eGPU on the MBP 16"

[Austin Nguyen]

Alright, I feel like no one has touched on this subject yet. Can someone on the forums validate compatibility between an eGPU unit and a MBP 16"

This is the only information I have been able to come across about the new mac. Vtudio's testing shows that performance on the MBP 16" with an eGPU is worst than that of the old MBP 15" connected to an eGPU.

1) Did the new Catalina update (10.15.2) fix/optimize eGPU performance w/ the MBP 16".
2) Was Vtudio's results due to a software/firmware issue, or it a hardware issue?
3) If it's a hardware issue (pci express lanes not adequate)

Hope to gain more information on this topic
Thank you,
Austin Nguyen

 

[tobyg]

Alright, I feel like no one has touched on this subject yet. Can someone on the forums validate compatibility between an eGPU unit and a MBP 16"

This is the only information I have been able to come across about the new mac. Vtudio's testing shows that performance on the MBP 16" with an eGPU is worst than that of the old MBP 15" connected to an eGPU.

1) Did the new Catalina update (10.15.2) fix/optimize eGPU performance w/ the MBP 16".
2) Was Vtudio's results due to a software/firmware issue, or it a hardware issue?
3) If it's a hardware issue (pci express lanes not adequate)

Hope to gain more information on this topic
Thank you,
Austin Nguyen

I tried it.
My biggest takeaway is even though you drag a window to a display that is plugged into the eGPU, the program doesn't run on the eGPU.

For instance, I was running Parallels and dragged my Windows VM over to one of my external monitors hooked up directly to the eGPU via DisplayPort. Graphics performance was horrible and sluggish.

On my Windows machine, the app/rendering of the app is done by the GPU that the window is activated on. I've used Alienware laptops with eGPU's via their Alienware Graphics Amplifier, and it does work this way.

On the Mac side, you have to tell the program to prefer the eGPU for it to run on the eGPU. You can verify this using the Activity Monitor, it will show you what GPU your app is using.

Performance was much better when I actually started Parallels when setting it to prefer the eGPU.
The problem there is, you can't run Parallels with multiple VM's and have one on your internal laptop screen/integrated GPU and another VM on an external screen hooked up to your eGPU without having the performance of one of them suffer.

But other than that, things seem to work ok. I've only tested the eGPU with macOS on my MBP. I have not tried the eGPU with Bootcamp.

Regarding the video above, maybe he didn't have the app forced to run on the eGPU? Let's hope that's all it was.

 

[esatamacmodular]

check out egpu.io it is a much better resource for eGPU than this site.

 

[jerryk]

I tried it.
My biggest takeaway is even though you drag a window to a display that is plugged into the eGPU, the program doesn't run on the eGPU.

For instance, I was running Parallels and dragged my Windows VM over to one of my external monitors hooked up directly to the eGPU via DisplayPort. Graphics performance was horrible and sluggish.

On my Windows machine, the app/rendering of the app is done by the GPU that the window is activated on. I've used Alienware laptops with eGPU's via their Alienware Graphics Amplifier, and it does work this way.

On the Mac side, you have to tell the program to prefer the eGPU for it to run on the eGPU. You can verify this using the Activity Monitor, it will show you what GPU your app is using.

Performance was much better when I actually started Parallels when setting it to prefer the eGPU.
The problem there is, you can't run Parallels with multiple VM's and have one on your internal laptop screen/integrated GPU and another VM on an external screen hooked up to your eGPU without having the performance of one of them suffer.

But other than that, things seem to work ok. I've only tested the eGPU with macOS on my MBP. I have not tried the eGPU with Bootcamp.

Regarding the video above, maybe he didn't have the app forced to run on the eGPU? Let's hope that's all it was.

Parallels is not a good test of a eGPU.

 

[tobyg]

Parallels is not a good test of a eGPU.

Did you read what I wrote?

My point is, the performance of using the app on a monitor plugged into an eGPU, but not actually running the program "on" the eGPU, causes performance issues. The person in the video may have the same issue.

And just to add to this, when I run Windows in a VM on a monitor connected to the eGPU AND running Parallels and setting the option to "Prefer the eGPU", I can play 4k videos in Chrome with 1% packet loss.

If I try the same thing, running on a monitor connected to the eGPU but NOT actually running the app on the eGPU, that same 4k video has 35% frame drops and a horrible slide show. And everything just feels sluggish and slower.

That's my experience so far. That's all I was sharing.

If you have more to share, please do.

 

[jerryk]

Did you read what I wrote?

My point is, the performance of using the app on a monitor plugged into an eGPU, but not actually running the program "on" the eGPU, causes performance issues. The person in the video may have the same issue.

