[SCRIPT] PurgeWrangler: AMD + NVIDIA eGPUs for all Thunderbolt Macs
 
Notifications
Clear all

[Solved] [SCRIPT] PurgeWrangler: AMD + NVIDIA eGPUs for all Thunderbolt Macs  

 of  181
  RSS

Jeroenvankeulen
(@jeroen-van-keulen)
Trusted Member
Joined: 3 years ago
 
Posted by: Nikolai Damm

Jeroenvankeulen, just to be sure, is it working with your RX 580, or did you by chance test with NVIDIA?

No, I only have my RX580 to test.

MBP11,4, i7-4980HQ, 256gb SSD, 16gbRAM Cathalina (Sonnet Breakaway box 350 + Gigabyte Radeon RX 580 Gaming 8GB)
MacBook7,1 A1342, P8600 120gb SSD 4gbRAM. Mojave (Hiking/Camping)
iMac7,1 A1224, T9500 CPU upgrade 120gb SSD 4gbRAM. HS
Mac mini3,1 P7350, 250gb SSD 8gbRAM. HS (server) iPhone10,3 64gb. iOS12 iPad5,3, 16gb. iOS12 iPod shuffle 2nd gen. 2gb
✧✧

 
2015 15" MacBook Pro [4th,4C,H] + RX 580 @ 16Gbps-TB2>TB3 (Sonnet Breakaway 350) + macOS 10.14 [build link]  


ReplyQuote
Diego Ferreyra
(@diego_ferreyra)
New Member
Joined: 3 years ago
 

I got my rmbp 2015 working with a rx580, but performance seem kind of bad, I think it might be cause of the TB2 connection but didn't find anywhere to compare. 
in Heaven 1440p I get about 24fps 600point while in windows the average for this card seem to be 55fps 1300points.

also I thought 10.13.5 had support for internal retina display using eGpu but it seems it keeps using iris pro? do I still need to do the hack for internal display?

is there a way to use the eGpu card for video decoding? 

Ill appreciate any thoughts! thanks in advance!

To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.

.

ReplyQuote
StellarPatellar
(@stellarpatellar)
Active Member
Joined: 3 years ago
 

I've un/reinstalled the purge-wrangler script on my mac mini (2012) running 10.13.5 and the OS doesn't see the attached AMD RX 560 at all. The external exclosure connected by TB1 doesn't power on when connected to the mac mini. Purge-wrangler indicates the patch is detected, however, if the patch is present then the enclosure should power on even if the OS doesn't recognize the RX 560 (automate-eGPU would fix that at that point). Pretty much out of ideas at this point -- any suggested next steps would be appreciated.

To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.

.

ReplyQuote
jim_survak
(@jim_survak)
Eminent Member
Joined: 4 years ago
 
Posted by: Diego Ferreyra

no, I test it on a windows pc connect it directly to the motherboard...

Ah, well that is our problem then. On a typical motherboard you're connecting the graphics card to a dedicated set of PCIe lanes (typically 16, somewhat less common is to see 8). With Thunderbolt you have 4 PCIe lanes that are shared - so everything that runs through PCIe is shared on that one bus. So really what you are seeing (as well as me in my TB1 set up) is that all the data which is being processed has to be queued through the shared PCIe bus. Since that bus has half as many lanes (perhaps 1/4 as many) and has a capped data transfer rate of theoretically 20Gbps (PCIe x16 on a Gen3 connection is almost 128Gbps; PCIe x16 Gen2 is 64Gbps).

Hopefully that clears up the raw performance discrepancy.

As far as actual GPU usage, that one is a little bit trickier. I'm going to assume 2 things:
1) That this "xrg software" is accurate (personally I use iStat Menus & my eGPU shows 100% utilization in games).
2) It's a matter of driver optimization (there's simply far better graphics support on Windows than any other OS) and that DirectX can get closer to the metal than Apple's own Metal can. This is contingent on the fact that the program(s)/game(s) you're using are designed to use Metal. If not, then you're using OpenGL (or less likely Vulkan); neither of which renders the same performance as DirectX, despite the gap having been narrowed in recent years.
All of that is to say: if the software on macOS isn't specifically designed to utilize the GPU properly then you won't get the same GPU utilization out of it as on a Windows system (assuming the TB connection wasn't bottlenecking on the PCIe bandwidth in the first place).

Now, if you were booting to Windows on Boot Camp we'd have a much more comparable set of data as we'd only have to worry about software differences.

