eGPU for Windows Gaming

[Nunyabinez]

I was hoping to find someone on here who has used an eGPU on their Mac to do Windows gaming.

I currently have a decent dedicated gaming machine. Here are the important specs.

i5 4570
256GB SSD
1TB HD
GTX 1080

What I'm contemplating doing, is to sell this rig and buy an eGPU enclosure, put a GTX 1080ti in it and run it from my 2014 Retina iMac into a 27" 4K Dell monitor.

My question is; Has anyone here used an eGPU from an iMac or MacBook Pro to drive an external monitor under Boot Camp to play Windows games?

And if so, how does it stack up to a dedicated Windows box? Can I expect to get the same FPS out of a 1080ti out of the eGPU?

Thanks in advance for any help here, especially someone with actual experience doing this.

 

[Huntn]

Can this setup be used to run Windows via Mac/Bootcamp?

 

[Crash0veride]

You'd definitely take a performance hit just based on the bandwidth of thunderbolt 2 (I'm assuming you'd be using), which if you're gaming with that lovely 1080Ti seems like a waste. eGPU.io has some stats: https://egpu.io/forums/mac-setup/pcie-slot-dgpu-vs-thunderbolt-3-egpu-internal-display-test/ I beleive stats for MBP use bootcamp as the unreal engine window shows as PC.

Another good resource with enclosure lists: https://egpu.io/setup-guide-external-graphics-card-mac/

I'd be inclined to sell that iMac, upgrade your gaming deck, and load up a hackintosh set-up (If you've got the tech savvy nature): https://www.tonymacx86.com/.

 

[justinhill]

You'd definitely take a performance hit just based on the bandwidth of thunderbolt 2 (I'm assuming you'd be using), which if you're gaming with that lovely 1080Ti seems like a waste. eGPU.io has some stats: https://egpu.io/forums/mac-setup/pcie-slot-dgpu-vs-thunderbolt-3-egpu-internal-display-test/ I beleive stats for MBP use bootcamp as the unreal engine window shows as PC.

Another good resource with enclosure lists: https://egpu.io/setup-guide-external-graphics-card-mac/

I'd be inclined to sell that iMac, upgrade your gaming deck, and load up a hackintosh set-up (If you've got the tech savvy nature): https://www.tonymacx86.com/.

The bandwidth restrictions really only come into play in a meaningful way if you're trying to drive the internal display with an eGPU. OP specifically stated he wanted to use an external display, and even the article you posted mentions this in the second paragraph.

 

[Crash0veride]

The bandwidth restrictions really only come into play in a meaningful way if you're trying to drive the internal display with an eGPU. OP specifically stated he wanted to use an external display, and even the article you posted mentions this in the second paragraph.

You can see in the first linked tests that there is still ~ 20% drop in performance on an external display on GTX 1080Ti on many of the games in 1080p. It's just not a 30% drop. And the second paragraph is talking about thunderbolt 2 vs 3 not internal vs external."on External Display the Thunderbolt Number is not a big deal". It does have a greater effect on higher frame rates more so than lower.

 

[justinhill]

You can see in the first linked tests that there is still ~ 20% drop in performance on an external display on GTX 1080Ti on many of the games in 1080p. It's just not a 30% drop. And the second paragraph is talking about thunderbolt 2 vs 3 not internal vs external."on External Display the Thunderbolt Number is not a big deal". It does effect higher frame rates more than lower though.

You're right, my bad.

 

[Huntn]

You'd definitely take a performance hit just based on the bandwidth of thunderbolt 2 (I'm assuming you'd be using), which if you're gaming with that lovely 1080Ti seems like a waste. eGPU.io has some stats: https://egpu.io/forums/mac-setup/pcie-slot-dgpu-vs-thunderbolt-3-egpu-internal-display-test/ I beleive stats for MBP use bootcamp as the unreal engine window shows as PC.

Another good resource with enclosure lists: https://egpu.io/setup-guide-external-graphics-card-mac/

I'd be inclined to sell that iMac, upgrade your gaming deck, and load up a hackintosh set-up (If you've got the tech savvy nature): https://www.tonymacx86.com/.

I vote Hachntosh, if it runs the MacOS reliably and not to much of a pain. What's interesting is Apple is talking about VR, so maybe in the near future it will be more gaming friendly but I can'T vouch for affordability.