And just to add to this, when I run Windows in a VM on a monitor connected to the eGPU AND running Parallels and setting the option to "Prefer the eGPU", I can play 4k videos in Chrome with 1% packet loss.

If I try the same thing, running on a monitor connected to the eGPU but NOT actually running the app on the eGPU, that same 4k video has 35% frame drops and a horrible slide show. And everything just feels sluggish and slower.

That's my experience so far. That's all I was sharing.

If you have more to share, please do.

As I read your post you were running Windows in a VM on a Mac or Bootcamp. If either of those are the case, I am not at all surprised there would be issues. Also, any observations with respect to using or not using a eGPU in those environments would not necessary be transferable to users that run MacOS Catalina only.

please correct me if I misunderstood your testing.

 

[tobyg]

As I read your post you were running Windows in a VM on a Mac or Bootcamp. If either of those are the case, I am not at all surprised there would be issues. Also, any observations with respect to using or not using a eGPU in those environments would not necessary be transferable to users that run MacOS Catalina only.

please correct me if I misunderstood your testing.

You are misunderstanding.
Because there are no issues specifically with Windows or a VM. It has to do with the app and where it's run. I very well could have said I was running an app called GobblyGook. Plug in "Your favorite app name" above over the word "Parallels" and it fits the same.

The point is, things work better with the eGPU if you make sure the app is also running on the eGPU. Just dragging the app to the display connected to the eGPU, doesn't make it run on the eGPU. And in fact, performance in this scenario was worse than running it purely on the MBP and having the monitors directly hooked up to the MBP. Does that make sense? I was getting better performance using the app on my 4k external monitor when the 4k display was hooked up directly to the MBP and no eGPU. Once I introduce the eGPU, I get even worse performance on the display that's not connected to the GPU that the app is currently running on. Maybe that's common sense to most people, but Windows doesn't seem to behave this same way.

Windows seems to shift the app to running on the GPU that's connected to the monitor you dragged it to. Mac doesn't seem to implement their GPU solution this way. I don't recall if this is the same in a Mac Pro with dual GPU's too, or if this is purely an eGPU issue.

And yes, it relates to anyone running macOS Catalina with an eGPU. Regardless of the app, if you don't force the app to run on the eGPU, just dragging it to the monitor connected to the eGPU doesn't make it run on the eGPU. And it actually makes it run slower than if you had the monitor connected directly to the MBP and running it off of the integrated graphics.

My guess is that when you're running on the integrated GPU and drag the window to a display connected to the eGPU, the way macOS handles this is by copying the framebuffer from the card it was rendered on (integrated GPU in this example) to the framebuffer on the eGPU. And vice versa. And this is slower of course, vs having the video card that's rendering the app connected to the display it's being displayed on.

I'm guessing that's just going to confuse most people.

 

[maik_is_here]

What you write makes a lot of sense to me. I have not tested it, but might do so.
One notable disadvantage of ‘moving’ apps to run on the eGPU is that once you disconnect the eGPU you have to close all the apps that run on it, too. I have frequent problems with this, that it hangs and would not disconnect. I do it so infrequently, that I decided to simply shut down my MAC every time I disconnect from eGPU.

 

[jerryk]

You are misunderstanding.
Because there are no issues specifically with Windows or a VM. It has to do with the app and where it's run. I very well could have said I was running an app called GobblyGook. Plug in "Your favorite app name" above over the word "Parallels" and it fits the same.

The point is, things work better with the eGPU if you make sure the app is also running on the eGPU. Just dragging the app to the display connected to the eGPU, doesn't make it run on the eGPU. And in fact, performance in this scenario was worse than running it purely on the MBP and having the monitors directly hooked up to the MBP. Does that make sense? I was getting better performance using the app on my 4k external monitor when the 4k display was hooked up directly to the MBP and no eGPU. Once I introduce the eGPU, I get even worse performance on the display that's not connected to the GPU that the app is currently running on. Maybe that's common sense to most people, but Windows doesn't seem to behave this same way.

Windows seems to shift the app to running on the GPU that's connected to the monitor you dragged it to. Mac doesn't seem to implement their GPU solution this way. I don't recall if this is the same in a Mac Pro with dual GPU's too, or if this is purely an eGPU issue.

And yes, it relates to anyone running macOS Catalina with an eGPU. Regardless of the app, if you don't force the app to run on the eGPU, just dragging it to the monitor connected to the eGPU doesn't make it run on the eGPU. And it actually makes it run slower than if you had the monitor connected directly to the MBP and running it off of the integrated graphics.

My guess is that when you're running on the integrated GPU and drag the window to a display connected to the eGPU, the way macOS handles this is by copying the framebuffer from the card it was rendered on (integrated GPU in this example) to the framebuffer on the eGPU. And vice versa. And this is slower of course, vs having the video card that's rendering the app connected to the display it's being displayed on.

