Whoa... Mac Mini much faster with eGPU NOT going to external monitor  

  RSS

Macbain80
(@macbain80)
New Member
Joined: 3 weeks ago
Posts: 3
December 7, 2018 1:25 am  

Results are way better if the monitor is plugged into another TB3 port on the computer instead...

http://barefeats.com/mac-mini-needs-egpu-pro-apps.html

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


ReplyQuote
mac_editor
(@mac_editor)
Noble Member Moderator
Joined: 2 years ago
Posts: 1596
December 7, 2018 2:50 am  

I believe he did not select the "Prefer External GPU" option for the Pro apps (I can speak for FCP/Motion) when using the display directly connected to the eGPU, hence the lackluster performance. In these pro apps, what's seen on-screen likely has a different rendering queue (which defaults to eGPU connected display without setting the preference as mentioned) while the video processing components used the built-in GPU (evident with similar performance to UHD 630 - setting the preference changes rendering for video processing as well). Poor testing. Basically, monitor is inconsequential for video processing. Where monitor matters is for things like timeline performance (from my experience with FCP/Motion).

purge-wrangler.shpurge-nvda.shset-eGPU.sh
----
Troubleshooting eGPUs on macOS
Command Line Swiss Knife
----
Multiple Build Guides


theitsage liked
ReplyQuote
Eightarmedpet
(@eightarmedpet)
Noble Member
Joined: 2 years ago
Posts: 1132

mac_editor liked
ReplyQuote
Pacinoho
(@pacinoho)
Active Member
Joined: 1 week ago
Posts: 6
December 9, 2018 1:20 am  

He did say it checked prefer external GPU so the test is legit.

The say I see it, previous MacBook Pro internal vs external test are mostly FPS test, that you need the display card to directly output to display, while the test above is the opposite. There are render test that data needs to go back to the CPU and write to files etc. So it make sense to not have the display using up bandwidth, memory and power of the GPU and the TB3 and you actually see an improved result. 

Also so I believe the Mac Mini has doubled the PCIe lanes than the MacBook Pro, so it has much better bandewidth allocation if the monitor is used on a different port.

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


ReplyQuote
mac_editor
(@mac_editor)
Noble Member Moderator
Joined: 2 years ago
Posts: 1596
December 9, 2018 2:16 am  

@pacinoho he didn’t do it when the monitor was connected to the eGPU so the test is incorrect.  https://twitter.com/barefeats/status/1068560357155688448?s=21

See the thread. In his reply to me when I pointed this out, he said that the preference is not needed if a monitor is connected to the eGPU. That is incorrect for compute (FCP export, for example). The BruceX test is a blaring red flag proving this. The Vega 64 could never do so poorly in any condition except if it wasn’t used to begin with. Note how the result nearly matches the iGPU’s.

Mac Mini doesn’t have double the PCIe lanes vs. MBP, and anyway TB3 is 4-lane only at best. See @itsage build guides for more information. 

As I said earlier, for compute monitor is irrelevant.

purge-wrangler.shpurge-nvda.shset-eGPU.sh
----
Troubleshooting eGPUs on macOS
Command Line Swiss Knife
----
Multiple Build Guides


ReplyQuote
Pacinoho
(@pacinoho)
Active Member
Joined: 1 week ago
Posts: 6
December 9, 2018 3:28 am  

Ah you are right. 

I must have confused it with some old 13” MBP that doesn’t have full speed TB3. Seems it’s not the case in 2018 models.

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


mac_editor liked
ReplyQuote
Pacinoho
(@pacinoho)
Active Member
Joined: 1 week ago
Posts: 6
December 9, 2018 6:47 am  

But hang on, what you said makes sense on the FCPX and Compressor results, but how about the DaVinci Resolve 4444XQ export result? I am sure in Resolve you can simply select the GPU you which to use, regardless of “prefer external” setting. So in that test the “indirect” beats the GPU driven monitor. Perhaps there is actually some case that this hold true, that connecting no GPU driven monitor will actually make an export faster, at least in Resolve?

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


ReplyQuote