PCIe generations generally double the speed of the slot over the previous generation. What that means for an end user is that a PCIe 5 x8 slot is roughly equivalent to a PCIe 4 x16 slot so you can assume that if the card worked well on a previous generation board then it should work just as well on a newer board with fewer lanes.
One of the main features of PCIe is also that the number of lanes is established and negotiated at startup so if a port only has x8 available, then both sides will go "oh, I only have 8 lanes available" and will carry on regardless, simply using less lanes that it originally expected.
At worst I might expect to see a difference in FPS measured in single digits, if you could measure any difference at all that wasn't attributable to simple measurement error.
From Wikipedia the transfer rates are as follows.
Version |
Transfer rate per lane |
x1 |
x2 |
x4 |
x8 |
x16 |
1.0 |
2.5 GT/s |
0.250 GB/s |
0.500 GB/s |
1.000 GB/s |
2.000 GB/s |
4.000 GB/s |
2.0 |
5.0 GT/s |
0.500 GB/s |
1.000 GB/s |
2.000 GB/s |
4.000 GB/s |
8.000 GB/s |
3.0 |
8.0 GT/s |
0.985 GB/s |
1.969 GB/s |
3.938 GB/s |
7.877 GB/s |
15.754 GB/s |
4.0 |
16.0 GT/s |
1.969 GB/s |
3.938 GB/s |
7.877 GB/s |
15.754 GB/s |
31.508 GB/s |
5.0 |
32.0 GT/s |
3.938 GB/s |
7.877 GB/s |
15.754 GB/s |
31.508 GB/s |
63.015 GB/s |
So instead of 63GB/s transfer rate you should get 31.5GB/s.
It absolutely should work at x8 and should work well but you will have a barely perceptible time difference in uploading texture and geometry data.
To fully fill a 24GB 4090s memory would take approximately 0.75 seconds instead of 0.425 seconds.
That a somewhat unrealistic workload for gaming and assumes that all the data is held in main RAM, in practice you will be bottlenecked by your storage device as the fastest NVMe drive is somewhere around 7.5GB/s.
For scientific workloads where you are processing data on the card and copying to RAM then you may well notice a difference, albeit very slight.