2018 15" ASUS ZenBook Pro (GTX1050) [8th,6C,H] + RTX 2080 Ti @ 32Gbps-TB3 (Razer Core) + Win10 1809 [itsage]
I’m continuing our investigation on lower-than-expected eGPU performance of 8th generation H-CPU hosts. After seeing the Dell XPS 9570 and its lack of PCH cooling, I exchanged it for the ASUS ZenBook Pro UX580GD. As seen in the photos below, this laptop has no cooling on the PCH chip either. The CPU and dGPU cooling system looks substantial. Even the soldered memory chips have a heat shield. It’s unfortunate Intel and most laptop vendors overlooked the PCH. I provided some cooling for the PCH by applying a 1mm-thick thermal pad square. This helped the PCH temp consistently stay below 70 degrees Celsius during eGPU use.
2018 15″ ASUS ZenBook Pro UX580GD – i7-8750H/GTX 1050 dGPU/UHD Graphics 630 iGPU/16GB RAM/512GB SSD
This ZenBook Pro has a very unique feature called ScreenPad. Its trackpad is an actual FHD touch screen. There are four modes of operation, ScreenPad apps, Extended monitor, Normal trackpad, and Disabled. The most useful mode is ScreenPad apps where you can quickly engage a selection of apps available for this touch screen. The trackpad functionality itself works great. It seems a little more sticky sliding my fingers on its surface due to the finish. Other than that I have no complaints.
One thing that caused an issue is the attachment of the GTX 1050 dGPU to this ScreenPad full time. I tried to disable the dGPU so that it’s more straight forward using the eGPU with iGPU. ASUS ScreenPad software would immediately error out and there’s no output on the ScreenPad (trackpad remained working). This ruined my impractical exercise of forcing the RTX 2080 Ti eGPU to accelerate this tiny 5.5″ FHD monitor 😛 . The RTX 2080 Ti eGPU was able to make use of all 5 display outputs. 1x USB-C powered the ASUS portable monitor, 1x HDMI powered the left LG 4K monitor, 1x DisplayPort powered the right LG 4K monitor, and 2x DisplayPorts powered the HP 5K monitor.
Another issue I encountered is the driver conflict between GTX 1050 dGPU and Nvidia eGPUs. If you were to simply hot-plug an Nvidia eGPU and use the existing GeForce drivers that Windows installed, the Nvidia eGPU would work but Nvidia dGPU would suffer error 43. To resolve this conflict, I used DDU to remove all Nvidia graphics drivers and changed the settings for Windows to not automatically install drivers. Once this was done, I made sure both Nvidia dGPU and eGPU showed up as “Microsoft Basic Display Adapters” in Device Manager before installing the latest Nvidia drivers.
The two Thunderbolt 3 ports in this ZenBook Pro have x4 PCIe connection but no cooling on the PCH chip is a hindrance on performance. Below are Unigine and 3DMark numbers through the HP 5K monitor.
It’s unfortunate Intel and PC vendors don’t pay more attention to Thunderbolt 3 in general and eGPU use in specific. We hope our discovery of performance deficiencies would alert users and manufacturers to take actions. If you have these Intel H-CPU laptops and the performance have not been up to your expectation, open the back cover and let us know what you find.
@itsage, an unusually low result there for Firestrike. Would you like to run the Forza 4 benchmark and drop a result into the following thread for comparison? Forza 4 is quite comprehensive and is sensitive to bandwidth.
@nando4 There’s thermal throttling with this ZenBook Pro. The same RTX 2080 Ti eGPU paired with 2018 13″ MacBook Pro got 21,818 in Fire Strike graphics score. Unfortunately I no longer have access to it. I will run the RTX 2080 Ti eGPU with 2018 15″ MacBook Pro this week.
Finally it comes my favorite theme 😆 Even system department from Razer has no clue about this. @itsage this is why I sold my 2080ti. Graphics performance is lame on my 8750H, compared to previous Blade 14 2017 and stealth 2016.
I still have the numbers.
Blade 15 8750H 2080ti
Blade stealth 7500U 2080ti
@nando4 i shared as well FH4 2080ti egpu with Blade 15 as host, I think it’s due pch things, not quite sure, the great aspect is only 20 degrees cpu improvement compared playing on dgpu and on stealth some cpu demanding title makes cpu throttling, but with Blade 15 there’s no such cpu throttle even though the graphics performance is lower.
Razer Blade 15 2018, i7 8750H 16gb ddr4 2666mhz 1tb 970Evo
Razer Blade Stealth 12,5 2016 i7 7500U 8gb ddr3 1866mhz 500gb 970Evo
MacBook Pro 13 mid 2012 i5 3210M 12gb ddr3 1600mhz 500GB HDD
Razer Core V2 rtx 2070 Black 2040mhz Boost clock
Previous setting: rtx 2080ti Asus Turbo, rtx 2080 xc gaming, rtx 2080 Fe, Zotac Mini 1080ti, Evga 1080 FTW2 gaming.
Dual Monitor Predator XB241YU 165Hz 1440p, LG l24UD58 60Hz 4K
Portable Monitor Magedok 1440p HDR 60Hz usbA/C monitor
I'm sorry if this isn't strictly relevant to this post but I was wondering two things specifically about the ASUS UX580GD.
1. Where did you buy the thermal pad from?
2. Do you think repasting the CPU could stop it from throttling as easily? If so, how would one do it?
I tried looking online for a guide but couldn't find one anywhere. Upon opening up the laptop I saw two wires (The ones coming straight at you as you look at the laptop) over the heatsink that I wasn't sure were removable or not. Any help would be greatly appreciated.
Again sorry for being off-topic.
@pb151 I had some spare thermal pads from heatsink kits. You can buy these from any online computer part stores. I would say to apply thermal pads on the PCH chip first before repasting the CPU/GPU and heatsink of the laptop.
PCH throttling due to the lack of heatsink is really unbelievable nowadays.
Dunno why designers keep making these kind of mistakes, without speaking about QC/QA teams letting this filter through the pipe.
From ages one could find at least a bare flat piece of metal onto the PCH to increase a bit the surface area, it cost less than 1$ and still do the job helping the die to cool down a bit.
Then if users guessed that having a thermal pad on top of the PCH to contact the chassis is a good idea, one can genuinely wonder why the designers didn't thought about right away.
Something so simple to get a validation for, yet implemented completely wrong!!!