Asus XG Station Pro + GTX 1660 with Gigabyte Aero 15Wv8 low performance?
I recently bought my Gigabyte Aero 15Wv8 with built-in GTX 1060 and TB3 x4 connector. I bought also an XG Station Pro eGPU enclosure and wanted to use my existing GTX 1660 card. The setup works, but performance is under expected level except for benchmarks. For example:
Unigine Superposition 1080p High:
Division 2 Ultra DX12:
- GTX 1060 Mobile 47fps
- GTX 1660 as EGPU 46fps and pulsating, drops in every 2. or 3. second
- GTX 1660 with Ryzen 2600 in PC 63fps
- GTX 1060 Mobile 51fps
- GTX 1660 as EGPU 37fps and pulsating like in Division, but a bit more acceptable
- GTX 1660 with Ryzen 2600 in PC 64fps
The GTX 1660 utilization is around 96-98%, so that seems to be fine. PCH temperature is around 55°C and PCI-E speed is x4. What else could be wrong? Driver is latest, thunderbolt firmware updated from Asus, although date is from 2018.03. I have read that 8th gen CPUs are not the best for EGPU, but that was mainly because of the lack of the PCH cooling. Any other idea?
I have a GTX 1660 in an enclosure as well. Don't have any of those games to run a comparison, but I'm not surprised you aren't seeing much performance improvement over a built-in 1060. As a whole, the 1660 isn't a massive upgrade over the 1060. It's certainly quicker (Anandtech estimates about 28% overall performance improvement over the 1060 3GB), but the difference isn't night and day. When you factor in the standard eGPU/TB3 penalty, you probably wouldn't see much difference except in (1) scenarios where the 1660's memory bandwidth improvement avoids a bottleneck, or (2) scenarios where the laptop's CPU is throttling because of the combined CPU/GPU thermal load and hence removing the GPU's heat allows higher CPU performance.
One thing I will say is you mentioned pulsating framerates every 2-3 seconds. I experienced the same thing in my setup before disabling the dGPU. If you haven't done so already, go to your Device Manager and disable the 1060. That smoothed out the performance of my system in games very substantially.
If the GTX 1660 has +28% over the mobile 1060, then actually I would expect overall +8% performance in games with the EGPU. Instead, I have minus 2-22% depending on games, which is under expectations.
Are you running the eGPU out to an external display, or back to the laptop's internal monitor? If the latter, you should expect an additional penalty because the signal has to pass back through the same cable.
Separately, the 28% is a very rough overall figure that doesn't capture all of the subtle differences between the cards in different applications. You can read the full Anandtech review to see how the 1660 really compares to both the 1060 3GB and 6GB in various benchmark/compute/gaming scenarios ( https://www.anandtech.com/show/14071/nvidia-gtx-1660-review-feat-evga-xc-gaming ). The 2-22% penalty could be accurate or the result of a non-representative sample size.
In general I don't imagine an external 1660 makes much sense over an internal 1060, absent one of the two scenarios outlined above (where either the 1060's memory bandwidth or the laptop's combined thermal load cause specific bottlenecks). A 1660 Ti would probably be the minimum card worth considering here. A GTX 1080 (if you can still find one at a discounted price) or RTX 2070 would be much more appropriate and should show very meaningful improvements in any use case.