A Call For Measurements: Isolating the Thunderbolt Effect.
@enjoy I'm adding FPS # and screen captures as I go. I'll switch the Thunderbolt connection to CPU PCIe slot after I'm done testing the SB PCIe slot.
Thank you! Heaven & Valley again show us close to 20% performance drop like in PCIe vs. Thunderbolt thread so that confirm all results even they are with different CPUs!
MacBook Pro (Retina, 15-inch, Later 2013) 3.2GHz Quad Core Intel i7-4750HQ / 8 GB 1600 MHz DDR3 / 256GB SSD + 1TB
✪ mini eGPU ● PCI Express vs. Thunderbolt ● Mac CAN game ● Gaming Laptops vs. MacBook Pro with eGPU
The shortlist of bandwidth sensitive games at https://egpu.io/forums/thunderbolt-enclosures/alert-akitio-node-half-h2d-bandwidth-issue/ would be a good place to do x16-desktop vs x4-desktop vs TB3-CPU vs TB3-PCH FPS performance comparison.
Great work @itsage, keep it up! I noticed that the minimum FPS number is higher in the eGPU than in the dGPU. For Unigine Heaven for example, it is 9 FPS vs 22 FPS. Do you think this is just some margin of error, so every time running gets different numbers?
I also have a question/suggestion: Nobody can run FPS>144 because of screen refresh, and most people can only do <60FPS, right? Then, in terms of performance for gaming, we should not really care about any FPS>144 or >60. Is there a way to cap the FPS via Vsync/Gsync when running the benchmark?
The reason I am asking is because I think then the average FPS could be much closer between dGPU and eGPU, as you reported in Superposition results in the table. In Heaven and Valley, the max FPS of the dGPU are much higher than for eGPU, so we don't want that to be driving our results. So the performance difference when FPS<144 are the ones we care about, is that correct? And obviously, as a big fan of eGPU, I don't want to conclude there are performance losses so fast :).
I filled out the benchmark table with FPS #s and links to screen caps. I will connect Thunderbolt 3 connection to CPU tomorrow and run the eGPU benchmarks again. I have a 4K FreeSync monitor to test higher resolution with an R9 Fury X next.
I switched the Thunderbolt EX3 card to use CPU PCIe slots. That did not work. There was no detection of the enclosure when Thunderbolt connection was routed through CPU PCIe slots.
@vava726 good point. I'm wondering if the Thunderbolt connection I'm using is at Thunderbolt 2 speed. Given the very low numbers in CUDA-Z and the the BIOS reporting Falcon Ridge controller. http://ark.intel.com/products/76719/Intel-DSL5320-Thunderbolt-2-Controller