Thunderbolt eGPU Possibilities for Early Intel Macs?

[TheShortTimer]

Thanks for updating your post with the 660 benchmarks.

I might be able to join you in a three-tier GPU comparison because I dug through my pile of parts and found this:

MSI AMD Radeon HD 7770 1GB GDDR5 DVI/HDMI/DisplayPort PCI-Express Video Card R7770-PMD1GD5

MSI AMD Radeon HD 7770 1GB GDDR5 DVI/HDMI/DisplayPort PCI-Express Video Card

www.amazon.co.uk

MSI HD 7770 GHz Edition Specs

AMD Cape Verde, 1000 MHz, 640 Cores, 40 TMUs, 16 ROPs, 1024 MB GDDR5, 1125 MHz, 128 bit

www.techpowerup.com

From what I can ascertain of the specs, the NVS-510 easily outclasses the HD 7770 in max resolution but the HD 7770's memory bandwidth is considerably faster at 72.00 GB/s vs 28.5 GB/s for the NVS-510.

I'd suggest checking out dortania's guide to supported AMD and NVIDIA GPUs and sticking to those natively supported in the version(s) of OS X/macOS you want to run, without messing with hacks or external drivers (such as Nvidia's web drivers which only work up to High Sierra).

According to Dortania's guide, the HD 7770 is supported up to Monterey with the help of the following:

Needed kexts

• Lilu.kext
• (opens new window)
• WhateverGreen.kext
• (opens new window)

Extras:

• radpg=15: Required for HD 7730/7750/7770/R7 250/R7 250X GPU initialization)

What do you think?

 

[Amethyst1]

From what I can ascertain of the specs, the NVS-510 easily outclasses the HD 7770 in max resolution but the HD 7770's memory bandwidth is considerably faster at 72.00 GB/s vs 28.5 GB/s for the NVS-510.

The 7770 runs circles around the 510. Both cards use 2012-era GPUs. The 7770 is a mid-range card; the 510 is very low-end. Both cards will do 3840×2160 at 60 Hz via DisplayPort, but the 7770 has a higher pixel clock limit (600 MHz?) than the 510 (540 MHz!), allowing it to also do 4096×2160 (confirmed by AMD) or 4096×2304 (unconfirmed; requires CVT-RBv2 timings) at 60 Hz.

According to Dortania's guide, the HD 7770 is supported up to Monterey with the help of the following:

The 7770 produces a black screen on Yosemite and later versions. So yep, some assistance from Lilu (which is already used by kryptonite) and WhateverGreen is required.

 

[Amethyst1]

Added results for a HP OEM Radeon RX 460 with 2GB VRAM. They are... weird. Nice to see Nexuiz finally staying playable using the Ultra preset but Heaven and Valley results are outright laughable, especially given the settings. Geekbench and Luxmark results look reasonable. I don't know why the Metal benchmark in the former reliably fails. I’ll have to do some tests on Mojave and on a more powerful Mac, as well as stick the card into a PC and run some expensive extensive tests on Linux or Windows. I did yank the heatsink to confirm the PCB really uses a RX 460 GPU (and not a 6450 or something with a modded VBIOS to misreport as one).

 

[TheShortTimer]

I'm lagging way behind but I'll catch up soon, I promise you that!

Does this mean that with the support of a more powerful GPU that an HTC Vive headset could be connected and VR stuff run on an i5 MBA/MBP?

 

[Amethyst1]

Does this mean that with the support of a more powerful GPU that an HTC Vive headset could be connected and VR stuff run on an i5 MBA/MBP?

Assuming the ancient dual-core CPU isn’t bottlenecking the GPU and VR setup too badly, I don’t see why not.

 

[Amethyst1]

Tested the RX 460 with my Late 2013 15" Retina MacBook Pro running Mojave. The results are much better.
Values in parentheses are for benchmark run on the indicated GPU but displayed on the other GPU.

2013 15" MacBook Pro - i7-4750HQ - macOS v10.14.6 18G9323
---------------------------------------------------------

Unigine Heaven Benchmark 4.0 Basic (OpenGL)
1280×720 windowed NoAA Medium

+---------------+---------------+---------------+
| Metric        | Iris Pro 5200 | Radeon RX 460 |
+---------------+---------------+---------------+
| FPS           |  35.7 (34.2)  |  61.9 (51.8)  |
| Score         |   899  (863)  |  1559 (1305)  |
| Minimum FPS   |  11.1 (11.2)  |  12.7 (11.6)  |
| Maximum FPS   |  71.9 (64.4)  | 110.3 (78.5)  |
+---------------+---------------+---------------+

Unigine Valley Benchmark 1.0 Basic (OpenGL)
1280×720 windowed NoAA Medium

+---------------+---------------+---------------+
| Metric        | Iris Pro 5200 | Radeon RX 460 |
+---------------+---------------+---------------+
| FPS           |  33.2 (32.2)  |  58.8 (48.5)  |
| Score         |  1388 (1347)  |  2459 (2030)  |
| Minimum FPS   |  17.4 (15.7)  |  27.3 (25.6)  |
| Maximum FPS   |  59.2 (65.3)  | 100.0 (75.9)  |
+---------------+---------------+---------------+


