bench.txt

Bench 1, Q35 Unpatched, root port: 1B36:000C
Lspci: x8/1.0, Nvidia: x1

  3DMark Time Spy:
    4009 4503
    3996 4549
    3999 4564

Bench 2, Q35 Unmodified patch [x16], root port: 1B36:000C     
Lspci: x8/3.0, Nvidia: Unknown bus      

  3DMark Time Spy:
    4115 4556                
    4128 4613                
    4122 4561                

Bench 3, Q35 Unmodified patch [x16], root port: 8086:3420
Lspci: x8/3.0, Nvidia: Unknown bus

  3DMark Time Spy:
    4116 4589
    4097 4544
    4102 4530

Bench 4, Q35 Modified patch [x8], root port: 8086:3420
Lspci: x8/3.0, Nvidia: x8/3.0

  3DMark Time Spy:
    4107 4564
    4104 4573
    4114 4569

  Unigine Heaven:
    2035 80.8
    2054 81.5
    2054 81.5

Bench 5, Q35 Modified patch [x8], root port: 1B36:000C
Lspci: x8/3.0, Nvidia: x8/3.0

  3DMark Time Spy:
    4107 4590
    4103 4578
    4109 4532

  Unigine Heaven:
    2063 81.9
    2060 81.8
    2061 81.8

Bench 6, 440fx
Lspci: x8/3.0, Nvidia: x0

  3DMark Time Spy:
    4112 4604
    4110 4592
    4098 4553

  Unigine Heaven:
    2059 81.7
    2056 81.6
    2056 81.6

Bench 7, Native
Nvidia: x16/3.0 (should be x8?)

  3DMark Time Spy:
    4196 4737
    4186 4713
    4264 4744

  Unigine Heaven:
    2116 84.0
    2138 84.9
    2135 84.7

CUDA bandwidth test: All cases except "Bench 1" show speeds consistent with x8 3.0 physical speed:
  
  Device 0 took 1029.413940 ms
  Average HtoD bandwidth in MB/s: 6217.129717
  Device 0 took 1019.704346 ms
  Average DtoH bandwidth in MB/s: 6276.329043
  Device 0 took 1190.849487 ms
  Average bidirectional bandwidth in MB/s: 10748.629559

Note: Nvidia drivers were reinstalled between each VM reconfiguration & MSI interrupts reenabled each time.

X16 patch: https://gist.github.com/gnif/e4c001b608347b0b86118a2647103378
X8 patch (apply after X16 patch): https://gist.github.com/aiberia/30196fd0b55556cb2840af4303d50216

Result with X8 patch applied: https://i.imgur.com/PIAduDO.png

Platform: Intel 6700k, nVidia GTX 1060 connected physically at PCIE 3.0 X8