euqlaog

@theitsage

Means an immediate license change to my script:

“No one is allowed to modify Apple’s copyrighted binary files.”

Plain text file editing sounds okay, but as to reverse engineering and binary modding, I am out of the equation. It would be endless patching and battling against Apple’s security updates, and Intel would be hunting bulls and bisons. DIYing is fun when the method shows signs of continuity. This a dead end.

There is a brighter side of the equation.

You're right, there is uncertainty regarding which TB3 enclosures are TB3 Mac compatible. Another aspect is AMD stability. PowerColor's Devil Box has been on the market for some time, is there any info on this?

1 hour ago, ikir said:

Any luck on macOS side?

Yes,

https://www.techinferno.com/index.php?/forums/topic/10718-2016-macbook-pros-and-egpus/&do=findComment&comment=152495

Metal acceleration does work with the RX 480, OpenCL doesn’t.

RX 460 provides full support:

https://www.techinferno.com/index.php?/forums/topic/10845-2015-13-mba-rx46016gbps-tb2-akitio-thunder2-108w-adaptor-osx-102-hedge/

There is a binary hack for uncertified TB3 enclosures on Netkas forum. Hopper Disassembler reveals many things, but I would not publish reverse engineered information that violates Apple’s SLA section [M].

21 hours ago, Morv said:

@goalque So it does work like Nvidia Optimus but with way better performance on the internal display? Which driver? Simply the current AMD driver or a specific driver?

21 hours ago, rwssoccer1 said:

Nice! So was this without any modifications to the PCIE...etc...

I also ran the Time Spy and 3DMark13, yielding ~94% and ~93% of the external monitor performance. Requires a workaround (hint: it’s one of the @Tech Inferno Fan’s suggested steps). Windows 10 downloads the AMD driver automatically, you can uninstall it in device manager and update the driver software to the latest by selecting a new one from the extracted package. I haven’t seen anyone else reporting “AMD XConnect Lite Technology” on Mac computer. This applies to iGPU only MBP. I suppose the dGPU model needs more configurations.

EDIT: it's renamed to "AMD XConnect Technology" with the latest drivers.

Awesome. Internal screen performance is ~96% of the external in 3DMark11. The total score was even better than the external due to physics.

http://www.3dmark.com/3dm11/11753719

This was through the TB3-TB2 adapter. It says “AMD XConnect Lite Technology” in the system tray.

1 hour ago, MacGamer said:

I may take the gamble. Worse case is that it fails and I can sell the GPU on eBay haha. 

 

Does the extended internal screen only work on AMD, and have you heard anything about performance? Thinking of a GTX 1070 (if it's possible) to drive games at 2880x1800 or a comparable AMD. 

It’s more meaningful to use an external monitor to achieve maximum performance. Haven’t heard about internal screen performance with an AMD dGPU. One German review sheds some light on this:

https://www.computerbase.de/2016-10/powercolor-devil-box-test/2/

XConnect technology may require a specific dGPU.

If doesn’t work, there are ready-made headless adapters if you want to try @LukeHero's method:

http://www.fit-pc.com/web/products/fit-headless/

http://www.newertech.com/products/hdmi_headless_video_adapter.php

18 minutes ago, Morv said:

 

Isn't there a chance it's going to work with an AMD GPU on the internal display due to XConnect? I don't have a AMD GPU to test it but the functionality of XConnect should be the same as Optimus by now, shouldn't it?

There might be a chance, but I’ve not found a switch yet. However, some hotfix enabled hot-plugging and surprise removal on MBP11,5 via TB2 (tried once, not sure about the current status).

"AMD Radeon Software Crimson Edition Hotfix drivers are compatible with AMD Discrete graphics products."

http://support.amd.com/en-us/download/desktop?os=Windows+10+-+64

AMD also provides BootCamp driver for MacBook Pro (2015+), but it’s rather old (release date 4/5/2016):

http://support.amd.com/en-us/download/desktop/bootcamp

1 hour ago, MacGamer said:

I'm very interested in playing on the Mac's internal display, but all the 2016 15" models have an AMD dGPU. Would there be better chances at getting a Radeon to work with the internal display, or am I out of luck?

