I'm putting together a computer for light gaming and AI tasks and planning to use two NVIDIA GeForce RTX 5060 Ti graphics cards along with an MSI MPG X870E Carbon WiFi motherboard. The specs say this motherboard supports x16x0x4 or x8x8x4 configurations, but I'm confused about the bandwidth distribution. The motherboard details mention that the PCI_E1, PCI_E2, and M.2_2 share bandwidth and that support varies depending on the CPU. If I install an M.2_2 device, will it throttle the performance of my GPUs? I'm pretty new to this, so I'd love any insights or advice!
2 Answers
Just a quick question—what type of software are you planning to use with both GPUs? Generally, for most AI applications, they typically run separate tasks on each GPU rather than combining powers. You can't really double the VRAM this way either, so it might not speed things up significantly. Considering getting a used 3090 may be better in some cases, depending on what you're running.
The PCI_E2 slot only supports x4 bandwidth. If you install a GPU in there, the PCI_E1 will drop to x8. Also, keep in mind that the x4 lanes on PCI_E2 are shared with the M.2_2 slot, which may affect your GPU speeds if you're using that M.2 slot. So, yes, using the M.2_2 will slow down your two GPUs if you want them to run at full capacity. It's a good idea to plan your component usage carefully!
I noticed that in the specs, it says there are two PCI_E1 slots, which might be misleading. Just confirm that there’s only one x16 slot to ensure your setup complies with the bandwidth constraints.
I'm mainly using Stable Diffusion along with ComfyUI and its Distributed Extension. I got both 5060 Ti's for $750, which is a decent deal compared to a used 3090 that I found.