Myths vs Benchmarks
https://www.reddit.com/r/Dell/comments/5rxclq/xps_9560_w_razer_core_1080_gtx_benchmark_requests/
Thanks for the article Douglas. I'm very excited to see the eGPU revolution finally reach consumers. And I agree we should all be watching the Latitude 5480, 5580 and Precision 5520 for viability. Their support of WiGig and better security options. Mind you, they don't have as bright a screen as the XPS 15, a low power pascal GTX 1050, nor the polished aluminum build. But they're still very interesting options that we'd love to have tested with eGPUs.
The article suggests that the Latitude 5480 is wired differently than the XPS 15, with all x4 PCIe lanes. Is there any evidence backing this up yet? I've been looking. So far, I've only found a second-hard account reportedly from @DellCarePRO listing some but not all Thunderbolt laptops (excludes AW, etc).
https://www.reddit.com/r/Dell/comments/5ufjk9/dell_thunderbolt_number_of_pcie_lanes/
eGPU performance (FPS) is not "half speed" with x2 lanes
That's a myth. x2 PCIe lanes provides 16GTps which rarely affects the utilization of even the most powerful GPUs. This all reality isn't intuitive given the complexities of eGPUs and Thunderbolt. But gaming isn't nearly so affected as many expect.
PCI scaling tests by TechPowerUp show that 16GTps showed at worse 7% impact on FPS and that's based on x4 PCIe 2.0 and its 20% overhead. Overhead in PCIe 3.0 is now just 1.8% so the impact should be even less with an XPS 15. PCIe scaling would really depend on the specific game and the specific moment within the game. As more benchmarks come in, you'll notice games are barely affected. The impact may be more noticeable in other PCIe applications like large SSD RAID arrays.
https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080_PCI_Express_Scaling/24.html
The article also suggests that the XG Station 2 is not viable until Dell certifies for eGPUs. That may or may not be true. To date, I believe only one XG Station 2 (@jackeyjoe) has been tried and there's some indications that the specific unit may be defective. We're still trying to confirm that.
https://egpu.io/forums/pc-setup/dell-xps-15-9560-asus-rog-xg-station-2-a-tragic-love-story/#post-2794
Also, there's some question if using the internal display has as much impact on the limited PCIe lanes or not. This too is the common belief. Yet both PCIe and Thunderbolt are bi-directional with their upstream and downstream bandwidth operating independently so this may not prove true. Again, we're looking to test this more.
Overall, I commend the article for increasing awareness of the wide range of options for eGPU. I believe eGPU capable Thunderbolt will become as common as USB is today. And I agree the Latitude 5480 is very interesting. But readers should note that these technologies are, as of yet, barely tested or understood. There is a lot of rumors and misinformation being promoted that needs ground-truthing.
https://thunderbolttechnology.net/sites/default/files/HBD16235_Thunderbolt_TB_r05.pdf