 

[Irishman]

I vote Hachntosh, if it runs the MacOS reliably and not to much of a pain. What's interesting is Apple is talking about VR, so maybe in the near future it will be more gaming friendly but I can'T vouch for affordability.

What about going with an upgraded base Mac Mini with TB 3 (once Apple ships it)? Then you can add a Sonnet eGPU box, and your GPU of choice, outputted to something like this Benq Zowie XL monitor https://www.bestbuy.com/site/benq-zowie-xl-series-24-lcd-hd-monitor/5514304.p?skuId=5514304 .

I think it would be hard to beat for the price.

 

[gkarris]

I found this good channel (the same guys as the magazine?) PC Gamer.

Since they are sponsored by MSI, they benchmark every GPU so you can see the relative performance for the price or how your GPU would run that game.

 

[martinm0]

The short answer is as sexy as eGPUs sound, they just can't compete with a built gaming PC from my experience.

I've dabbled with eGPUs here and there over the last few years. While they are handy to have if you only have a laptop or a non-gaming Mac, the losses in FPS and overall bandwidth due to the use of Thunderbolt (TB1 and TB2 only in my tests) were too much of a hit coupled with the high price of the enclosure and GPU. After toying with the Akitio Thunder2 for a year, I retired it and built a Mini ITX gaming PC that ran circles around my MacBook Pro 13" and eGPU.

Then after running the Mini ITX build in a tiny Lian Li PC-TU100 case, I decided I wanted to move up to the GTX 1080 full size, and built a Ncase M1 that can fit a full size GPU in a small mini ITX case. At this time the Akitio Node started popping up as a all in one eGPU with PSU and cables (I had to kluge all that together in the Thunder2 box). I ordered up the Akitio Node, and when it arrived, it was larger than the Ncase M1 gaming PC!

I did some testing and I don't have any numbers laying around, unfortunately, but there was a good amount of loss in the GTX 1080 performance between the full built, i6700k CPU, gaming machine and running it from a 2015 rMBP 13 with eGPU.

In the end, the Akitio was noisy, expensive ($300), and actually bigger than my Ncase M1 (which now has a 1080ti installed), so I scrapped the idea of eGPUs for now and just use a dedicated gaming PC.

One last comment - my tests were not ideal since I'm benching against a quad core CPU in the PC, and a dual core in the rMBP, but that is another argument against the eGPU if you don't have a quad core Mac and think you'll get everything out of your 1080 or 1080ti, and you won't. That will be another performance drop due to the lower hardware.

 

[gkarris]

The short answer is as sexy as eGPUs sound, they just can't compete with a built gaming PC from my experience.

I've dabbled with eGPUs here and there over the last few years. While they are handy to have if you only have a laptop or a non-gaming Mac, the losses in FPS and overall bandwidth due to the use of Thunderbolt (TB1 and TB2 only in my tests) were too much of a hit coupled with the high price of the enclosure and GPU. After toying with the Akitio Thunder2 for a year, I retired it and built a Mini ITX gaming PC that ran circles around my MacBook Pro 13" and eGPU.
...

I hear that TB3 is a vast improvement... I hear also the bottleneck is framerate but you can get higher resolutions to make up for it...

 

[martinm0]

I don't really recall that TB3 made that huge of an improvement (but I didn't have a TB3 port on my Mac, so I never was able to test).

I'm trying find the articles I was reading about TB3 and what you really gain over TB2, but this is the only one I'm finding quickly enough:

https://egpu.io/forums/mac-setup/pcie-slot-dgpu-vs-thunderbolt-3-egpu-internal-display-test/

In here, it shows that even with a TB3 connected eGPU and testing on an external monitor was still a 20% drop. With the internal display (sending the signal back over the TB cable into the laptop) is showing a 30% drop (at least for the GTX 1080 and 1080ti).

At least in this test all things are equal (hardware wise), so you see the true drops from going internal to eGPU. For the OP that is looking to make use a the iMac, there will be a hardware change, plus additional costs for the eGPU, which I personally think just doesn't provide enough performance for the money spent (and replacing an already fine gaming machine).

