2013 Mac Pro (FP D300 x2) [3rd,12C,E] + RP 580 @ 16Gbps-TB2>TB3 (Blackmagic eGPU) + macOS 10.15.4 // TB Display with TB3 to TB2
2013 Mac Pro (12 Core with 2x Firepro D300), 64GB RAM, 2TB SSD, 6x TB2 interfaces
Blackmagic eGPU w/ Radeon Pro 580
- Connection to Apple Thunderbolt Display via original Apple TB3 to TB2 adapter
No enough posts for being able to attach pictures, I made plenty of it...
- Follow first installation step for purge-wrangler, step 2A: Disable SIP
- Connect the eGPU via Apple Adapter to the Thunderbolt 0 bus (lowest row), otherwise the Mac won't boot
- First leave the TB Display connected directly to the Mac one or two rows higher than the eGPU
- Continue with steps 3 and 4 of the purge-wrangle beginner's guide. It may take a while, be patient until it offers to reboot.
- After reboot, the eGPU should be detected. You can find the symbol to eject the eGPU
- If you plan to leave at least one Monitor connected directly to the Mac, you're done here.
- If your only and primary display shall be connected to the eGPU, first you won't see an image if you switch the monitor from the Mac to the eGPU. The Mac seems to remember that the primary display was connected directly, if there's no display connected to the Mac, your primary display is missing
- To overcome that, I connected the Thunderbolt Display to the eGPU and another display via HDMI directly to the Mac. After booting, the Thunderbolt Display comes also up. If you then change the primary display to the TB Display, next boot the Mac remembers that. You can disconnect the 2nd display now. You won't see the boot screen, but after booting, the TB is now also available even if there's no other display connected
Even though speed was not my main motivation for the eGPU, I did some comparisons and found a strange behaviour
Geekbench 5 (Metal):
- FirePro D300: 23120
- RX580 (Display connected to Mac): 39679
- RX580 (Display connected to eGPU): 30220 (Why is it slower in that case?)
Unigine Heaven (OpenGL, 1920x1080 fullscreen, Quality: Medium, everything else disabled):
- FirePro D300: 1217
- RX 580 : 1724
My motivation to buy an eGPU was not gaming as probably many others. My main point was to be able to attach a 5K Display to my old Mac Pro. Previously there were some displays available with dual TB2, but those died out meanwhile.
Also I'm hoping to increase the rendering speed of FCPX, but currently I can't see an influence...
- Unclear why the eGPU get's slower when a monitor is attached to it
- The thermal management of the Blackmagic eGPU is clearly noise oriented. It derates quite early. If you rather think about overclocking than noise level, it's the wrong choice. But even compared to the very silent Mac Pro: After 10mins of Unigine Heaven, the D300 get's significantly warmer and you can hear the fan speed increase of the Mac Pro. If it runs on the eGPU, there's almost no temperature change and no increased fan speed.
- The image quality of the Thunderbolt Display appears to be better if it's connected to the eGPU. I know it's a digital connection, but I made some macro pictures. Unfortunately, I'm not allowed to attach these here, but you see the difference.
- FCPX appears to be not influenced at all. Speed is almost the same as with the D300. Also GPU load is almost not there. I render in 4k 10bit HDR HVEC codec.
- If the only monitor you have is attached to an eGPU (I assume this is not specific for my configuration), the monitor comes up only at the end of the boot sequence. Before it's just black...
- All features integrated in the Thunderbolt Display work fine: Facetime camera, Speaker control, microphone...
Next step for me: Purchase a high res monitor and try it with the eGPU, but this may take a while...
Update April 30th, 2020:
It didn't take as long as expected to purchase an Ultrafine 5k as I got a great opportunity. As already mentioned in this forum, it works without any issue to connect this monitor to the BM eGPU. The full resolution was immediately set, but I changed the setting to most space otherwise the space is the same as with the Thunderbolt Display.
I guess it is important to use the TB3 cable included in the monitor, as I saw also reports where people had to use SwitchResX to switch to full resolution. Maybe this happens when using the USB-C cable included.
- While the Thunderbolt Display came up just after booting, the Ultrafine immediately comes up, shows the apple and the bar, then disappears until booting is complete (strange)
- I repeated the benchmarks and while Unigine didn't show anything different, my concern with Geekbench 5 was solved: 44000. Obviously the conversion to TB1 on the display side has a big influence on the benchmark (still not clear to me why)
- The Thunderbolt Display is still a great Display. Colors of the Ultrafine are better, but it's not a shocking difference. The resolution is far better, this is the main benefit of this change
- I'm very picky in regards to Display coating. I know this forum is about eGPUs, but if you are also concerned: At least the latest version 27MD5KL-B has a glossy surface like an iMac. Please don't start a discussion about this, I know everybody has a different opinion, but for me this is important
- All functions of the Ultrafine Display work fine: camera, loudspeaker (with Mac control) and Mic
@mkc2lr, Thank you for sharing this build. My one recommendation is to not use GeekBench for assessing performance. It's very inconsistent and worthless when compared on version to the next. In my review of the Blackmagic eGPU, I concluded it's only good for your exact plan, powering the LG UltraFine 5K. If you're looking for performance boost, there are much better options. For example an ASUS XG Station Pro + RX 5700 XT will be about the same price, same noise level with the right GPU cooler, and much higher performance.
@itsage, Thanks for your comment. I totally agree that the BM eGPU is not the first choice for performance. The only reason I used Geekbench is that it works fine if the display is not connected to the eGPU. Generally if the eGPU runs slower when the monitor is attached, I don't understand.
I'd expect the other option worse where the eGPU computes and the other GPU needs to get the data back and computes the picture. In all reviews I read, this was actually confirmed.
I didn't find a logic explanation for the behavior in my case. Maybe you're right. It's just related to the specific benchmark I used...
@mkc2lr, glad to see another 6,1 eGPU user. I'm running a 6,1 with D700 also with a Blackmagic EGPU, just the Vega 56 flavor with an LG Ultrafine 5k (and Pro Display XDR because it just happens to be home with me in quarantine from the office). I honestly haven't had many issues with the setup, I wish you the same luck! I was surprised to see you needed to connect another monitor to the eGPU screen to light up? I'm not sure I've encountered that one before, but can confirm the bottom port is where mine is plugged in. I must have just blindly put it there and not known there would have been an issue otherwise. Strange. @itsage, any reason why that port makes eGPU work or boot properly?
@rawweb, Oh, wow this is the BM Pro eGPU in your case. Not bad !
I had the same luck, the Ultrafine 5k worked immediately without any issues. I updated the article as today (earlier than expected) I bought the Ultrafine Display 🙂
@rawweb, My guess is the Thunderbolt bus #0 to where the bottom two Thunderbolt 2/Mini-DisplayPort ports connect is also shared with the HDMI port. The firmware prioritizes this Thunderbolt bus and the associated monitor outputs so that the computer has boot screen.