+---------------+---------------+---------------+
| Metric        | Iris Pro 5200 | Radeon RX 460 |
+---------------+---------------+---------------+
| GB4 OpenCL    |       26250   |       69750   |  
| GB5 Metal     |        5149   |       20511   |     
| GB5 OpenCL    |       18883   |       18580   |
| LuxMark Ball  |        2546   |        5505   |
+---------------+---------------+---------------+

I wonder if the 2011's CPU or High Sierra is to blame for the 460's abysmal performance in Unigine. The UI often exhibits heavy glitches on the display hooked up to the eGPU in High Sierra on the 2011 which are more apparent when "Reduce transparency" is disabled and occur with both AMD and NVIDIA eGPUs. There are no glitches in Mojave with the RX 460 on the newer machine.

Update — kryptonite is the culprit. No issues with the older purge-wrangler script on the 2011 MBP.

 

[TheShortTimer]

Assuming the ancient dual-core CPU isn’t bottlenecking the GPU and VR setup too badly, I don’t see why not.

I came across this interesting webpage by someone at UC San Francisco that covers various tests with Macs and VR and was taken aback by the following sentence:

Performance. I chose an AMD graphics card because Apple does not support Nvidia graphics cards.

Unless I've missed something here, Apple do support Nvidia's products because how else could so many of their cards automatically work in macOS without the need to install drivers?

 

[Amethyst1]

Unless I've missed something here, Apple do support Nvidia's products because how else could so many of their cards automatically work in macOS without the need to install drivers?

macOS' built-in drivers only work with "Kepler" [and older] GPUs. "Maxwell" and "Pascal" GPUs need Nvidia's web drivers which only run in High Sierra and older versions. Thus, "Maxwell" and "Pascal" don't work in Mojave and newer versions. Newer Nvidia GPUs don't work at all in macOS.

 

[TheShortTimer]

Newer Nvidia GPUs don't work at all in macOS.

I was about to ask why but the answers were out there for me to find with a quick web search.

 

[Amethyst1]

I wonder if the 2011's CPU or High Sierra is to blame for the 460's abysmal performance in Unigine.

It's not High Sierra's fault, since the 460's performance is as expected in High Sierra on the 2013 MBP.

I was about to ask why

...because AMD rul3z

I did some more benchmarks with the 460 on the 2013 MBP today (added to post #31), demonstrating the performance hit when accelerating the internal display is reduced to 16≈17% in Unigine over Thunderbolt 2.
When doing the reverse, i.e. running stuff on the iGPU and displaying it via the eGPU (LOL!), the hit is reduced to 5≈10%.
The Iris Pro 5200 was the first decent Intel iGPU and is quite capable of providing 30 fps if you keep the settings reasonable low.
The temperature difference (as reported by Heaven and Valley) is even more impressive: 45°C using the eGPU and an external display, around 50°C using the eGPU to accelerate the internal display, 74°C using the iGPU and the internal display. Using an eGPU is a very reasonable way to keep the MBP cooler... and, more importantly, quiet when doing graphically intense stuff.

 

[TheShortTimer]

My turn!

Its taken me a while to get everything sorted out but better late than never.

Initially I had Catalina running on the MBA and needed to revert to High Sierra because thanks to a heads up from @Amethyst1 - I was made aware that for the meantime, Kryptonite isn't compatible with Mojave and its successors. Hopefully this situation will change soon.

Setting up on OpenCore volume on the MBA's SSD was much easier than I expected, although it would have been helpful if the author of this guide had explained that you don't have to worry about labelling the volume because Kryptonite will rename it anyway.

Also this was unnecessary and caused confusion for me:

tmutil deletelocalsnapshots /

...because I'm not even running Time Machine in the first place - as I eventually realised.

Anyhow, once the Nvidia web drivers were downloaded, the computer restarted and I pressed Ctrl + Enter to assign the OpenCore volume as the default booting drive.

So far, so good!

High Sierra has reached the log-in screen: it's time to connect the eGPU enclosure to the MBA's Thunderbolt port, log-in and see what happens.

Everything appears to be recognised with a dual display mode: the internal GPU handles the MBA's LCD whilst the NVS 510 handles the output to my TV. Let's delve a bit further.

It's a 50 inch TV not 31.5 but at least the manufacturer has been recognised correctly. Pedantry aside, let's continue.

As yet I haven't run any benchmark tests but here's my observations so far with the NVS 510:

*Finder windows suffering from glitching and tearing when they're dragged around and only correct themselves after you've clicked elsewhere and then returned to them. Can this be rectified?

*1440p windowed on YouTube works fairly well. Full screen playback causes severe lagging. 4K mode is just too much.