If you're going to choose AMD eGPU, you can drag a windowed app to the extended internal screen, and its contents will be accelerated by the eGPU.

If someone has forgotten:

Quote

I promised Optimus automation for a MBP11,5, and it is provided as “Disable dGPU ON/OFF” option. You will see black screen about 10 seconds, but the internal display is activated as soon as Windows has loaded.

This functionality is included in the demo.

As @Tech Inferno Fan noted, I'm confident that MBP13,3 behaves similarly as the MBP11,5 (M370X dGPU) regarding gmux, and HD530 can be activated manually as well for Nvidia Optimus. Someone just has to try.

On 11/12/2016 at 7:45 PM, fakebanana said:

I've also had more progress with 10.12.2 beta 2 - goalque's EFI utility now detects graphics cards plugged into the Razer Core. Unfortunately, that's as far as I can get (choosing to "Load OS X", or enabling "Warm restart" returns a "Not Supported" error - maybe this is a limitation of the demo?). automate-eGPU.sh still doesn't detect anything.

Sorry for the late reply. Yes, it’s a limitation of the demo. It seems that people are reporting about an error code 12 issue, especially with Maxwell series cards connected to a MBP13,1.

I can confirm that my EFI app resolves this on Windows 10. Unfortunately, I cannot confirm Razer Core’s functionality and everything about this project is confidential. Here’s a couple of public 3DMark11 (still my favourite) benchmarks, with an Apple TB3 (USB-C) to TB2 adapter:

RX 480: http://www.3dmark.com/3dm11/11748177

GTX 980: http://www.3dmark.com/3dm11/11751133

Stay tuned.

@Miller

I start to feel like a 24-hour advisory service. Works with -a mode, the external monitor set as primary. Avoid 750Ti without a 6pin. No more questions please. Please read through the thread. I’ve finished this, other developers can continue.

Did you set the external monitor as primary after login?

I wrote a small script for you. Reinstall macOS 10.12.1 (disk name must be "Macintosh HD", SIP disabled, FileVault disabled). Give execute permission "chmod +x", run with sudo,
and restart (eGPU monitor disconnected).

#!/bin/sh
personalities=("ATI\ Support" Controller AMDBaffinGraphicsAccelerator)
kexts=(AMDSupport AMD9500Controller AMDRadeonX4100)
for (( i=0; i < 3; i++ ))
do
personality="Add :IOKitPersonalities:"${personalities[$i]}":IOPCITunnelCompatible bool true"
plist="/System/Library/Extensions/"${kexts[$i]}".kext/Contents/Info.plist"
/usr/libexec/PlistBuddy -c "$personality" "$plist" 2>/dev/null
done
rm /Volumes/Macintosh\ HD/System/Library/PrelinkedKernels/prelinkedkernel 2>/dev/null
rm /Volumes/Macintosh\ HD/System/Library/Caches/com.apple.kext.caches/Startup/kernelcache 2>/dev/null
touch /System/Library/Extensions
kextcache -update-volume /Volumes/Macintosh\ HD

If the startup progress bar proceeds slowly, even if the eGPU monitor is unplugged, there is something wrong in the kext cache.

If it goes through normally, but the UI on the eGPU screen is sluggish, AMDRadeonX4100.kext is probably not loaded. I am out of this script project, but I believe there are other programmers who can help.

2 hours ago, Hedge said:

It looks like the script doesn't yet account for the new AMDRadeonX4100.kext

 

Has anyone successfully used/modified goalque's auto-egpu script for RX 460 cards or would I be the first?

My script doesn’t support RX series cards. RX 460 is fully supported by the macOS 10.12.1+. You only need to add a couple of key-value pairs.

Remember to backup (rsync is more reliable than cp with recursive and preserve flags).

Finally, you have to update kext caches. This is the critical part.

If you choose the manual route, you will probably have to keep the eGPU monitor disconnected during the startup.

