The performance drop for external GPUs is a lot simpler: they are operating over an effective x4 link, not z16, and they are connected to the chipset, not the CPU directly (so compete for access over the same x4 link that all the other chipset-connected peripherals have access to, like your SSDs). Testing at different PCIe bandwidths shows a 5%-10% penalty from the link speed reduction.
If you are feeding the rendered image back over the PCIe bust to the host system for display, there will be an additional hit to performance due to bandwidth used on passing images back. Depending on how the driver implements things, this may be an outsized hit if there is a wait while the finished frame is sent back before instructions for the next frame are sent.
If you are feeding the rendered image back over the PCIe bust to the host system for display, there will be an additional hit to performance due to bandwidth used on passing images back. Depending on how the driver implements things, this may be an outsized hit if there is a wait while the finished frame is sent back before instructions for the next frame are sent.
