eGPU Performance Loss - PCI Express vs. Thunderbolt  

Yes @ozzy

Thunderbolt1, Thunderbolt2 and Thunderbolt3 have the same performance if you use External Display (Monitor) for 1080p. This thread is so important for users with Thunderbolt1. If they use External Display (Monitor) and quad-core processor they are still in the game 🙂

Posted by: enjoy

 

Yes @ozzy

Thunderbolt1, Thunderbolt2 and Thunderbolt3 have the same performance if you use External Display (Monitor) for 1080p. This thread is so important for users with Thunderbolt1. If they use External Display (Monitor) and quad-core processor they are still in the game 🙂

   

The situasion resume is perfect for 1080p the situation is sharp. Thunderbolt 1 is not able to efficently manage 2k or 4k. Am I right ?

Posted by: vava726

 

The situasion resume is perfect for 1080p the situation is sharp. Thunderbolt 1 is not able to efficently manage 2k or 4k. Am I right ?

You can see from that chart:

For 1440p (2560x1440) the Thunderbolt1 will perform also the same like Thunderbolt2 & Thunderbolt3 but maybe with some losses. It is not recommended to run 4k with Thunderbolt1 with Thunderbolt2 will be OK but again we will have some drops! For 4k Thunderbolt3 is strongly recommended!

Something like this:

• Thunderbolt1 - 1680x1050 Internal Display / 1080p External Display Monitor
• Thunderbolt2 - 1080p Internal Display / 1440p External Display
• Thunderbotl3 - 1440p Internal Display / 4k External Display

4k and eGPU with 1080Ti = YES:

So 4k on TB1 external display is not possible? Even for browsing the web? I don't need to be able to play games.

You can run an 8K display from an eGPU connected via Expresscard. 4K output from an eGPU connected via TB1 works just fine, even for running games. See my sig 🙂

If you use the Thunderbolt1 port as a mini-DisplayPort connection, however, then you cannot output 4K60 to your monitor.

Thanks, I would thinks so.
Not sure why Enjoy is writing it's not possible.

Posted by: Fred.

 

Thanks, I would thinks so.
Not sure why Enjoy is writing it's not possible.

   

My post is about remaining bandwidth and especially performance drop and gaming with 4k with Thunderbolt1!

enjoy, I still do not understand why you are making the (completely false) statement that you cannot play at 4K via an eGPU connected over Thunderbolt1. Are you talking about an internal display use case (if any TB1 systems have 4K internal displays, that is)?

Posted by: Yukikaze

 

enjoy, I still do not understand why you are making the (completely false) statement that you cannot play at 4K via an eGPU connected over Thunderbolt1.

   

I told it's not recommended, because users will have big performance drop because of Thudnerbolt 1..

Can you please post your results for Fire Strike Ultra bench?

Posted by: enjoy

 

Posted by: Fred.

 

Thanks, I would thinks so.
Not sure why Enjoy is writing it's not possible.

   

My post is about remaining bandwidth and especially performance drop and gaming with 4k with Thunderbolt1!

   

No worries man. Not possible is different from possible with limitations / performance drop 😉
Not all of us are gamers, We just want a 4K monitor on our Macbook. Maybe you should make make a difference between Gaming and web browsing or something.

@Fred if you interested i can confirm that if we have higher resolution we will have less performance drop i have full test of FireStrike 1.1 (1080p) , FireStrike Extreme 1.1 (1440p) & FireStrike Ultra 1.1 (4k):

• 1080p - 24% performance drop Internal Display

3DMark FireStrike 1.1 (INTERNAL DISPLAY - Optimus) 1080p / MacBook Pro (Retina, 15-inch, Later 2013) with eGPU with GTX 1060 6GB Thunderbolt 2 (16Gbps):

3DMark FireStrike 1.1 (basic test) Desktop PC (126Gbps) the same processor 1080p -   http://www.3dmark.com/fs/9891027

