TB3 (4-lane) better performance during iGPU only mode vs. dGPU; reasons?
TL:DR - Why is there better eGPU performance (as measured by SuperPosition) when laptop is set to iGPU only mode vs dGPU-only or Hybrid (Optimus) modes?
-4x TB3 Connection
-Active TB3 cable
-Laptop AC Adapter plugged in for all testing
I recently obtained an Intel 10th Gen Laptop w/ i7-10875 and a RTX 2070S Max-P from Eluktronics. Since I had a previous eGPU, I decided to compare the performance on this laptop in dGPU only mode, iGPU only mode, and Optimus (hybrid) mode. I can set this option in the BIOS.
My hypothesis was that in dGPU only mode, the eGPU would perform better. However, as measured by Superposition by Unigine, I found that the eGPU performance was better by ~5% on average when using the laptop in iGPU only mode.
Any theories on why this behavior is the case? I figured the weaker iGPU would be slower when serving as a pass-through. I also ordered a ADT-LINK RG43SG to test with the same system and see if any further performance gains can be optained.
Any theories on why this is the case? The common wisdom points to a the dGPU only mode being better for performance (though worse for battery life)
What is the external monitor connected to? What display (external or laptop) is being used in each test?
The external monitor is hooked to the GPUs Display port out. Each test was performed with this configuration.
Would I get better performance if I connected the external monitor to the laptop's HDMI out? My initial thoughts are in doing so, wouldn't I run into the issue of needing to loop back display information through the TB3 cable?
If the display is always connected to the eGPU then I don't know why iGPU performs better than dGPU or Hybrid - except maybe iGPU uses the least power of the three so it runs the coolest and therefore the CPU can use more power?
Appreciate all the insight! @joevt and @mini-i5, you guys both seem to have pointed in the right direction. Going off @mini-i5's suggestion about cooler laptop offering more headroom, I decided to use fanboost to really crank up the fans to keep CPU temps at a minimum (while on iGPU only mode). This actually resulted in slightly lower benchmark scores (by about ~3%) than when I let the fans go normally.
My hypothesis is that with the fanboost, the system draws more power overall which means less is available to the CPU and therefore, slightly worse performance is seen. Waiting for a ADT-LINK RG43SG to arrive but I wonder if the same performance will be the case once I can cut TB3 out of the equation and rely solely on the m.2 NVME PCI-E x4 connector?
@ishikawa_goemon, M.2 slot connection is very close to an internal x4 PCIe connection. Due to no Thunderbolt 3 overhead, the performance is more predictable in games and software. You can read @nando4's build for more detailed explanation.