*On the subject of Ultra HD, 4K H264/265 MP4 videos actually play somewhat in windowed mode (unlike under the internal HD 3000, which cannot cope with them at all) but with extremely noticeable lagging, sluggishness and sometimes, tearing. I suspect that the MBA's 2011 1.6 Ghz dual core i5 falls slightly short of the horsepower necessary to provide fluid playback. Feedback on this would be appreciated. I'll have to see whether the 2.3 Ghz dual core i5 on my 2011 MBP can pull it off - I'd wager that it can.

*The 4K videos saw a slight improvement (with tearing issues) by switching to the NVS 510 as the sole GPU for the MBA's LCD and my TV, which was curious.

Moving onto the other two GPU's, the 7770 briefly powers up with a spin from the fan when connected at the log-in screen and then causes the MBA to lock up and kernel panic:

I've rebooted several times and reconnected the Thunderbolt at the log-in screen but the outcome is always the same.

The GT 630 provides a dual display with a small area of garbage in the top corner of the TV. You can see the mouse pointer and move it around the TV but beyond that the display is completely black.

 

[Amethyst1]

Thanks a lot for sharing your experience

Initially I had Catalina running on the MBA and needed to revert to High Sierra because thanks to a heads up from @Amethyst1 - I was made aware that for the meantime, Kryptonite isn't compatible with Mojave and its successors.

This only applies to “unsupported” installations of Mojave and up. kryptonite is compatible with Mojave and up on officially supported machines.

Anyhow, once the Nvidia web drivers were downloaded,

You don’t actually need them for Kepler cards.

Finder windows suffering from glitching and tearing when they're dragged around and only correct themselves after you've clicked elsewhere and then returned to them.

Same here on the 2011 MBP; no such issues on the 2013 (Haswell) MBP — neither on High Sierra nor on Mojave.

The 4K videos saw a slight improvement (with tearing issues) by switching to the NVS 510 as the sole GPU for the MBA's LCD and my TV, which was curious.

The MBA’s LCD is still driven by the HD 3000, despite About This Mac saying otherwise.

I can do some UHD video testing with the GTX 660 Ti and RX 460. Do you have a link to a freely available video I could use as a “benchmark?”

 

[TheShortTimer]

Thanks a lot for sharing your experience

One good turn deserves another.

This only applies to “unsupported” installations of Mojave and up. Kryptonite is compatible with Mojave and up on officially supported machines.

That means I could use Kryptonite on my 2012 MBP with Mojave and beyond.

You don’t actually need them for Kepler cards.

Understood.

Same here on the 2011 MBP;

Interesting - does this mean that there's something about the 2011 hardware which is causing this issue?

no such issues on the 2013 (Haswell) MBP — neither on High Sierra nor on Mojave.

I'll have to upgrade my 2012 MBP to High Sierra or Mojave and check whether it applies to that machine as well.

The MBA’s LCD is still driven by the HD 3000, despite About This Mac saying otherwise.

I won't pay attention to that information in future.

I can do some UHD video testing with the GTX 660 Ti and RX 460. Do you have a link to a freely available video I could use as a “benchmark?”

Yes, the Demolandia website has loads of (legally) free to download UHD content - I chose this one at random. It played with sound on VLC but the frames quickly broke up and froze for the most part.

 

[Amethyst1]

That means I could use Kryptonite on my 2012 MBP with Mojave and beyond.

Yep.

Interesting - does this mean that there's something about the 2011 hardware which is causing this issue?

Maybe Thunderbolt 1 plays a role in it too; I'll test on the 2012 13" retina MBP I recently got which has an Ivy Bridge dual-core and Thunderbolt 1.

I won't pay attention to that information in future.

About This Mac > Overview always reports the GPU which is driving the main display, but on my 2015 iMac, About This Mac > Displays correctly reports the iGPU driving its internal display and the eGPU driving the external(s). That's on Mojave though. I also get this nifty icon in the menu bar on both High Sierra and Mojave with the RX 460, but not with an NVIDIA eGPU, which allows gracefully disconnecting the eGPU. It's worth noting that High Sierra calls the RX 460 "AMD Radeon Pro 460" there, but it's identified correctly everywhere else.

 

[TheShortTimer]

Maybe Thunderbolt 1 plays a role in it too; I'll test on the 2012 13" retina MBP I recently got which has an Ivy Bridge dual-core and Thunderbolt 1.

It goes without saying that I'll be interested in your findings.

After a bit more tinkering, I've discovered that the glitching extends to dragging almost anything around the external display: be it Finder windows or applications such as Firefox, VLC or QuickTime - they all leave a "trail" behind them. Drive icons seem to be (more or less) ok though. Also the external desktop doesn't update or refresh correctly.

I've managed to solve this by switching back from a mirrored configuration to separate displays: the glitching no longer occurs whilst dragging items around the external display area.

EDIT: this proved to be a short-lived solution because the glitching has persisted in dual display mode.