• 1440 - 8% performance drop Internal Display

3DMark FireStrike Extreme 1.1 (basic test) INTERNAL DISPLAY (Optimus) 1440p / MacBook Pro (Retina, 15-inch, Later 2013) with eGPU with GTX 1060 6GB Thunderbolt 2 (16Gbps):

3DMark FireStrike Extreme 1.1 (basic test) Desktop PC (126Gbps) the same processor 1440p -   http://www.3dmark.com/fs/10811461

• 4k - 8% performance drop Internal Display

3DMark FireStrike Ultra 1.1 (basic test) INTERNAL DISPLAY (Optimus) 4k  / MacBook Pro (Retina, 15-inch, Later 2013) with eGPU with GTX 1060 6GB Thunderbolt 2 (16Gbps):

3DMark FireStrike Ultra 1.1 (basic test) Desktop PC (126Gbps) the same processor 4k -    http://www.3dmark.com/fs/11705916

So I can confirm - Performance drop is lower = if your FPS is lower - CONFIRMED! only 8% performance drop for Thunderbolt2 (16Gbps) Internal Laptop Display  vs. Desktop PC (126Gbps)

Posted by: enjoy

@Fred if you interested i can confirm that if we have higher resolution we will have less performance drop i have full test of FireStrike 1.1 (1080p) , FireStrike Extreme 1.1 (1440p) & FireStrike Ultra 1.1 (4k):

That's Interesting! 😀
I currently don't have a 4k monitor and I was going to wait for the new Asus 4k UHD monitor. So I can't test.
Still figuring out things too. All new to me.

Those results are all for internal monitors though. And I completely agree that it would be unfeasible to get 4k on an internal monitor with TB1 bandwidth.  However, most TB1 computers (Mac's anyway) can't actually even output to a 4k external display because of the version of HDMI/Display Port/Thunderbolt built into them.  But having an eGPU with a DisplayPort 1.2/1.4 port would easily allow this to work.

There is a big question about gaming in 4k though, as you likely would need a 1080Ti to properly do it, and there is an honest/open question as to whether the TB1 bandwidth would limit the datafeed to the 1080Ti causing it to not be able to perform as well as it could.  I don't know if we have any good data on this at the moment.  All I can say is that with a 1070 on my TB1 computer, I got ~73 FPS in 1080p Heaven benchmark, and ~45 FPS in 1440p Heaven benchmark.  These are definitely faster than enjoy got on his similar computer with a 1060 at 1080p (73 vs. 55 FPS), so it does show that the 1070 is definitely worth it and adds significant incremental value on a TB1 computer at 1080p on an external monitor.  I don't know if a 1080ti card would also though.

I don't have a 4k monitor, and the trend shows my 1070 likely wouldn't be powerful enough to drive gaming in 4k (although it could certainly drive a 4k display for non-gaming).  I don't know how my stats compare to a similar processor video card at TB3 speeds or internal, and therefore how you could expect a 1080Ti to perform for gaming at 4k compared to how it does in a desktop.

All I can say is that with a 1070 on my TB1 computer, I got ~73 FPS in 1080p Heaven benchmark, and ~45 FPS in 1440p Heaven benchmark.

Interesting. Going off the OP in this reddit post, they scored 94.8fps @ 1080p, and 56.2fps @ 1440p. So looking at the delta between your TB1 scores and their desktop scores, it appears you have a 23% loss at both resolutions, which leads me to think that TB1 is in fact your primary bottleneck. (Your secondary bottleneck would likely be the CPU; OP's post has an Ivy Bridge CPU, while others in their thread have more powerful CPUs and are also getting higher scores. Regardless, you can't do anything about the TB1 bottleneck so the CPU doesn't really matter unless it's thermal throttling hard)

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. 

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

Posted by: ozzy

 

That basically means the new CPU and TB3 vs TB1 makes basically no difference on an external monitor.

   

The big asterisk behind that sentence is that this is for Heaven benchmark only. Heaven is not a very bus-intensive benchmark compared to a modern game...it's computationally heavy, sure, but modern games have a ton of texture paging and resource management that is happening over the PCIe bus in addition to draw calls. 