There are plenty of incorrect guides on the web - one suggesting "sudo touch /Extensions" command, which is literally the same "typo" as in my old implementation guide, two years ago. Nothing interesting in /Extensions folder since it does not exist. I suggest people learn what the commands do. The "man" is a good starting point.

3 hours ago, theitsage said:

@goalque  The “whoosh” is music to my ears! Thank you for your guidance. I tried running the metal test but got the following error:

<unknown>:0: error: no such file or directory: 'x86_64-apple-macosx10.12'

 

Hmm.. check that Xcode.app exists in /Applications folder and put "sudo" in front of the swiftc command:

sudo swiftc -sdk $(xcrun --show-sdk-path) -target x86_64-apple-macosx10.12 -o metaltest metaltest.swift

and you will need to type "agree" if it shows Xcode license agreement.

@theitsage

I guess you heard the “whoosh” sound

The text “Metal: Supported” is misleading - at least in previous builds.

The macOS 10.12 supports both v1 and v2 Metal feature sets. To confirm whether RX 470 is supported, follow these steps:

1) Install Xcode + command line tools: xcode-select --install

2) nano ~/Desktop/metaltest.swift

3) Copy-paste:

import Metal
var supportedDevices: [MTLDevice] = MTLCopyAllDevices()
if (supportedDevices.count > 0)
{
    for device in supportedDevices {
        var isSupported_v1 = device.supportsFeatureSet(MTLFeatureSet.osx_GPUFamily1_v1)
        var isSupported_v2 = device.supportsFeatureSet(MTLFeatureSet.osx_GPUFamily1_v2)
        print("\(device.name), supported (v1): \(isSupported_v1), supported (v2): \(isSupported_v2)")
    }
}
else
{
   print("No supported devices found")
}

4) ctrl+X -> Y
5) cd ~/Desktop

6)

swiftc -sdk $(xcrun --show-sdk-path) -target x86_64-apple-macosx10.12 -o metaltest metaltest.swift

7) ./metaltest

It’s worth to try IndigoBench when OpenCL works. Remember it’s only a half of the 32 compute units. HD 7970 keeps the lead in OpenCL due to optimized drivers.

I've ceased development of this script in favor of my EFI app. However, others are invited to continue its development on GitHub, for personal and non-commercial use only.

https://github.com/goalque/automate-eGPU/blob/master/SCRIPT-LICENSE.txt

If the copyright name list grows too long, you can reform its appearance by listing all the copyright holders in separate lines, for example:

* *** automate-eGPU.sh v1.0.0 ***
*
*  Copyright 2016 by My Name <[email protected]>
*  Copyright 2015, 2016 by Goalque <[email protected]>
*

I recommend encryption. It doesn’t prevent rip off, but makes it a bit more difficult. The encrypt/decrypt mechanism is pretty trivial. I am sure that there are programmers out there who know how pipelines and I/O redirections work. You should be able to generate new encrypted copyright notices easily in the Terminal.

Please do not release anything under the MIT (v0.9.8), it would be VOID under the Apple’s SLA:

http://images.apple.com/legal/sla/docs/macOS1012.pdf

1 hour ago, ayrat.m said:

Hello. I wanted to ask what you think about BizonBox3 device?

http://bizon-pc.com/ru/bizonbox3-egpu.html/

Previous version BizonBox 2 was based on AKiTiO Thunder2 PCIe Box

 

Akitio is a brand of inXtron, they are basically the same company. Looks like Bizon uses their TB3 board. Intel's certification requires complete TB products.

Akitio Node is already listed at:

http://www.span.com/product/AKiTiO-Thunderbolt-PCIe-Thunder3-Node-NODE-T3IA-AKTU-1x-TB3-USB-C-1x-PCI-Express-v3-0-x16-FHHL-x4-speed-75W-~60061

1 hour ago, theitsage said:

I love how there's no Late 2016 MacBook Pro in any of Bizon TB3 enclosure photos yet they claim full plug-n-play. The port layout in the rear looks identical to AKiTiO Thunder3 and Thunderbolt 3 dev kit. - http://www.inxtron.com/products/thunderbolt3-hdk

This HDK is not for resale and is not intended for the retail market.

