eGPU Performance Loss - PCI Express vs. Thunderbolt
 
Notifications
Clear all

eGPU Performance Loss - PCI Express vs. Thunderbolt  

 of  10
  RSS

ozzy
 ozzy
(@ozzy)
Estimable Member
Joined: 3 years ago
 
Posted by: P-Mac

You could probably test 4K for us by using DSR, even if you don't have a 4K monitor. Enable the 2.25x DSR factor in the NVIDIA Control Panel on a 1440p monitor to reveal 3840x2160 resolution as selectable to any benchmark / 3D app. 

Here are the results in 4k resolution using the 2.25x DSR factor.  It's 20.7 FPS, which interestingly is not 25% of the 73.2 FPS I got at 1080p (if it scaled 100% linearly I'd expect 18.3 FPS, since it is 4x the number of pixels).

4k External 1070
2012 15" MacBook Pro Retina (GT650M) [3rd,4C,Q] + GTX 1070 @ 10Gbps-TB1>TB3 (AKiTiO Node) + macOS 10.13 & Win10 [build link]  

ReplyQuote
enjoy
(@enjoy)
Reputable Member
Joined: 4 years ago
 

I found this in Youtube with the i5 CPU ( ):

So you have 15% performance drop maybe 20% (because of CPU) on External Display (Monitor) at 4k - Thunderbolt 1 is in the game but ONLY ON EXTERNAL DISPLAY (MONITOR)!

MacBook Pro (Retina, 15-inch, Later 2013) 3.2GHz Quad Core Intel i7-4750HQ / 8 GB 1600 MHz DDR3 / 256GB SSD + 1TB
mini eGPUPCI Express vs. ThunderboltMac CAN gameGaming Laptops vs. MacBook Pro with eGPU

 
2012 13" MacBook Air [3rd,2C,U] + GTX 1060 @ 10Gbps-TB1 (AKiTiO Thunder2) + macOS 10.12 [build link]  


ReplyQuote
enjoy
(@enjoy)
Reputable Member
Joined: 4 years ago
 

@P-Mac i will answer with video:

I love eGPU because i love my MacBook Pro Retina (not new model) and to can game all new games 1080p with mini-eGPU and GTX 1060 6GB all to the max on my Internal Retina Display is just a pleasure that i can't describe, but we can't hide that we have performance drop, for me that performance drop is not a problem but for some users maybe IS a big problem and we need to show them that, because eGPU is not for everyone and it is not cheap. I don't play online games so 60FPS, 144FPS is not my target but for other users will be a huge problem to not reach them... I know that Heaven benchmarks are not strictly perfect results because the CPUs are not the same and some benchmarks are with overclocked GPUs other not but i think they are not bad example of performance drop at all Smile


Pictures from the video:

.... AKiTiO Node is very very BIG wow Smile

MacBook Pro (Retina, 15-inch, Later 2013) 3.2GHz Quad Core Intel i7-4750HQ / 8 GB 1600 MHz DDR3 / 256GB SSD + 1TB
mini eGPUPCI Express vs. ThunderboltMac CAN gameGaming Laptops vs. MacBook Pro with eGPU

 
2012 13" MacBook Air [3rd,2C,U] + GTX 1060 @ 10Gbps-TB1 (AKiTiO Thunder2) + macOS 10.12 [build link]  


ReplyQuote
enjoy
(@enjoy)
Reputable Member
Joined: 4 years ago
 

This is the new Superposition benchmark and the results are amazing, because it is heavy benchmark and low FPS give us lower performance drop or we don't have a performance drop on Internal or External Display... 

Desktop PC (PCIe @126Gbps)

CPU: Intel I7 6700K

GPU: MSI GTX 1060 Gaming X 6GB

RAM: Corsair Dominator Platinum 16GB

Motherboard: MSI Z170A Gaming M5

eGPU (Thunderbolt 2 @16Gbps)

CPU: Intel i7-4750HQ

GPU: EVGA GTX 1060 6GB SC

RAM: 8 GB 1600 MHz DDR3 

MacBook Pro (Retina, 15-inch, Late 2013)

  1. Desktop PC with MSI GTX 1060 Gaming X 6GB
  2. eGPU Internal Laptop Display with EVGA GTX 1060 6GB SC
  3. eGPU External Display (Monitor) with EVGA GTX 1060 6GB SC