I also noticed that the gamma levels for the NVS 510 at their standard setting were too low and they had to be cranked up quite a bit in order to reach what I would consider to be normal. I'll probably put the GT 630 into my Mac Pro 1,1 and see if I can get the 7770 to work with Kryptonite and the MBA - I should also test it with the 2012 MBP and see whether a kernel panic occurs on that machine as well.

On a side note, unfortunately there is no audio pass through to the eGPU so I'll have to connect the audio jack directly to the TV.

 

[Amethyst1]

After a bit more tinkering, I've discovered that the glitching extends to dragging almost anything around the external display: be it Finder windows or applications such as Firefox, VLC or QuickTime - they all leave a "trail" behind them. Drive icons seem to be (more or less) ok though. Also the external desktop doesn't update or refresh correctly.

That matches my experience. It might also explain why the RX 460 performs abysmally (worse than the NVS 510) on the 2011 MBP.

I should also test it with the 2012 MBP and see whether a kernel panic occurs on that machine as well.

What's the actual panic message?

On a side note, unfortunately there is no audio pass through to the eGPU so I'll have to connect the audio jack directly to the TV.

This works for me with the RX 460 on Mojave (via DisplayPort); I'll do some tests with the NVS 510 and on High Sierra. Maybe your DisplayPort-to-HDMI adapter is to blame for that?

 

[Amethyst1]

It goes without saying that I'll be interested in your findings.

So I've done some initial testing on the Late 2012 13" Retina MacBook Pro with a dual-core Ivy Bridge. No glitches with the RX 460 on High Sierra and performance in Heaven and Valley is OK. Heaven's score (1518) is just 2.7% lower than on the quad-core Haswell, but Valley's (1705) is 30.7% lower. Audio via DisplayPort works fine with the 460.

 

[TheShortTimer]

That matches my experience. It might also explain why the RX 460 performs abysmally (worse than the NVS 510) on the 2011 MBP.

On the subject of perfomance, I have some further updates: since upgrading both the 2011 MBA and MBP to High Sierra, I've played a couple of 4K videos on both machines in full screen with QuickTime using their internal HD 3000 GPU's. Both videos are encoded in H.264 with a resolution of 3840 x 2160. Neither machine can cope with Video #1 but Video #2 plays perfectly on the MBP and near flawlessly on the MBA - with just a few seconds of lagging at the beginning when viewed on the latter.

The NVS 510 is unable to play either of the videos - even on the MBP which has far greater horsepower than the MBA. This is insane: the technically superior GPU is outperformed by its inferior counterpart.

What's the actual panic message?

Here you go - in its entirety:

