Explaining PCIe lanes helps clarify if running two GPUs with 16 free lanes creates a limitation.
Explaining PCIe lanes helps clarify if running two GPUs with 16 free lanes creates a limitation.
I'm preparing my initial self-built PC and figuring out the best motherboard. I'm considering an AMD Ryzen 7 5800X but unsure which RAM size is right. I stumbled upon information about ENTHUSIAST PCIe lanes. My goal is to add two graphics cards plus a video capture card. The CPU only supports 24 lanes, with 4 already used by the motherboard and another 4 for an M.2 drive. That leaves 16 lanes for the GPUs. Would this cause a bottleneck? Also, if I need a second M.2 drive, does this relate to AMD's Smart Access Memory Technology and affect performance?
What gpu sare you using? For gaming, there is almost no reason to get 2 gpus now. Normally 2 drives at x8 will have basically no performance hit for most uses. The chipset has more pcie lanes for things like a capture card and other m.2 drives.
They don't require 16 lanes, but they'll utilize 16. AM4 PCIe connections have a simplified setup: each Ryzen CPU provides 24 lanes, with 16 going to the main PCIe slot, 4 to an M.2 card, and 4 to the chipset. The chipset then rearranges those lanes into additional pathways—either 8x/4x or 8x/8x depending on the board. In some configurations you can even achieve 8x/4x/4x. This doesn't increase speed; it just allows more peripherals to connect. Let's return to the 16 lanes in the main slot, which can be divided further—either 8x/4x for budget boards or 8x/8x for premium models. Occasionally you'll see 8x/4x/4x, though that's unusual. For 3-way SLI, you need a platform with enough lanes for 8x/8x/8x, since SLI demands eight lanes per GPU. The 2 GT 710s won't be limited by 3.0 8x—they're unlikely to be affected by 3.0 1x. However, 2 3090s might face some limitation with 3.0 8x, as that would cut their max transfer rate in half (3090s reach up to 4.0 16x; 3.0 16x = 4.0 8x). Ultimately, the lanes pass through the chipset, not the CPU.
Your setup works well for both gaming and VMs. The 6800 can handle gaming, while the GTX 1060 stays free for virtual machines. The issue is about M.2 slots—most modern motherboards support dual-channel configurations, so they can share the lanes efficiently. Each slot typically needs two lanes, but with proper configuration, both can run smoothly together.
Choose x8 x8 for both GPUs or x16 only in the top slot based on distance from the CPU. Some boards allow a second PCIe x16 slot which might connect to the chipset, possibly causing a bottleneck—this depends on whether the virtual machine prioritizes GPU speed. The second M.2 slot is usually linked to the chipset and could hit bandwidth limits, though it won’t impact the first M.2 slot.
It seems to be a 6800XT paired with an older GTX 1060 setup. Would that create a performance issue with 4.0 16x lanes, or could you consider upgrading later?
The virtual machines in use are compatible with shared GPUs or virtual GPU configurations, eliminating the need for an additional GPU. This arrangement should function properly, and running at 8x will not pose any problems.
Consider setting up a QEMU VM on Linux with GPU acceleration while running macOS. You may need to purchase a new GPU since NVIDIA isn't supported in macOS, or use the Intel 1060 for Linux. If you plan to add another card, like a capture card, think about how it will fit within your system and whether it can be installed alongside the existing setup.
It mainly comes down to the specific board you're using, not just the CPU. The board controls how PCIe lanes are arranged. Make sure your board allows GPU passthrough—most won’t handle it well. Almost every board has a dedicated slot for the chipset, so it shouldn’t impact the GPU. I’m leaning toward a M1 Mac; it should run macOS smoothly, and Apple is likely to phase out Intel support soon, with some ARM-only features already available on macOS.