2019 13" MacBook Pro [8th,4C,U] + RX 5700 XT @ 32Gbps-TB3 (Netstor HL23T-Plus) + Win10 1903 [itsage]
The RX 5700/XT graphics cards don’t currently have drivers in macOS so this build is in Windows 10 Boot Camp only. With the 2019 13″ MacBook Pro running 1903, I can hot-plug both Nvidia and AMD eGPU without any workarounds. All four Thunderbolt 3 ports are working well. I installed Thunderbolt Control Center to monitor device detection just in case there’s issues. It’s not neccessary on a Mac in Boot Camp.
Mid 2019 13″ MacBook Pro – i7-8569U/Iris Plus Graphics 655 iGPU/16GB RAM/512GB SSD
Netstor HL23T-Plus + RX 5700 XT + .5m Thunderbolt 3 cable
I recently learned it’s possible to hot-plug AMD eGPU in Boot Camp using Windows 10 1903 version. As soon as the RX 5700 XT eGPU was connected, the MacBook Pro would freeze for about two minutes while it detected new components and configured drivers. Audio and Network icons in the Taskbar notification area shown as disabled during this process. Trackpad and keyboard are non-functional also. Once the Thunderbolt 3 connection was established, everything resumed as normal. I tried Radeon drivers version 19.7.1 first then the latest 19.7.3. There’s a slight improvement in compute performance.
|Version 19.7.1||Version 19.7.3||eGPU & 5K Same Controller|
Besides testing internal display mode, I wanted to run loopback mode by connecting a 5K monitor through a Thunderbolt 3 port. First run was through the same controller as the eGPU then second run had it through a separate controller. Last run is through the external monitor directly through the RX 5700 XT eGPU.
|Internal Display||Loopback Same Controller||Loopback Separate Controller||External Monitor|
It seems the Radeon drivers for RX 5700/XT are not well optimized for eGPU use just yet. Unlike Polaris and Vega cards, the performance loss for Navi cards is much greater through internal display vs external monitor. Loopback mode showed some interesting result as well. I reran 3DMark several times to make the results were consistent. At the moment it’s best to use RX 5700 XT through an external monitor directly.
Ahhhhhhh rats, maybe I should cancel my order then.
Cracking write up as ever, thanks for testing out loop back too! It’s crazy how far eGPU implementation has come when you compare this to the earlier 2016 MacBook builds.
Comparing the benchmarks to yourVega 56 ones it looks like the 5700xt actually performs a little worse in loopback! Argh.
@eightarmedpet Yes the drivers need more work imo. I tested the RX 5700 XT in discrete mode using the Threadripper desktop and it runs very well. I also did some quick eGPU testing with this X399 desktop.
Thunderbolt 3 monitor output was straight-forward with the Gigabyte GC-Titan-Ridge AIC too – feed 2x DisplayPort cables into the AIC then run a Thunderbolt 3 connection to 5K monitor. In my test I needed a Thunderbolt 3 to dual DisplayPort adapter to use the HP Z27q.
|RX 5700 XT eGPU||RX 5700 XT dGPU|
Wow, another bunch of interesting test results... looks like Navi isnt quite ready for eGPU usage. Wondering if a Gigabyte Vega 64 would be a better option...
Do you have any idea why RX 5700 XT comparison on desktop and eGPU vary so much?
If I'm not wrong, eGPU performance loss on Superposition is less than 10%, where in tests like Firestrike and Timespy eGPU loss is 30~40%.
In Unigine Heaven you have an eGPU loss of around 15%.
From my personal experience with AORUS Gaming Box with MacBook Pro 15 on Windows, eGPU performance loss is around 15-20% on 1080p resolutions and 10 - 15% on 1440p or 4k resolution. Do you have this 10-15% results also?
I'm wondering how it performs in real modern games, like Shadow of Tomb Raider or Assasin's Creed Odyssey built-in benchmarks.
Assuming Apple will support Navi cards in macOS Catalina sooner than latter, Radeon 5700 XT is a good option for people who need to choose 1 single GPU that works perfect both on Mac and windows. But only if eGPU loss follow GeForce GTX patterns (10-15% on average on 1440p and 4k resolutions)
@precision01 To clarify I was running the RX 5700 XT eGPU through loopback mode on X399 Desktop. It’s one of the worst case possible so the performance difference between eGPU vs dGPU is huge. If the eGPU was through external monitor directly it would perform much better. You can see MacBook Pro + RX 5700 XT eGPU in external monitor mode is fairly good.