Anonymous UUID: D75C2BD7-AEE9-924A-641E-F128285BDE60

Mon Dec 6 09:31:01 2021

*** Panic Report ***
panic(cpu 0 caller 0xffffff7fad231577): "displayModeDidChange failed: set_display_mode_and_vram failed!
"@/BuildRoot/Library/Caches/com.apple.xbs/Sources/GPUDriversAMD/GPUDriversAMD-1.68.25/Common/GLKernel/ATI/Accel/AMDAccelDisplayMachine.cpp:321
Backtrace (CPU 0), Frame : Return Address
0xffffff80867c3650 : 0xffffff8029e6ae76
0xffffff80867c36a0 : 0xffffff8029f94484
0xffffff80867c36e0 : 0xffffff8029f86654
0xffffff80867c3750 : 0xffffff8029e1ce60
0xffffff80867c3770 : 0xffffff8029e6a8ec
0xffffff80867c38a0 : 0xffffff8029e6a6ac
0xffffff80867c3900 : 0xffffff7fad231577
0xffffff80867c3a30 : 0xffffff7fad19c478
0xffffff80867c3a60 : 0xffffff7fad1c93b9
0xffffff80867c3aa0 : 0xffffff7faae119ae
0xffffff80867c3b50 : 0xffffff7faadfe83c
0xffffff80867c3b90 : 0xffffff7faae08e33
0xffffff80867c3bb0 : 0xffffff7faae0d03b
0xffffff80867c3c30 : 0xffffff7fade33c08
0xffffff80867c3c70 : 0xffffff7faae040c2
0xffffff80867c3cc0 : 0xffffff802a46bbdf
0xffffff80867c3d10 : 0xffffff802a4c86a7
0xffffff80867c3d70 : 0xffffff8029f41292
0xffffff80867c3dc0 : 0xffffff8029e70380
0xffffff80867c3e10 : 0xffffff8029e4d6ed
0xffffff80867c3e60 : 0xffffff8029e601fb
0xffffff80867c3ef0 : 0xffffff8029f7033d
0xffffff80867c3fa0 : 0xffffff8029e1d666
Kernel Extensions in backtrace:
com.apple.iokit.IOGraphicsFamily(519.23)[81D774E8-DD11-3866-B90A-E855215EDB60]@0xffffff7faadf5000->0xffffff7faae3dfff
dependency: com.apple.iokit.IOPCIFamily(2.9)[F763C275-306F-358F-B6B0-0FD93486BD5F]@0xffffff7faa694000
com.apple.iokit.IOAcceleratorFamily2(378.28)[D342FE41-0457-3586-ABB0-38E8E3947C4F]@0xffffff7fad191000->0xffffff7fad228fff
dependency: com.apple.driver.AppleMobileFileIntegrity(1.0.5)[A4F333D5-7EC0-332F-A23B-558A0AC169A8]@0xffffff7faadb5000
dependency: com.apple.iokit.IOSurface(211.15)[37BCC686-AF99-3381-AC4F-E54A347E0F40]@0xffffff7facab3000
dependency: com.apple.iokit.IOPCIFamily(2.9)[F763C275-306F-358F-B6B0-0FD93486BD5F]@0xffffff7faa694000
dependency: com.apple.iokit.IOGraphicsFamily(519.23)[81D774E8-DD11-3866-B90A-E855215EDB60]@0xffffff7faadf5000
com.apple.kext.AMDRadeonX4000(1.6.8)[9C20D204-D0FC-3A0D-B96E-25FA33F9798F]@0xffffff7fad229000->0xffffff7fad6e3fff
dependency: com.apple.iokit.IOSurface(211.15)[37BCC686-AF99-3381-AC4F-E54A347E0F40]@0xffffff7facab3000
dependency: com.apple.iokit.IOPCIFamily(2.9)[F763C275-306F-358F-B6B0-0FD93486BD5F]@0xffffff7faa694000
dependency: com.apple.iokit.IOGraphicsFamily(519.23)[81D774E8-DD11-3866-B90A-E855215EDB60]@0xffffff7faadf5000
dependency: com.apple.iokit.IOAcceleratorFamily2(378.28)[D342FE41-0457-3586-ABB0-38E8E3947C4F]@0xffffff7fad191000
com.apple.kext.AMDFramebuffer(1.6.8)[CFBAFADB-04FF-3788-9CC9-F1618EEE6579]@0xffffff7fade32000->0xffffff7fade6dfff
dependency: com.apple.iokit.IOACPIFamily(1.4)[95DA39BB-7C39-3742-A2E5-86C555E21D67]@0xffffff7faab5b000
dependency: com.apple.iokit.IOPCIFamily(2.9)[F763C275-306F-358F-B6B0-0FD93486BD5F]@0xffffff7faa694000
dependency: com.apple.kext.AMDSupport(1.6.8)[ECE58EB7-A85B-32C8-9BDB-1D8325E7EA0E]@0xffffff7facf34000
dependency: com.apple.iokit.IOGraphicsFamily(519.23)[81D774E8-DD11-3866-B90A-E855215EDB60]@0xffffff7faadf5000

BSD process name corresponding to current thread: WindowServer
Boot args: -lilubeta -krybeta krygpu=NVDA krytbtv=1

Mac OS version:
17G14042

Kernel version:
Darwin Kernel Version 17.7.0: Fri Oct 30 13:34:27 PDT 2020; root:xnu-4570.71.82.8~1/RELEASE_X86_64
Kernel UUID: 5E83A13A-32F5-3604-8591-50E2F2F70DC6
Kernel slide: 0x0000000029c00000
Kernel text base: 0xffffff8029e00000
__HIB text base: 0xffffff8029d00000
System model name: MacBookAir4,1 (Mac-C08A6BB70A942AC2)