I do wish I had more access to do additional testing, but really in the end size and noise were my biggest issues. It's strange that my my mini ITX build is smaller and more quiet than the Akitio Node eGPU case and video card. Figured if I were to use an eGPU as a portable gaming rig with my rMBP, it would actually be larger and more pieces than just toting my Ncase M1 build directly.

 

[gkarris]

I don't really recall that TB3 made that huge of an improvement (but I didn't have a TB3 port on my Mac, so I never was able to test).

I'm trying find the articles I was reading about TB3 and what you really gain over TB2, but this is the only one I'm finding quickly enough:

https://egpu.io/forums/mac-setup/pcie-slot-dgpu-vs-thunderbolt-3-egpu-internal-display-test/

In here, it shows that even with a TB3 connected eGPU and testing on an external monitor was still a 20% drop. With the internal display (sending the signal back over the TB cable into the laptop) is showing a 30% drop (at least for the GTX 1080 and 1080ti).

I thought I saw somewhere, yes you get that drop, but only fps. But if you have a powerful enough card, you can up the resolution and still deal with only that same drop and not with the drop related to increasing the resolution...

 

[star-affinity]

I don't really recall that TB3 made that huge of an improvement (but I didn't have a TB3 port on my Mac, so I never was able to test).

I'm trying find the articles I was reading about TB3 and what you really gain over TB2, but this is the only one I'm finding quickly enough:

https://egpu.io/forums/mac-setup/pcie-slot-dgpu-vs-thunderbolt-3-egpu-internal-display-test/

In here, it shows that even with a TB3 connected eGPU and testing on an external monitor was still a 20% drop. With the internal display (sending the signal back over the TB cable into the laptop) is showing a 30% drop (at least for the GTX 1080 and 1080ti).

At least in this test all things are equal (hardware wise), so you see the true drops from going internal to eGPU. For the OP that is looking to make use a the iMac, there will be a hardware change, plus additional costs for the eGPU, which I personally think just doesn't provide enough performance for the money spent (and replacing an already fine gaming machine).

I do wish I had more access to do additional testing, but really in the end size and noise were my biggest issues. It's strange that my my mini ITX build is smaller and more quiet than the Akitio Node eGPU case and video card. Figured if I were to use an eGPU as a portable gaming rig with my rMBP, it would actually be larger and more pieces than just toting my Ncase M1 build directly.

Ouch, I didn't know the drop was that big when going via Thunderbolt. Was hoping it would be more around 10%.

I wonder why it's that big of a drop? Isn't Thunderbolt 3 supposed to be using PCI-E in a way? Is there some kind of overhead going on that could perhaps improve in the future (via software)?

 

[Crash0veride]

I wonder why it's that big of a drop? Isn't Thunderbolt 3 supposed to be using PCI-E in a way? Is there some kind of overhead going on that could perhaps improve in the future (via software)?

Thunderbolt 3 is equivalent to PCIEx4, internal cards are PCIEx16 (for the first card), so four times the bandwidth. Internal monitors take an extra hit because they have to share the bandwidth with the return data, while external does not. Also maybe latency is a factor?

The x# refers to how many CPU PCI lanes it's using. That's why 2016+ 13' MacBook pros are only full speed ports on one side - fewer PCI lanes on those CPUs.

 

[star-affinity]

Thunderbolt 3 is equivalent to PCIEx4, internal cards are PCIEx16 (for the first card), so four times the bandwidth. Internal monitors take an extra hit because they have to share the bandwidth with the return data, while external does not. Also maybe latency is a factor?

The x# refers to how many CPU PCI lanes it's using. That's why 2016+ 13' MacBook pros are only full speed ports on one side - fewer PCI lanes on those CPUs.

It’s just that here bandwidth doesn't seem to matter much:

The bandwidth of an eGPU's PCIe slot is severely restricted compared to the internal x16 PCIe 2.0 slot #1 of a Mac Pro tower. However, it does not necessarily restrict the performance of the graphics card.

If bandwidth was a critical performance factor, the Mac Pro tower would be TWO TO FOUR TIMES FASTER than the Macs with eGPUs. In reality, the performance gap was much smaller than the bandwidth gap.

http://barefeats.com/bandwidth.html

But it seems it does when higher resolution displays and faster GPUs are involved(?)