2018 Mac mini: Core i3, 8GB RAM, Sonnet 350 (aftermarket 550w PSU)+XFX Radeon RX-480 8GB Black Edition
2012 Mac mini: Core i7, 16GB RAM, Toshiba 1TB SSD, Seagate 1TB HDD
Custom: Ryzen 7 1700, 16GB Corasir DDR4-3200MHz RAM, 2x Corsair 500GB Neutron SSD, Seagate 3TB SSD, EVGA Nvidia 980Ti
YT:


ReplyQuote
mac_editor
(@mac_editor)
Famed Member Moderator
Joined: 4 years ago

ReplyQuote
Diego Ferreyra
(@diego_ferreyra)
New Member
Joined: 3 years ago
 

no, I test it on a windows pc connect it directly to the motherboard, still a 50% difference seems to much and its consistent among different resolutions and settings its always 50% of what is should be, try it 1440p, 1080p and 1080p with much lower settings. another thing I noticed its the gpu usage is really low. not sure about the accuracy of the xrg software but it also seems to decrease with the settings
1440p 30% usage and 50% of expected performance compared to windows pc.
1080p 24% usage and 50% of expected performance compared to windows pc.
1080p lower settings 19% usage and 50% of expected performance compared to windows pc.

To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.

.

ReplyQuote
Guru Nightmare
(@gurunightmare)
Active Member
Joined: 4 years ago
 

Great on MOJAVE Beta + Mac Mini 2014 (late) + Akitio Node + Rx 580!   🙂 

Mac Mini Late 2014 + MSI Radeon RX580 Gaming X [email protected]>TB3 (AKiTiO Node)

 
2014 Mac Mini [4th,2C,U] + RX 580 @ 16Gbps-TB2>TB3 (AKiTiO Node) + macOS 10.13 & Win10 [build link]  


ReplyQuote
Diego Ferreyra
(@diego_ferreyra)
New Member
Joined: 3 years ago
 
Posted by: Guru Nightmare

Great on MOJAVE Beta + Mac Mini 2014 (late) + Akitio Node + Rx 580!   Smile

how is your performance, any bench you can provide? Smile

To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.

.

ReplyQuote
jim_survak
(@jim_survak)
Eminent Member
Joined: 4 years ago
 
Posted by: Diego Ferreyra

no, I test it on a windows pc connect it directly to the motherboard...

Ah, well that is our problem then. On a typical motherboard you're connecting the graphics card to a dedicated set of PCIe lanes (typically 16, somewhat less common is to see 8). With Thunderbolt you have 4 PCIe lanes that are shared - so everything that runs through PCIe is shared on that one bus. So really what you are seeing (as well as me in my TB1 set up) is that all the data which is being processed has to be queued through the shared PCIe bus. Since that bus has half as many lanes (perhaps 1/4 as many) and has a capped data transfer rate of theoretically 20Gbps (PCIe x16 on a Gen3 connection is almost 128Gbps; PCIe x16 Gen2 is 64Gbps).

Hopefully that clears up the raw performance discrepancy.

As far as actual GPU usage, that one is a little bit trickier. I'm going to assume 2 things:
1) That this "xrg software" is accurate (personally I use iStat Menus & my eGPU shows 100% utilization in games).
2) It's a matter of driver optimization (there's simply far better graphics support on Windows than any other OS) and that DirectX can get closer to the metal than Apple's own Metal can. This is contingent on the fact that the program(s)/game(s) you're using are designed to use Metal. If not, then you're using OpenGL (or less likely Vulkan); neither of which renders the same performance as DirectX, despite the gap having been narrowed in recent years.
All of that is to say: if the software on macOS isn't specifically designed to utilize the GPU properly then you won't get the same GPU utilization out of it as on a Windows system (assuming the TB connection wasn't bottlenecking on the PCIe bandwidth in the first place).

Now, if you were booting to Windows on Boot Camp we'd have a much more comparable set of data as we'd only have to worry about software differences.

2018 Mac mini: Core i3, 8GB RAM, Sonnet 350 (aftermarket 550w PSU)+XFX Radeon RX-480 8GB Black Edition
2012 Mac mini: Core i7, 16GB RAM, Toshiba 1TB SSD, Seagate 1TB HDD
Custom: Ryzen 7 1700, 16GB Corasir DDR4-3200MHz RAM, 2x Corsair 500GB Neutron SSD, Seagate 3TB SSD, EVGA Nvidia 980Ti
YT:


ReplyQuote
RoniAyala
(@roniayala)
New Member
Joined: 3 years ago
 

Hi @mac_editor.
first of thanks for all your effort and time.
My system is working but I’m not able to Hotplug or restart my computer with the egpu connected to my MBP (it just freeze). Plus my MBP keeps turning off every time I eject the egpu and close the MBP. The only way it works is connecting the egpu once I’m logged in.

this is my setup:
late 2015 15” MBP
MANTIZ VENUS
AMD RX Vega 64
High Sierra 10.13.5

thank you very much!

To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.

.

ReplyQuote
 of  181