At this point in time, it’s likely driver optimization or lack thereof for the Navi cards. Polaris was the generation that launched with XConnect/eGPU support so the drivers for them in both Windows and macOS are very good. You can see a clear difference in 3DMark tests between the internal display vs. loopback mode in MBP results. There are other factors with the host computers as well. Thunderbolt 3 connection routing also makes a difference.
I see. Thanks for clarification.
So, we must compare, your first post (MacBook Pro results) with your next benchmark post (Desktop results).
So, results would be:
Superposition Macbook: 14357 (-9%)
Superposition Desktop: 15733
Valley Macbook: 4230 (-5%)
Valley Desktop: 4444
Heaven Macbook: 2307 (-9%)
Heaven Desktop: 2632
Timespy Macbook: 6702 (-32%)
Timespy Desktop: 8860
Firestrike Macbook: 15435 (-40%)
Firestrike Desktop: 21676
If my interpretation of the results are right, 5700 XT eGPU is performing very well on Unigine benchmarks, but horribly on 3Dmark.
Am I looking at something wrong here?
Similar to the RX 5700/XT cards, Nvidia RTX cards do not have drivers in macOS. I paired an RTX 2070 Super to the Mantiz Venus and ran some tests last night. While the internal display mode performance is better than the RX 5700 XT, external monitor mode is very similar. I will find time to run lookback mode soon.
|Internal Display||Loopback||External Monitor|
Hi, I got around Error 12 by booting into Mac and rebooting into windows, but doing that only results in an Error 43. Is this indicative of damage to my enclosure and/or graphics card? This is my first setup and it’s making me a bit anxious. Can anyone help?
I am currently looking at getting this Sapphire 5700xt ( https://www.newegg.com/sapphire-radeon-rx-5700-xt-100416p8gl/p/N82E16814202349 ) and putting it inside a Mantis Venus, attached to a Dell (S2719DGF) 1440p, 144hz, 1ms response monitor.
It will be hooked up to a 2018 Macbook Pro 13" I5, 8gb. I'll bootcamp it in the interim since there this is no MacOS support. Any glaring pitfalls I should look out for? I will be using this card for some light gaming (classic wow) and just wanted to be sure I could push ~144fps on high settings to a 1440p monitor. I was reading the 5700xt was optimized for 1440p, but it seems they are not optimized? Would it be beneficial to get a different card or monitor for my use?
@graphicsnonwizard That will be a nice build. While the Radeon drivers aren’t well optimized for eGPU use in internal display mode, external monitor mode should be good. There’s some loss through the Thunderbolt 3 interface but I think you will get nice results with your 144Hz monitor at 1440p. Let us know how it works for you.
@precision01 Like what @Mac_editor said, check your Windows 10 OS Build version to make sure it’s not 18362.329 & 18362.356. We found them to be causing error 12 for most Mac running Boot Camp. Don’t disable PCIe Controller x16 1901 in your system. It will make the internal display sluggish.
When you say some loss through the TB3 port, what exactly are you referring too? Do you think I will get decent framerate with this setup or am I really limiting myself with the TB3 port?
Different connections have different bandwidths (for you this doesn't matter since TB3 is the fastest that your MacBook Pro can do):
PCIe 1.0 x1 = 2 Gbps
PCIe 2.0 x1 = 4 Gbps
PCIe 3.0 x1 = 7.877 Gbps, PCIe 2.0 x2 = 8 Gbps
PCIe 3.0 x2 = 15.75 Gbps, PCIe 2.0 x4 = 16 Gbps
TB3 = 22 Gbps
PCIe 3.0 x4 = 31.5 Gbps, PCIe 2.0 x8 = 32 Gbps
PCIe 3.0 x8 = 63 Gbps, PCIe 2.0 x16 = 64 Gbps
PCIe 3.0 x16 = 126 Gbps
PCIe 4.0 x16 = 252 Gbps
Higher bandwidth gives higher performance. Google "pci express scaling site:techpowerup.com"
You can see that dividing the bandwidth by 5 does not divide the performance by 5, so you should be ok.
For some games (like Deus Ex: Mankind Divided), performance drop when lowering bandwidth decreases when resolution increases.
For others (like Civilization VI), the opposite is true.
Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts
Do either of you guys have a link to the correct build I am to download to get the best performance out of an external display? I am afraid I am going to grab the wrong build.
And if I should avoid KB4515384, .329 & .356 what should I get? And What are those numbers referencing?