Introduction

NVIDIA earlier this week launched the GeForce RTX 4090 "Ada," its flagship next-generation graphics card. We have a boatload of reviews of these cards for you (9, counting the Founders Edition). As is customary with each new GPU generation, we do several feature articles closely related to the architecture, and one of them has to be PCI-Express scaling—to test by just how much the flagship graphics card's performance relies on the available bus bandwidth, and how much its performance is affected on machines with older PCIe standards or narrower PCIe bus interfaces. This should prove particularly important for those building 13th Gen Intel Core machines who plan to use PCIe Gen 5 NVMe SSDs.



The NVIDIA GeForce RTX 4090 "Ada" graphics card features a PCI-Express 4.0 x16 host interface, even though its power architecture meets PCIe Gen 5 standards. Gen 4 x16 gives it 32 GB/s per-direction bandwidth. The card also supports the PCI Resizable BAR feature, which lets the CPU see all 24 GB of its video memory as a single addressable block, instead of funneling everything through a tiny 256 MB aperture window. NVIDIA's decision to go with PCIe Gen 4 host-interface is a bit curious, given that the CPU platforms have moved on to Gen 5 from both Intel and AMD. Intel has been supporting PCIe Gen 5 since 2021, with its 12th Gen Core "Alder Lake" processors; while AMD recently launched its Ryzen 7000 "Zen 4" processors with PCIe Gen 5 support. There's something interesting about these two platforms, and it concerns the PCIe Gen 5 NVMe SSDs that are about to debut this November.

While the Ryzen 7000 "Zen 4" processors put out PCIe Gen 5 both for the x16 PEG and CPU-attached x4 NVMe interfaces. On the Intel side, both the 12th Gen and upcoming 13th Gen Core "Raptor Lake" put out PCIe Gen 5 only for the x16 PEG slot, the CPU-attached NVMe slots are still Gen 4. In order to not lose this platform advantage to AMD, Intel is now allowing motherboard designers to add Gen 5 M.2 NVMe slots on their upcoming 700-series motherboards, but they're going about this by stealing PCIe lanes from the x16 PEG slot. This means that when 13th Gen Core users install an SSD in that M.2 slot, it will redirect eight lanes that were used by the x16 graphics slot, and use those on the M.2 slot. So those building a bleeding-edge Core i9-13900K + RTX 4090 gaming PC with a Gen 5 NVMe SSD will have to content with the RTX 4090 running at PCI-Express x8 bandwidth. No such problem on the Ryzen 7000 platform, as the CPU-attached NVMe slots are Gen 5, both on the X670/B650 "E" and non-E chipsets; without eating into the PCI-Express Graphics link configuration in any way.

This is where it becomes important to see if the GeForce RTX 4090 suffers any performance loss with PCI-Express 4.0 x8. While unlikely, some RTX 4090 buyers may still be on PCIe Gen 3 platforms, such as 10th Gen Core "Comet Lake" or Ryzen 2000 series. PCIe Gen 4.0 can be had with 11th Gen Intel Core onward, and Ryzen 3000 "Zen 2" onward. There are some exceptions such as the Ryzen 5000G APUs that only offer PCIe Gen 3. The Gen 2 x16 and Gen 1.1 x16 numbers should be highly relevant to those wanting to use external GPU boxes with their RTX 4090, as the 40 Gbps or 80 Gbps Thunderbolt/USB4 connections powering these boxes typically end up offering this kind of bandwidth to the graphics card inside.

In this review, we are comparing the RTX 4090 in its native PCI-Express 4.0 x16 setting, with PCI-Express 3.0 x16 (identical bandwidth to PCI-Express 4.0 x8), and older generations of PCIe from an academic curiosity. We are forcing older PCIe generations using settings in the motherboard's UEFI setup program.