I'm guessing that's just going to confuse most people.

OK I see my confusion. Thanks. I was assuming you had set the Mac to use the eGPU and the external monitors are connected to the eGPU since the external monitors and heat/fans when in clamshell (ie. display shut) are what most people are complaining about.

Also, there is a way to tell Mac application apps to prefer the eGPU, https://support.apple.com/en-us/HT208544 see section "Use the preferred External GPU option" and "set the egpu connected display as primary display"). Can you try clamshell and this with your apps and see if it has an impact?

 

[bsbeamer]

eGPU with MBP16 works. Drivers for RX 5700 XT are not optimized, however. Getting ~15-20% less METAL scores in benchmarks in 10.15.1 and 10.15.2 vs RX580 in High Sierra 10.14.6. Real world performance is close to 33% less. For a GPU with more compute than RX580, something needs to be fixed.

 

[BigMcGuire]

eGPU with MBP16 works. Drivers for RX 5700 XT are not optimized, however. Getting ~15-20% less METAL scores in benchmarks in 10.15.1 and 10.15.2 vs RX580 in High Sierra 10.14.6. Real world performance is close to 33% less. For a GPU with more compute than RX580, something needs to be fixed.

Sorry to bring back an old thread but... this explains my experience recently. I have a Sonnet EGPU (550w) with an RX 580 that I put into a Windows PC build I did before the virus hit. I have a 4k monitor and wanted to be able to game on the Windows PC > 50fps with high graphics...... so I went out and bought an RX 5700 XT and ... wow big difference in speed. Where the RX 580 could barely do 45fps, the RX 5700 XT was pulling 90-100fps - ON WINDOWS.

I had the bright idea of trying to put the RX 5700 XT in my Sonnet EGPU thinking, great, I bet I can play games on high settings now on my MacBook! .... I got worse FPS (~5-10fps) than the Rx 580 and the RX 5700 XT graphics card was SCREAMING max fans the entire time. Major WTF?! I couldn't even talk over the screaming GPU - I thought I had melted it.


What brought me to this thread is that I was trying to figure out if running an EGPU helps my 2017 MBP 13' TB run cooler or not. Just using battery temps, it does not appear to have much of a difference between the EGPU and integrated graphics doing normal computing with an external 4k.

For a real scientific study, I'd need to pull out iStat Menus and do an exact comparison.

For me, the noise of the Sonnet EGPU's power supply fan drives me nuts. I like the 100% silence (no fans) of the 2017 MBP 13' to the 4k monitor.

 

[bsbeamer]

Sonnet 650 OC is much quieter than the first generation 550 eGPUs. Don’t know what version you have and hard to tell unless you know purchase date.

Would suggest making sure you’re using the stock, native, BIOS that is not modified. Also ensure your specific GPU is on Sonnet compatibility list.

macOS drivers for RX580 are polished. Even people with MP7,1 have better experience with RX580 than W5700 for certain tasks

 

[smbu2000]

I didn’t really notice a difference between PSUs, but can’t you replace the PSU in the Sonnet with any SFX PSU?
For a short while I had a Sonnet eGFX Breakaway Box (350W) and I swapped out the PSU with a 650W Enermax Revolution SFX PSU (ERV650SWT) to use with it.
I don’t have the Sonnet eGPU box anymore though.

As far as GPUs: On my previous base model 2019 15” MBP the Vega64 (Powercolor Red Devil Vega 64) that I used with it was super loud and the fans were running at full blast. When I switched over to a 5700XT (Sapphire Nitro+ 5700XT) the fans were much quieter on the GPU connected to my 4k monitor.
(That was on my current Asus XG Station Pro, not the Sonnet.)
I haven’t tried testing it on my current base model 16” MBP though so I’m not sure how it compares to my previous 2019 15” MBP with the eGPU.

I think the eGPU box might help the 15“/16” models more as they don’t have to rely on their dGPUs as much to push pixels on the external monitor, which turns on the MBP fan fairly quickly.
Someone correct me if I’m wrong.

 

[MacosaurusNEXT]

Noise wise I can recommend the Razer Core X with the MSI RX580, after swapping the bit noisy stock fans with some nice Be Quiets (120mm, 80mm inside PSU - void warranty!! ), it's super quite, 120mm & GPU fans stop after init procedure on normal desktop use (web, YouTube etc.), connected to an 4k Samsung via HDMI, 60hz, sadly enough it's only way to run my 2019 16" i9 with an external display w/o jetplane noise on the desk

 

[richinaus]

Egpu definitely helps with reducing the load on the MBP. I find mine is much better in boot camp.

a few things I have learned so far :

1. Egpu are slightly flakey on some apps. My MBP runs better without it sometimes.
2. Nvidia RTX wipes the floor with AMD. Same apps in boot camp with a RTX 2080 are massively better than my w5700 in macos.
3. They are super loud.
4. A big pain to be connecting and disconnecting.

