With the surge of graphic-intensive apps and games, one of the components that greatly benefit from PCIe’s high transfer rate, high bandwidth, low latency, and efficient handling of large chunks of data is the GPU. There are a lot of grounds to cover to understand the technology behind PCIe fully, but at its most basic, it consists of PCIe lanes that link one component to another. Let’s discuss the dynamics of PCIe lanes and how they relate to a GPU’s performance.
PCIe lanes serve as the highway of data traveling to and from high-speed components like GPUs and SSDs and the processor or chipset. Each lane consists of two pairs of copper wires; one pair is for sending out data, and the other is for receiving data. The more lanes there are, the higher the aggregate bandwidth that can be delivered. PCIe 5.0, the latest generation of PCIe interface, has an aggregate bandwidth of 128Gbps on a x16 implementation, twice that of its predecessor, PCIe 4.0.
When you hear about PCIe lanes, the lanes on the PCIe slots instantly come to mind, but PCIe lanes are more than just the slots. PCIe lanes also operate USB ports, SATA ports, M.2 ports, and some I/O ports. PCIe lanes link these components to the motherboard’s core components like the CPU and RAM. Do you know that the CPU and the motherboard’s chipset each have their own set of PCIe lanes?
The total number of PCIe lanes available on a chipset depends on the motherboard’s architecture, but it normally ranges from 4 to 24. You can check the total available PCIe lanes on your motherboard by checking out the manufacturer’s specifications. Chipset lanes are primarily allocated for USB ports, SATA ports, M.2 slots, onboard Ethernet, and some PCIe expansion slots. Chipset lanes can be shared among the components if the total number of lanes is limited. For example, SATA ports may share the lanes with M.2 slots, but both cannot use the lanes at the same time. One component will be inoperable, while another component will be using the PCIe lanes.
Because of this limitation, chipset lanes do not fully handle a GPU’s high-bandwidth demand. To take some load off the chipset and thus avoid bottlenecks, CPUs have their own PCIe lanes linked directly to the PCIe expansion slots.
The PCIe lanes on the processor are mainly tied to the primary and secondary expansion slots. These slots are normally the closest to the CPU. The primary slot is usually the PCIe x16 expansion slot, while the secondary slot is the x8 slot, but it could vary according to the manufacturer’s design. Because of their high bandwidth demand, GPUs normally occupy these two slots. Depending on the manufacturer, the CPU also provides lanes linked to other components. For example, Intel’s processors have PCIe lanes linked directly to the PCIe x16 expansion slot, while AMD’s have PCIe lanes reserved for other components like SATA ports and M.2 ports aside from the expansion slot. The CPU lanes are not shared between devices, so we can expect optimum performance from the device connected to the PCIe slot directly connected to the CPU.
PCIe Lanes and GPUs
Determining the number of lanes your chipset and CPU supports is important, especially if you’re a PC builder. This way, you would not only know whether the PCIe interface will be saturated, but you would also future-proof your system by knowing which devices would work best with it now and in the future.
GPUs usually require the highest bandwidth and transmission rate among the expansion cards considering that they process large amounts of graphical data. Games, cinematic applications, machine learning, and artificial intelligence are just some of the applications that rely on graphics. For this reason, the top PCIe x16 slot, which has the highest number of lanes and thus can provide the highest bandwidth, is typically reserved for GPUs. Depending on your workload, a PCIe x8 slot can also provide the bandwidth that GPUs require. You can still use a PCIe x4 slot, but you can expect a drop in performance since it has fewer lanes.
PCIe lanes, however, are not the only driver of bandwidth; the PCIe version also plays an important role. Newer versions of PCIe have double the speed and bandwidth of their predecessor. A PCIe 5.0, for example, has double the bandwidth of a PCIe 4.0 on a x16 slot and on any other slot, for that matter. Therefore, a PCIe 5.0 GPU can have the same bandwidth on a PCIe 5.0 x8 slot and on a PCIe 4.0 x16 slot.
Most computer systems can handle multiple GPUs to handle multiple displays. Likewise, standards such as SLI and Crossfire merge multiple GPUs and make them work as one. No matter the case, if two GPUs occupy both the x16 and x8 slots, the top slot’s lanes will be reduced according to the PCIe lanes made available by the CPU. For example, if the CPU provides 16 PCIe lanes to the expansion slots, and two GPUs use both the x16 and x8 slots, the first slot will be reduced to eight lanes since the CPU has a max number of 16 lanes linked to the expansion slots. In most cases, if the processor is on PCIe 4.0, there will be no significant difference in the performance of your GPU also has the same PCIe version as the processor. If either the processor or the GPU has a lower PCIe version than the other, there will be some impact on the performance since the capabilities will be downgraded to that of the lower version.
To sum up, the number of lanes a GPU will use depends on the workload, the number of PCIe lanes from the CPU, and the PCIe version used. Sixteen lanes are ideal for a single GPU, but generally, a x8 slot is already sufficient for regular consumer workload and even for graphic-intensive games and apps. For multi-GPU configuration, it is recommended to use CPUs with a higher number of lanes to avoid saturating the PCIe lanes. PCIe 4.0 is recommended for modern games and graphic-intensive apps, but even PCIe 3.0 can still handle data traffic just fine. Lower versions may no longer be able to handle today’s demand for graphical rendering. For excellent performance, use the PCIe x16 or PCIe x8 slot for your GPU while considering the PCIe version used. It is also best to consult the motherboard’s PCIe specifications to know the max number of lanes the GPU can use for either single or multiple GPU configurations.