 

the video from which i get the Desktop PC results

MacBook Pro (Retina, 15-inch, Later 2013) 3.2GHz Quad Core Intel i7-4750HQ / 8 GB 1600 MHz DDR3 / 256GB SSD + 1TB
mini eGPUPCI Express vs. ThunderboltMac CAN gameGaming Laptops vs. MacBook Pro with eGPU

 
2012 13" MacBook Air [3rd,2C,U] + GTX 1060 @ 10Gbps-TB1 (AKiTiO Thunder2) + macOS 10.12 [build link]  


ReplyQuote
ozzy
 ozzy
(@ozzy)
Estimable Member
Joined: 3 years ago
 
Posted by: P-Mac

 

Posted by: ozzy

 

Posted by: enjoy

 

@ozzy you can run 3DMark FireStrike Ultra 1.1 without 4k monitor too like me!

   

I downloaded 3DMark, but looks like I have to pay to run FireStrike.  I ran the Time Spy benchmark that was included and it got 4716 in their default settings (1440p I believe).

   

Sigh...c'mon Enjoy, at least give the guy some info to work with besides just saying it's possible. 

@ozzy - You can run FireStrike Basic from the demo version of 3DMark, you just click over to the benchmarks tab and you'll notice it's not locked. However, since this is within the context of testing for per-resolution performance deltas, the demo version of 3DMark won't help as FireStrike Ultra and Extreme are locked behind the pay wall, and that's the only way to enable higher render resolutions for those tests.

You can run Heaven / Valley for free, as they allow you to change the render resolution, and if you enable the 2.25x DSR factor in the NVIDIA Control Panel on a 1440p monitor, 3840x2160 should become a selectable resolution for those benchmarks in full screen, allowing you to put down numbers in 4K for your GPU. Smile

   

Ok - I didn't pay for the benchmark, but I did run the Fire Strike normal one just so you guys could see/compare if interested.  This is at 1080p.  Here are the results:

Firestrike 1080p 1070 External
2012 15" MacBook Pro Retina (GT650M) [3rd,4C,Q] + GTX 1070 @ 10Gbps-TB1>TB3 (AKiTiO Node) + macOS 10.13 & Win10 [build link]  

enjoy liked
ReplyQuote
enjoy
(@enjoy)
Reputable Member
Joined: 4 years ago
 

Thanks to @itsage we finally have result with the same CPU in the same Desktop PC with attached to the motherboard Thunderbolt 3 port eGPU (AKiTiO Node with GTX 980Ti 6GB Hybrid) and the results shows the same close to 20% percent performance drop with External Display for Thunderbolt 3 (32Gbps)!

GTX 980Ti 6GB Hybrid in Desktop PC (PCIe slot 126Gbps)

GTX 980Ti 6GB Hybrid in eGPU (Thunderbolt 3 32Gbps) External Display - 16% performance drop


GTX 980Ti 6GB Hybrid in Desktop PC (PCIe slot 126Gbps)

GTX 980Ti 6GB Hybrid in eGPU (Thunderbolt 3 32Gbps) External Display - 16% performance drop for 1080p again on FireStrike 1.1 and lower for Time Spy because it is 4k benchmark!


So after that test we can confirm that all results we saw so far for 1080p in this thread are true story even they are with different CPUs and some GPUs are overclocked! 

BENCHMARKS DESKTOP PC (126 Gbps)
GTX 980TI 6GB
EGPU THUNDERBOLT 3 (32Gbps)
EXTERNAL MONITOR | GTX 980TI 6GB
     
Unigine Valley 3807 (91.0 FPS) 3250 (77.7 FPS)
Unigine Heaven 2279 (90.5 FPS) 1918 (76.1 FPS)
Unigine Superposition 3806 (28.5 FPS) 3653 (27.3 FPS)
3DMark Time Spy 5878 (38.0 FPS) 5342 (34.4 FPS)
3DMark Fire Strike 15997 (49.7 FPS) 13233 (49.2 FPS)
     
Metro Last Night Redux 138.1 FPS 93.5 FPS
Rise of the Tomb Raider 59.9 FPS 58.1 FPS
Tom Clancy's The Division 59.8 FPS 46.8 FPS

  

enjoy Smile


I want to repost this user post, to show him that the results are now with the same CPU!