Really I think if we want this power a desktop really is the best solution with a lower powered laptop.
Its just too much of a compromise.

Then the issue is which desktop, as a mid range Mac doesn’t exist [without a built in monitor].....but that’s another discussion.

 

[thedocbwarren]

Question for you as this is the direction i wish to go. I'm considering the Razer Core Chroma with an 5700 XT. I planned to plkug my 4K Samsung into it full time. Is this something that works well as in open up in the moring (not rebooting) and working throughout the day? I'm aware it's a pareiferial so it's likely to have the connect/disconnect by app. Does it work like a connected external display is what I mean?

Egpu definitely helps with reducing the load on the MBP. I find mine is much better in boot camp.

a few things I have learned so far :

1. Egpu are slightly flakey on some apps. My MBP runs better without it sometimes.
2. Nvidia RTX wipes the floor with AMD. Same apps in boot camp with a RTX 2080 are massively better than my w5700 in macos.
3. They are super loud.
4. A big pain to be connecting and disconnecting.

Really I think if we want this power a desktop really is the best solution with a lower powered laptop.
Its just too much of a compromise.

Then the issue is which desktop, as a mid range Mac doesn’t exist [without a built in monitor].....but that’s another discussion.

 

[richinaus]

Question for you as this is the direction i wish to go. I'm considering the Razer Core Chroma with an 5700 XT. I planned to plkug my 4K Samsung into it full time. Is this something that works well as in open up in the moring (not rebooting) and working throughout the day? I'm aware it's a pareiferial so it's likely to have the connect/disconnect by app. Does it work like a connected external display is what I mean?

Yes it does work like an external display and I have a 24” 4K LG plugged into mine.

However, I never have it permanently attached as I found it only benefitted me on certain apps, and a lot of the time it was not worth the hassle. I am going to be selling my w5700 soon because of this, most likely, as the RTX kills it for my uses.

For example all day today I had no requirement for it in any of my work.
Make sure you know what apps you will be using before jumping in, as it can be more hassle than it is worth [also depends on your current computer of course].

 

[MacosaurusNEXT]

Question for you as this is the direction i wish to go. I'm considering the Razer Core Chroma with an 5700 XT. I planned to plkug my 4K Samsung into it full time. Is this something that works well as in open up in the moring (not rebooting) and working throughout the day? I'm aware it's a pareiferial so it's likely to have the connect/disconnect by app. Does it work like a connected external display is what I mean?

Here is what I do with my Razer Core X (Core Chroma should be same), I use a 4k Samsung like you as a permanent display most of the time.

Before I sent my MacbookPro to sleep I detach the eGPU via menu bar symbol, screen goes black, with shortcut option+command+eject on my magic keyboard I sent the MBP to sleep. PSU fan does keep running, because it charges my MBP. You have to switch of the eGPU by hand if you find this annoying.

 

[thedocbwarren]

Good to know. I'm looking for something that is more powerful than the 5500m for graphics and light gaming now and then. I did plan to use for FCP as well.

Most of my work, however, is software development. I do want to start using machine learning but Mac isn't the platform for this so I use the cloud. My desire is to Mac game where able vs bootcamp.

My intention is to have my external monitor run entirely off the eGPU and leave the laptop on. The idea is to unleash the CPU from the thermal sharing when I really push the system (which happens with some of my own apps.)

Yes it does work like an external display and I have a 24” 4K LG plugged into mine.

However, I never have it permanently attached as I found it only benefitted me on certain apps, and a lot of the time it was not worth the hassle. I am going to be selling my w5700 soon because of this, most likely, as the RTX kills it for my uses.

For example all day today I had no requirement for it in any of my work.
Make sure you know what apps you will be using before jumping in, as it can be more hassle than it is worth [also depends on your current computer of course].

[automerge]1594663349[/automerge]
Would consider the fan replacement as described here as well. I'm disappointed in one way that we can't use NVIDIA GPUs in macOS but I can't fully blame Apple for that, it's a shared blame that punishes us in the end.

Here is what I do with my Razer Core X (Core Chroma should be same), I use a 4k Samsung like you as a permanent display most of the time.

Before I sent my MacbookPro to sleep I detach the eGPU via menu bar symbol, screen goes black, with shortcut option+command+eject on my magic keyboard I sent the MBP to sleep. PSU fan does keep running, because it charges my MBP. You have to switch of the eGPU by hand if you find this annoying.