GeForce RTX 40 series rumored to stick with a PCIe 4.0 interface

PCIe 5.0 made its debut with Intel's 12th gen CPU and motherboard range, and AMD is set to add support to its Zen 4 platform later this year. It was widely assumed that next gen graphics cards would support PCIe 5.0, but a new rumor suggests that Nvidia's next gen Ada GPUs will not support the new standard.

The rumor comes from the reliable kopite7kimi (via Videocardz). It's a bit of a surprise given that Nvidia's enterprise-oriented Hopper architecture does support PCIe 5.0. There's also the fact that RTX 40 will support 12+4 pin 12VHPWR PCIe 5.0 power connectors. But then, so does the RTX 3090 Ti and that certainly doesn't support PCIe 5.0.

If Nvidia does decide to stick with PCIe 4.0, it's not likely to have any impact. It's more of a box to check for marketing value. A current high-end GPU loses next to nothing when run on a 16x PCIe 3.0 system vs 4.0, so the difference will be even less compared to PCIe 4.0 vs 5.0. 

If Nvidia does omit PCIe 5.0 support, it might save a valuable few watts. It's believed that RTX 40 cards will consume unprecedented amounts of power, so any saving that can be made means an extra handful of watts can be budgeted elsewhere.

Of course, this is just a rumor. Nvidia is notoriously tight-lipped when it comes to details about its unreleased products. It's interesting to see the rumor mill ramping up though, indicating that the cards aren't too far from release.

It's believed that the cards have entered the testing phase, meaning the GPUs themselves have finished their design phase. Now it's up to finding a balance between things like yields and die harvesting, clock speeds, power requirements and PCB designs.

RTX 40 cards are expected to be built with TSMC's 4N node, which has been tweaked to suit Nvidia specifically. RTX 40 cards will feature a large increase in core count and cache size.

We can expect the RTX 40 series to be released in the second half of 2022. They will go head to head with AMD's MCM RDNA3 models around the same time.