Posted by: gtosi

 

Maybe we should not compare the same graphics card in different machines. A quick search shows that for the same desktop graphics card it is normal to see up to 20-30% differences in Unigine performance depending on the rest of the computer (especially CPU).

MacBook Pro (Retina, 15-inch, Later 2013) 3.2GHz Quad Core Intel i7-4750HQ / 8 GB 1600 MHz DDR3 / 256GB SSD + 1TB
mini eGPUPCI Express vs. ThunderboltMac CAN gameGaming Laptops vs. MacBook Pro with eGPU

 
2012 13" MacBook Air [3rd,2C,U] + GTX 1060 @ 10Gbps-TB1 (AKiTiO Thunder2) + macOS 10.12 [build link]  


ReplyQuote
Pimodius1
(@pimodius1)
New Member
Joined: 3 years ago
 
I have an internal 4k display, so it is not possible to play on 4k Will using an eGPU?
and is the bandwidth higher or lower if you play on a higher resolution on an internal display?

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

.

ReplyQuote
ImDe
 ImDe
(@imde)
Active Member
Joined: 3 years ago
 

Guys, can you explain why there is no difference between thunderbolt 1,2,3 on external monitor at 4k in your tests? (you show only 20% drop Vs desktop PCIe) 

 

And how does it work with thunderbolt 3 with jhl6240 (PCIE 3.0 x2) on external 4k? Remaining bandwith, overall speed like with thunderbolt 2?

2017 13" HP Spectre X360 [8th,4C,U] + GTX 1050 @ 32Gbps-TB3 (Lenovo TB3 GFX Dock) + Win10 [build link]  

ReplyQuote
enjoy
(@enjoy)
Reputable Member
Joined: 4 years ago
 
Posted by: itsage

I finally got this test bench going without error 12. Here are some numbers comparing x16 PCIe slot, X4 through PCH PCIe slot, and X4 through PCH Thunderbolt 3. The eGPU is a Razer Core + GTX 980 Ti.

Z170X Test Bench PCI Express (Desktop PC) Thunderbolt 3 eGPU
Unigine Valley 95.1 FPS 83.7 FPS
Unigine Heaven 90.7 FPS 79.0 FPS
Unigine Superposition 93.2 FPS 86.9 FPS
3DMark Time Spy 36.0 FPS 33.5 FPS
3DMark Fire Strike 88.2 FPS 71.7 FPS
Rise of the Tomb Raider 60.0 FPS 58.5 FPS
Tom Clancy's GhostRecon 87.9 FPS 56.6 FPS
Shadow of Mordor 128.5 FPS 97.3 FPS

Thanks to the hard work and tests from @itsage we finally have full battle PCI Express vs. Thunderbolt 3 - You can see again that if you can get more FPS you will have more performance drop. The performance drop is still 15-20% of Unigine Heaven for FULL HD resolution!

MacBook Pro (Retina, 15-inch, Later 2013) 3.2GHz Quad Core Intel i7-4750HQ / 8 GB 1600 MHz DDR3 / 256GB SSD + 1TB
mini eGPUPCI Express vs. ThunderboltMac CAN gameGaming Laptops vs. MacBook Pro with eGPU

 
2012 13" MacBook Air [3rd,2C,U] + GTX 1060 @ 10Gbps-TB1 (AKiTiO Thunder2) + macOS 10.12 [build link]  


ReplyQuote
Atty
 Atty
(@atty)
Active Member
Joined: 3 years ago
 

I think I've read over most of the posts but I have a few questions. 

I'm considering getting a eGPU set up for my 2017 15'' MacBook Pro (highest spec'd model). I'd consider a 1070 or 1080, depending on research and probably the Sonnet eGPU box.

I really want the set up to run in Windows 10 with Xbox Play anywhere games (Forza, Halo, Gears, etc) and mainly racing simulators (rFactor 2, Assetto Corsa, Projects Cars 2, etc). 

I'm curious how the external monitor testing is done. Is it with the laptop in Dock mode, so lid closed and all video output to the external? 

Has anyone done input lag testing to determine the lag introduced with all this I/Os? 

Any other information would be super helpful, I was going to post about some of these topics but didn't know which forum to use and this topic covers one of my big concerns which is performance loss against a PC. I'm doing this instead of building a PC and I don't want to lose so much performance that the overhead isn't worth it for the investment versus just building a PC. 

 

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

.

ReplyQuote
 of  10