System uptime in nanoseconds: 51632835380
last loaded kext at 40511133296: com.apple.kext.AMDFramebuffer 1.6.8 (addr 0xffffff7fade32000, size 245760)
loaded kexts:
io.egpu.Kryptonite 1.0.1
as.vit9696.Lilu 1.5.6
com.apple.kext.AMDFramebuffer 1.6.8
com.apple.kext.AMDRadeonX4000 1.6.8
com.apple.kext.AMD7000Controller 1.6.8
com.apple.filesystems.autofs 3.0
com.apple.driver.AudioAUUC 1.70
com.apple.filesystems.msdosfs 1.10
com.apple.driver.AppleHWSensor 1.9.5d0
com.apple.driver.AppleTyMCEDriver 1.0.3d2
com.apple.driver.ApplePlatformEnabler 2.7.0d0
com.apple.driver.AGPM 110.23.37
com.apple.driver.AppleMikeyHIDDriver 131
com.apple.driver.AppleGraphicsDevicePolicy 3.20.18
com.apple.AGDCPluginDisplayMetrics 3.20.18
com.apple.driver.AppleHV 1
com.apple.iokit.IOUserEthernet 1.0.1
com.apple.driver.AppleHDAHardwareConfigDriver 281.52
com.apple.driver.AppleUpstreamUserClient 3.6.5
com.apple.driver.ApplePolicyControl 3.20.18
com.apple.driver.AppleHDA 281.52
com.apple.driver.AppleMikeyDriver 281.52
com.apple.iokit.IOBluetoothSerialManager 6.0.7f22
com.apple.driver.pmtelemetry 1
com.apple.driver.AGDCBacklightControl 3.20.18
com.apple.Dont_Steal_Mac_OS_X 7.0.0
com.apple.driver.AppleIntelHD3000Graphics 10.0.4
com.apple.driver.AppleBacklight 170.12.6
com.apple.driver.eficheck 1
com.apple.driver.AppleThunderboltIP 3.1.1
com.apple.driver.AppleLPC 3.1
com.apple.driver.AppleSMCPDRC 1.0.0
com.apple.driver.AppleOSXWatchdog 1
com.apple.driver.AppleSMCLMU 211
com.apple.driver.ACPI_SMC_PlatformPlugin 1.0.0
com.apple.driver.AppleFIVRDriver 4.1.0
com.apple.driver.AppleIntelSlowAdaptiveClocking 4.0.0
com.apple.driver.AppleMCCSControl 1.5.5
com.apple.driver.AppleIntelSNBGraphicsFB 10.0.4
com.apple.iokit.IOBluetoothUSBDFU 6.0.7f22
com.apple.iokit.SCSITaskUserClient 404.30.3
com.apple.driver.AppleUSBTCKeyEventDriver 254
com.apple.driver.AppleUSBTCButtons 254
com.apple.driver.AppleUSBTCKeyboard 254
com.apple.driver.AppleFileSystemDriver 3.0.1
com.apple.filesystems.hfs.kext 407.50.6
com.apple.AppleFSCompression.AppleFSCompressionTypeDataless 1.0.0d1
com.apple.BootCache 40
com.apple.AppleFSCompression.AppleFSCompressionTypeZlib 1.0.0
com.apple.AppleSystemPolicy 1.0
com.apple.filesystems.apfs 748.51.0
com.apple.driver.AirPort.Brcm4331 800.21.30
com.apple.driver.AirPort.Brcm4360 1250.20.1a4
com.apple.driver.AppleAHCIPort 329.50.2
com.apple.driver.AppleSmartBatteryManager 161.0.0
com.apple.driver.AppleRTC 2.0
com.apple.driver.AppleACPIButtons 6.1
com.apple.driver.AppleHPET 1.8
com.apple.driver.AppleSMBIOS 2.1
com.apple.driver.AppleACPIEC 6.1
com.apple.driver.AppleAPIC 1.7
com.apple.driver.AppleIntelCPUPowerManagementClient 220.50.1
com.apple.nke.applicationfirewall 186
com.apple.security.TMSafetyNet 8
com.apple.security.quarantine 3
com.apple.driver.AppleIntelCPUPowerManagement 220.50.1
com.apple.kext.AMDRadeonX4030HWLibs 1.0
com.apple.iokit.IOAcceleratorFamily2 378.28
com.apple.kext.AMDRadeonX4000HWServices 1.6.8
com.apple.kext.AMDSupport 1.6.8
com.apple.driver.AppleThunderboltPCIUpAdapter 2.1.3
com.apple.driver.AppleHPM 3.1.3
com.apple.driver.AppleIntelLpssI2CController 3.0.60
com.apple.driver.AppleIntelLpssDmac 3.0.60
com.apple.driver.AppleIntelLpssI2C 3.0.60
com.apple.kext.triggers 1.0
com.apple.iokit.IOAVBFamily 683.1
com.apple.plugin.IOgPTPPlugin 680.15
com.apple.iokit.IOEthernetAVBController 1.1.0
com.apple.driver.AppleSSE 1.0
com.apple.driver.AppleGraphicsControl 3.20.18
com.apple.driver.DspFuncLib 281.52
com.apple.kext.OSvKernDSPLib 526
com.apple.iokit.IOSurface 211.15
com.apple.iokit.IOSerialFamily 11
com.apple.driver.AppleBacklightExpert 1.1.0
com.apple.iokit.IONDRVSupport 519.21
com.apple.AppleGPUWrangler 3.20.18
com.apple.AppleGraphicsDeviceControl 3.20.18
com.apple.driver.AppleHDAController 281.52
com.apple.iokit.IOHDAFamily 281.52
com.apple.iokit.IOAudioFamily 206.5
com.apple.vecLib.kext 1.2.0
com.apple.driver.IOPlatformPluginLegacy 1.0.0
com.apple.driver.IOPlatformPluginFamily 6.0.0d8
com.apple.iokit.IOSlowAdaptiveClockingFamily 1.0.0
com.apple.driver.AppleThunderboltEDMSink 4.1.4
com.apple.driver.AppleThunderboltDPOutAdapter 5.5.5
com.apple.driver.AppleSMBusController 1.0.18d1
com.apple.driver.AppleSMBusPCI 1.0.14d1
com.apple.iokit.IOGraphicsFamily 519.23
com.apple.iokit.BroadcomBluetoothHostControllerUSBTransport 6.0.7f22
com.apple.iokit.IOBluetoothHostControllerUSBTransport 6.0.7f22
com.apple.iokit.IOBluetoothHostControllerTransport 6.0.7f22
com.apple.iokit.IOBluetoothFamily 6.0.7f22
com.apple.driver.AppleUSBMultitouch 262
com.apple.driver.usb.IOUSBHostHIDDevice 1.2
com.apple.driver.usb.cdc 5.0.0
com.apple.driver.usb.networking 5.0.0
com.apple.driver.usb.AppleUSBHostCompositeDevice 1.2
com.apple.driver.usb.AppleUSBHub 1.2
com.apple.filesystems.hfs.encodings.kext 1
com.apple.driver.AppleThunderboltDPInAdapter 5.5.5
com.apple.driver.AppleThunderboltDPAdapterFamily 5.5.5
com.apple.driver.AppleThunderboltPCIDownAdapter 2.1.3
com.apple.driver.AppleXsanScheme 3
com.apple.iokit.IOAHCIBlockStorage 301.40.2
com.apple.driver.AppleThunderboltNHI 4.7.3
com.apple.iokit.IOThunderboltFamily 6.7.10
com.apple.driver.AppleUSBMergeNub 900.4.1
com.apple.iokit.IO80211Family 1200.12.2
com.apple.driver.mDNSOffloadUserClient 1.0.1b8
com.apple.driver.corecapture 1.0.4
com.apple.iokit.IOAHCIFamily 288
com.apple.driver.usb.AppleUSBEHCIPCI 1.2
com.apple.driver.usb.AppleUSBHostPacketFilter 1.0
com.apple.iokit.IOUSBFamily 900.4.1
com.apple.driver.usb.AppleUSBUHCIPCI 1.2
com.apple.driver.usb.AppleUSBUHCI 1.2
com.apple.driver.usb.AppleUSBEHCI 1.2
com.apple.driver.AppleUSBHostMergeProperties 1.2
com.apple.driver.AppleEFINVRAM 2.1
com.apple.driver.AppleEFIRuntime 2.1
com.apple.iokit.IOSMBusFamily 1.1
com.apple.iokit.IOHIDFamily 2.0.0
com.apple.security.sandbox 300.0
com.apple.kext.AppleMatch 1.0.0d1
com.apple.driver.DiskImages 480.60.3
com.apple.driver.AppleFDEKeyStore 28.30
com.apple.driver.AppleEffaceableStorage 1.0
com.apple.driver.AppleKeyStore 2
com.apple.driver.AppleUSBTDM 439.70.3
com.apple.driver.AppleMobileFileIntegrity 1.0.5
com.apple.iokit.IOUSBMassStorageDriver 140.70.2
com.apple.iokit.IOSCSIBlockCommandsDevice 404.30.3
com.apple.iokit.IOSCSIArchitectureModelFamily 404.30.3
com.apple.iokit.IOStorageFamily 2.1
com.apple.driver.AppleCredentialManager 1.0
com.apple.driver.KernelRelayHost 1
com.apple.iokit.IOUSBHostFamily 1.2
com.apple.driver.usb.AppleUSBCommon 1.0
com.apple.driver.AppleBusPowerController 1.0
com.apple.driver.AppleSEPManager 1.0.1
com.apple.driver.IOSlaveProcessor 1
com.apple.iokit.IOReportFamily 31
com.apple.iokit.IOTimeSyncFamily 680.15
com.apple.iokit.IONetworkingFamily 3.4
com.apple.driver.AppleACPIPlatform 6.1
com.apple.driver.AppleSMC 3.1.9
com.apple.iokit.IOPCIFamily 2.9
com.apple.iokit.IOACPIFamily 1.4
com.apple.kec.pthread 1
com.apple.kec.Libm 1
com.apple.kec.corecrypto 1.0

