Introduction
With the release of the NVIDIA Blackwell GPUs and RDNA 4-based Radeon 9000-series GPUs, we finally have consumer video cards that support the PCIe 5.0 standard. Although we have had motherboards with support for it for some time now, we didn’t have any devices other than storage that took advantage of it. This leads now to the question: What impact does PCIe 5.0’s increased bandwidth have on GPU performance in content creation applications?
PCI Express (abbreviated PCIe or PCI-e) is a technology used to connect various internal computer devices to the motherboard. The physical connectors and communication schema are used for drives, GPUs, and add-in cards like RAID or HBA cards and network cards. Since 2003, we have seen a variety of revisions and updates to the standard. Currently, the most common PCIe specification seen on new high-end motherboards is PCIe 5.0 at 16x, though often with some 4.0 lanes available.
The primary difference between PCI Express versions is transfer rate. A PCIe connection between devices has two defining features: the number of lanes and the PCIe version. Most slots on the motherboard have between four and sixteen lanes (x4, x8, or x16), with the occasional x1 or x2 slot. Each of these lanes has a maximum transfer rate, defined by the PCIe version. Since PCIe 3.0, each new version has doubled this transfer rate.
As an example, PCIe 5.0 supports up to 32 GT/s per lane. So, an x16 slot has 16 lanes each at 32 GT/s for a maximum throughput of 64 GB/s. If that same slot were using the ePCIe 4.0 protocol, it would have 16 lanes at 16 GT/s for a throughput of up to 32 GB/s. Alternatively, you could achieve 32 GB/s with x8 lanes at PCIe 5.0.
At present, consumer desktop motherboards tend to feature limited “free” PCIe lanes. We tend to be disappointed with the quantity, connectivity, and placement of PCIe slots on motherboards. Although it depends on the specific price point being targeted, many boards will have a primary 5.0 x16 slot, and then only a few other slots, typically at 4.0 x4 or even 3.0 x1. One reason we like the ASUS ProArt boards we often carry is in part due to the PCIe slot layout and support.
However, this isn’t merely motherboard vendors being cheap. Instead of maximizing add-in card support, they are typically dedicating many of the available PCIe lanes (from the CPU or chipset) to additional features like M.2 slots, USB ports, and Ethernet/WiFi. The drawback to this is that once a GPU is installed, there may be no way to add a second at full bandwidth (if at all). Even our preferred boards require that the GPU be run at x8 if we want to install most add-in cards or additional GPUs.
Given that add-in cards (GPUs or otherwise) may need to run at lower-than-maximum bandwidth due to PCIe lane availability concerns, it is reasonable to ask what the cost is. How much performance is lost when video cards are operated at less than their maximum PCIe bandwidth?
An Illustration of the PCIe Slot Problem
MSI MAG X870 Tomahawk Gigabyte X870 Aorus Elite ASUS ROG Strix X870E-E MSI B650 Gaming Plus
... continue reading