I am sure that the internals are the same.

1 hour ago, rozhkov said:

They ignored my questions about "plug and play"

They use my script. Looks good though. +$350 extra price for the shiny enclosure without thermal certification or electrical safety label.

They don’t care about hardware certification or software licenses, unstoppable unless Apple/Intel lawyers come tapping on their shoulders.

There are too many factors to make final conclusions from CUDA-Z.

Razer Core needs a firmware update in order to be Late 2016 13” MBP compatible on macOS/OS X. This cannot be fixed by software.

Akitio claims only 6-8% difference compared to the motherboard:

https://www.akitio.com/faq/305-node-is-the-performance-identical-for-internal-and-external-gpus

And 2-6% drop compared to a 20Gbps cable:

https://www.akitio.com/faq/306-node-can-i-use-a-20-gbps-thunderbolt-3-cable

Sounds a contradiction. I guess the test was not CPU intensive.

A quad core CPU is still needed in CPU intensive gaming. A dual core i5-6200U (XPS 13) drops the frame rate by nearly 15fps compared to the i7-6700HQ in Tomb Raider benchmark:

https://www.computerbase.de/2016-10/powercolor-devil-box-test/4/

5 minutes ago, fakebanana said:

Tried today hooking up my 13" 2-port 2016 MBP via an active 40Gb/s TB3 cable - same results from system_profiler, "Up to 20 Gb/s x1". eGPU still not detected via EFI utility or anything else.

And says “Unsupported”? Then it’s a firmware issue. Akitio Node is compatible:

https://www.akitio.com/blog/articles/akitio-thunderbolt3-devices-macbook-pro

2 hours ago, theitsage said:

 

I can also confirm the two USB-C ports on the base Late 2016 MBP show 20 Gbps.

 

I've tried my RX 470 eGPU which has been working great with an 11" MBA and 17" MBP. Unfortunately there's no Acceleration or Metal Support when connected to TB3 MBP. All the supporting kexts loaded and external monitor work through the eGPU though. Any ideas?

Not alarming if the speed activates to 20Gb/s x2 with a Belkin 40Gb/s cable and TB3 chassis.

This should lead you on the right track:

https://www.techinferno.com/index.php?/forums/topic/10057-automate-egpu-efi-mac-bootscreen-on-egpu/&do=findComment&comment=151531

However, I suppose that you’ve already done so. Hopefully the problem is due to the bidirectional TB3-TB2-adapter. How did you check Metal acceleration? If OpenCL works, the -clpeak option may give us clues about transfer bandwidth in GBps (enqueueWriteBuffer/enqueueReadBuffer).

2 hours ago, Tech Inferno Fan said:

 

@goalque, where did you find that picture?  That shows 2x20Gbps 'physical' channels.

https://m.reddit.com/r/apple/comments/5a6z8s/my_experience_with_the_new_13_mbp_and_the_razer/?utm_source=mweb_redirect&compact=true

https://thunderbolttechnology.net/sites/default/files/Thunderbolt3_TechBrief_FINAL.pdf

It confirmed my suspicions.

2 hours ago, fakebanana said:

This is a 13" with 2 TB ports. It's supposed to have full Thunderbolt 3 performance on both ports.

The picture explains it:

20Gb/s x2.

There is something really wrong in Thunderbolt firmware handshake. Passive TB3 cables provide transfer data up to 20Gb/s. Would be intriguing to see if the speed is shown as 20Gb/s x2 with the Belkin 40Gb/s cable, when the TB3 device is correctly detected. It should work:

http://forums.macrumors.com/threads/base-13-macbook-pro-benchmarks-and-egpu.2010461/#post-23825958

My script is discontinued and may provide only a partial support with newer AMD cards.

Open up Terminal, and provide the output of the following command:

system_profiler SPThunderboltDataType | sed -e '/Port (Upstream)/,/Cable/!d'

No 40 Gb/s string, right?