EOF
Model: MacBookAir4,1, BootROM 135.0.0.0.0, 2 processors, Intel Core i5, 1.6 GHz, 2 GB, SMC 1.74f4
Graphics: Intel HD Graphics 3000, Intel HD Graphics 3000, Built-In
Graphics: NVIDIA NVS 510, NVIDIA NVS 510, PCIe
Memory Module: BANK 0/DIMM0, 1 GB, DDR3, 1333 MHz, 0x802C, 0x384A54463132383634485A2D314734473120
Memory Module: BANK 1/DIMM0, 1 GB, DDR3, 1333 MHz, 0x802C, 0x384A54463132383634485A2D314734473120
AirPort: spairport_wireless_card_type_airport_extreme (0x14E4, 0xE9), Broadcom BCM43xx 1.0 (5.106.98.102.30)
Bluetooth: Version 6.0.7f22, 3 services, 27 devices, 1 incoming serial ports
Network Service: Wi-Fi, AirPort, en0
PCI Card: NVIDIA NVS 510, Display Controller, Thunderbolt@132,0,0
PCI Card: pci10de,e1b, Audio Device, Thunderbolt@132,0,1
Serial ATA Device: APPLE SSD TS128C, 121.33 GB
USB Device: USB 2.0 Bus
USB Device: Hub
USB Device: AS2115
USB Device: USB 2.0 Bus
USB Device: FaceTime Camera (Built-in)
USB Device: Hub
USB Device: Apple Internal Keyboard / Trackpad
USB Device: BRCM20702 Hub
USB Device: Bluetooth USB Host Controller
Thunderbolt Bus: MacBook Air, Apple Inc., 8.1
Thunderbolt Device: eGFX Breakaway Box 550, Sonnet Technologies, Inc., 1, 25.1

The 7770 causes my 2011 MBP to swiftly lock up at the log-in screen (no matter how fast you attempt to type the password before this happens!) and then reboot - sans kernel panic screen.

