/tn/7"

Really too bad that they ran only one test, and it was a small model.

Testing PCIe bandwidth with a model that fits entirely into VRAM (quantized Phi-3 Mini at 2.39 GB on RTX 5090 with 32GB VRAM) is stupid because there won't be any memory transfer over PCIe beyond the initial model load. They should have tested a large MoE model like Qwen3-235B-A22B-GGUF, where the difference will be huge.

[dead]