Although we approached this review with little hopes of a revelation, we must admit we're a bit surprised at just how little the situation has changed with high-end graphics cards saturating the PCI-Express bus. The GeForce GTX 1080 is about 30% faster than the Radeon R9 Fury X from last year's PCI-Express scaling review. We were expecting whatever little performance drops we saw last year as we lowered PCI-Express bandwidth to become more pronounced since the GTX 1080 is a faster card being tested on newer games. This, however, doesn't seem to be the case.
When averaged across all our games, you lose virtually no performance as you go down from PCI-Express 3.0 x16 to PCI-Express 3.0 x8, no matter the resolution. Performance doesn't even drop with newer DirectX 12 and Vulkan games, including titles like "DOOM," which are known to utilize virtual texturing ("mega textures," an API feature analogous to Direct3D tiled-resources). If anything, mega textures has reduced the GPU's bandwidth load on the PCI-Express bus. There is, similarly, no noticeable performance loss (1-2%) between PCI-Express 3.0 x16 and PCI-Express 2.0 x16. This should come as a relief to both those gaming on older platforms such as Intel "Sandy Bridge" and AMD FX and those considering the expensive Intel HEDT platform just for its PCI-Express 3.0 x16 assurance when installing two cards in SLI. There are certain tests that load the PCI-Express bus more than others. "Far Cry: Primal" posts steeper performance losses as we switch between gen 3.0 x16 and gen 2.0 x8 at 1080p resolution.
Performance losses begin to be noticeable as you get down to PCI-Express 2.0 x8, PCI-Express 3.0 x4, and below. Even here, the frame-rate drops are within 5-10% of PCI-Express 3.0 x16. If that makes a difference between "playable" and "slideshow" for you, you have something to consider. PCI-Express 1.1 x16 still has sufficient bandwidth with performance similar to PCI-Express 2.0 x8. As you switch to gen 1.1 x8 and gen 1.1 x4, the performance loss begins to become more noticeable. Even in the slowest PCI-Express mode, the GTX 1080 isn't much slower than a GTX 1070 running at Gen 3.0 x16.
An interesting trend we noticed is that frame-rate losses are more pronounced at the lower 1920 x 1080 resolution rather than the higher Ultra HD resolution. This is due to the higher frame rate at the lower resolution, which requires more PCIe bandwidth. The frame rates in such cases, with the GTX 1080, are still too high for you to worry about. Perhaps it makes a difference for some if they're gaming on fast 144 Hz monitors.
We expected the chipset-linked PCI-Express 3.0 x4 (physical x16) slot to be the weakest option for you since this setup is sub-optimal and feeds on your chipset's bus bandwidth. The performance loss for this option is there with 5-8%, which isn't that significant. We'd still not recommend using such a slot since it slows down other devices in your system, such as your SSD, and conversely, other bandwidth-hungry devices can slow down your graphics card, causing frame-rate drops. It's also important to mention that installing graphics cards in such slots could cause the graphics card to heat up by limiting airflow due to its position at the very bottom.
Just out of curiosity, we investigated the impact of switching between the three PCI-Express generations on power draw. We are happy to report that switching to older PCI-Express generations made no difference to power draw. The measurements only differ in percentage, and to an expected extent due to slightly different frame rates which have the GPU stay idle a bit longer.
We hope our data helps settle a lot of flame wars on the forums. PCI-Express 3.0 x8 is just fine for this generation of GPUs, and so is PCI-Express 2.0 x16. You lose about 4% in performance at PCI-Express 2.0 x8 and PCI-Express 3.0 x4, but that's no deal breaker. If you're still gaming on a PCI-Express 1.1 platform, congratulations, your future-proofing has worked for many years, but it may now be time to upgrade.