This works for me with the RX 460 on Mojave (via DisplayPort); I'll do some tests with the NVS 510 and on High Sierra. Maybe your DisplayPort-to-HDMI adapter is to blame for that?

It's a fairly reputable brand (Belkin) and it definitely provides audio directly to the TV from the Thunderbolt ports on my 2011 MBP and MBA, 2012 MBP and the DisplayPort on my 2010 MBA. There is another adapter that I can try instead though.

So I've done some initial testing on the Late 2012 13" Retina MacBook Pro with a dual-core Ivy Bridge. No glitches with the RX 460 on High Sierra and performance in Heaven and Valley is OK. Heaven's score (1518) is just 2.7% lower than on the quad-core Haswell, but Valley's (1705) is 30.7% lower. Audio via DisplayPort works fine with the 460.

Thanks for that. I set up Kryptonite on my 2012 15" non-Retina MBP and the installer warned me that the GT 650M would have to be disabled if I wanted to use an external display. I opted to retain it and use the LCD instead. The 7770 didn't cause a lockup and kernal panic but it didn't appear to be available as an option.

Your experience with successful audio points towards a potential error with my configuration. Please let me know whether you can achieve an audio passthrough with your NVS 510. I might need to look at an additional card because presently the 510 is my only functioning one (within the eGPU set up). Expensive hobby!

 

[Amethyst1]

*Finder windows suffering from glitching and tearing when they're dragged around and only correct themselves after you've clicked elsewhere and then returned to them. Can this be rectified?

YES

I just tried the older purge-wrangler script which has a different approach and a couple of shortcomings compared to kryptonite, but eGPUs finally work properly in High Sierra on the 2011 MBP: no glitches, proper performance in Unigine! I feel like a total idiot now... because I'm also using purge-wrangler on my main system (kryptonite didn't exist when I set it up)!

 

[TheShortTimer]

YES

I just tried the older purge-wrangler script which has a different approach and a couple of shortcomings compared to Kryptonite...

What are the shortcomings?

…but eGPUs finally work properly in High Sierra on the 2011 MBP: no glitches, proper performance in Unigine! I feel like a total idiot now... because I'm also using purge-wrangler on my main system (Kryptonite didn't exist when I set it up)!

Should I erase the Kryptonite partitions and try PurgeWangler instead? Do I need to erase them in order to use it?

 

[Amethyst1]

What are the shortcomings?

See here.

Should I erase the Kryptonite partitions and try PurgeWangler instead? Do I need to erase them in order to use it?

That's not necessary but you need to boot "normally" to use an eGPU after running purge-wrangler.

 

[TheShortTimer]

I reached as far as Stage 6 on my MBA:

If running macOS 11 or newer, you must also disable ART:
csrutil authenticated-root disable

This fails with the following error message:

csrutil: invalid command authenticated-root

Google is my friend but I find any leads on how to resolve this. I tried to use the command from Terminal in my El Capitan USB installer but I received the same error message.

 

[Amethyst1]

Are you on macOS 11? I thought you are on High Sierra. That command isn’t required for it.
I've updated the benchmark results for the RX 460 in post #24 now that I'm using purge-wrangler. Apart from Unigine, they're pretty much the same.

 

[Amethyst1]

Yes, the Demolandia website has loads of (legally) free to download UHD content - I chose this one at random. It played with sound on VLC but the frames quickly broke up and froze for the most part.

I just tried this exact video and it plays virtually flawlessly using the RX 460 and VLC, with just a couple dropped frames right at the beginning. CPU load is about 25%, GPU load around 20≈25%. This is a stark contrast to the unwatchable slideshow the HD 3000 subjects me to.

 

[TheShortTimer]

Are you on macOS 11? I thought you are on High Sierra. That command isn’t required for it.
I've updated the benchmark results for the RX 460 in post #24 now that I'm using purge-wrangler. Apart from Unigine, they're pretty much the same.

A classic case of (me) failing to pay full attention whilst reading. In my haste I misread it as macOS 10.11.

There's a major update on the way!

I just tried this exact video and it plays virtually flawlessly using the RX 460 and VLC, with just a couple dropped frames right at the beginning. CPU load is about 25%, GPU load around 20≈25%. This is a stark contrast to the unwatchable slideshow the HD 3000 subjects me to.

Great stuff!

I'll discuss this more in my next post - along with some developments...