Theoretical Maximum GPU benchmark score on PCIe 2.0 slot

[hunt103]

So I'm working on a YouTube video where I'm testing cards ranging from a 1080 down to my 1050ti on every single PCIe slot. My methodology of using those slots without changing hardware is to cover the contacts on the graphics card to simulate x8, x4, x2, and x1 slots and then BIOS to simulate 2.0 and 1.0.

My actual question is what is the highest 3D benchmark capable on a 2.0 x16 slot?

(Each card is benchmarked thrice on each slot)
My GTX 960:

Slot	960
3.0 x16	5992.5
3.0 x8	5895.5
3.0 x4	5033.5
3.0 x2	3941.5
3.0 x1	3448.5

Based off of these results, it looks as if the slot starts to become a bottleneck on 3.0x4 slot which is a 1.0 x16 equivalent. But here's the funny thing; my GTX 1060 has very similar results with just higher benchmark results....(system would lock up on the 1060 after 3.0 x4 ) These results are intriguing because I was able to achieve a higher score on 3.0 x4 even though that's where the GTX 960 ran into bottlenecking.

Slot	960	1060
3.0 x16	5992.5	9730.5
3.0 x8	5895.5	9732.5
3.0 x4	5033.5	7617.5
3.0 x2	3941.5	dnf
3.0 x1	3448.5	dnf
Percentile Drop	1.62	-0.02
	16.00	21.72
	34.23	dnf
	42.45	dnf

When you factor in for errors in the data you see that there is for, all intents and purposes, no difference between 3.0 x16 and 3.0 x8 on cards that benchmark at roughly the same as the 1060 and slower.

Does it seem like my testing methodology might not isolate issues within the GPU in running on sub PCIe 1.0 x16 bandwidth? Or is it even worth benchmarking cards that are crippled to that degree? From what I've found in my research is that the 1070 and 1080 are some of the only cards that are actually capable of maxing out the bandwidth on a 2.0 PCIe slot.

[David (PassMark)]

You didn't mention which CPU you are using. Using the fastest CPU should put more load on the video card, and thus you might also notice more scaling as you move from a x16 to x8 slot. The 1080 TI might also place more demands on the bus and you might see some scaling there.

I was able to achieve a higher score on 3.0 x4 even though that's where the GTX 960 ran into bottlenecking

The score dropped from 9732.5 to 7617.5, this is a lower score, not a higher score.

system would lock up on the 1060 after 3.0 x4

Maybe the insulation of the pins wasn't working as well as you expected. Or maybe you blocked some power pins and the video card was under powered.

no difference between 3.0 x16 and 3.0 x8

I don't have time to dig out the article, but other people did similar experiments in years gone by with SLI running with two x8 slots (rather than two x16 slots) and most of the time there wasn't a big difference for gaming. It depends a bit on the game, the CPU and the cards in use however.

I wouldn't assume a 3.0 x1 slot is the same as a 1.0 x16 slot. Maybe in theory they are, but in real life I would imagine they don't behave exactly the same. (But does it matter as no one is using PCIe 1.0 now anyway)