I would wager that Heaven (and for that matter, Superposition) testing belies the bandwidth limitations of Thunderbolt versions when in real-world gaming scenarios. 

Case in point, my TXp (2017) can pull a score within 10% of a desktop-installed TXp in Superposition, but when playing GTA V, I see framerate drops lower than my buddy's desktop-installed GTX 1080. Both of us running 1080p, and both of us in similar scenarios (as we're playing multiplayer together and are looking at the same scenes in the same location in-game, etc.) 

Clearly, the Titan should be wiping the floor with the 1080, but it's not the case in GTA V because it's constantly paging textures and geometry data in and out, and this overhead starts to make Thunderbolt's bandwidth limitation more apparent. 

Posted by: P-Mac

 

Posted by: ozzy

 

That basically means the new CPU and TB3 vs TB1 makes basically no difference on an external monitor.

   

The big asterisk behind that sentence is that this is for Heaven benchmark only. Heaven is not a very bus-intensive benchmark compared to a modern game...it's computationally heavy, sure, but modern games have a ton of texture paging and resource management that is happening over the PCIe bus in addition to draw calls. 

I would wager that Heaven (and for that matter, Superposition) testing belies the bandwidth limitations of Thunderbolt versions when in real-world gaming scenarios. 

Case in point, my TXp (2017) can pull a score within 10% of a desktop-installed TXp in Superposition, but when playing GTA V, I see framerate drops lower than my buddy's desktop-installed GTX 1080. Both of us running 1080p, and both of us in similar scenarios (as we're playing multiplayer together and are looking at the same scenes in the same location in-game, etc.) 

Clearly, the Titan should be wiping the floor with the 1080, but it's not the case in GTA V because it's constantly paging textures and geometry data in and out, and this overhead starts to make Thunderbolt's bandwidth limitation more apparent. 

   

Very good point - this is not necessarily in a real-world scenario but for that specific benchmark.  I'll try to do the 3DMark benchmark a bit later today.

Posted by: P-Mac

 

All I can say is that with a 1070 on my TB1 computer, I got ~73 FPS in 1080p Heaven benchmark, and ~45 FPS in 1440p Heaven benchmark.

Interesting. Going off the OP in this reddit post, they scored 94.8fps @ 1080p, and 56.2fps @ 1440p. So looking at the delta between your TB1 scores and their desktop scores, it appears you have a 23% loss at both resolutions, which leads me to think that TB1 is in fact your primary bottleneck. (Your secondary bottleneck would likely be the CPU; OP's post has an Ivy Bridge CPU, while others in their thread have more powerful CPUs and are also getting higher scores. Regardless, you can't do anything about the TB1 bottleneck so the CPU doesn't really matter unless it's thermal throttling hard)

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. 

   

Interesting - that makes sense (assuming our 1070 clock speeds were the same).  A ~20% loss sounds about right (as you said, their CPU is slightly better), with people on TB3 reporting more like a 10-15% loss if I remember correctly.  23% loss is definitely not bad for a 5-year old laptop though 🙂

Posted by: ozzy  

 

23% loss is definitely not bad for a 5-year old laptop though 🙂

Not bad at all. It's why we do what we do. 

I told that many times but future eGPU users wants to view this for sure:

• GTX 1060 - HIGH/ULTRA gaming with 1080p on External Display (TB1, TB2, TB3) or Internal Display (TB2, TB3)
• GTX 1070 - ULTRA gaming with 1080p on External Display (TB1, TB2, TB3) or Internal Display (TB2, TB3)
• GTX 1070 - MID/HIGH gaming with 1440p on External Display (TB1, TB2, TB3)  or Internal Display (TB2, TB3) 
• GTX 1080 & 1080TI - HIGH/ULTRA gaming with 1440p on External Display (TB1, TB2, TB3) or Internal Display (TB2, TB3)
• GTX 1080 & 1080TI MID/HIGH gaming with 4k only External Display (TB1, TB2, TB3)

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).

Posted by: enjoy

I told that many times...

I'm trying to prevent misinformation and incorrect conclusions being drawn by your barrage of redundant data, when the fact of the matter is that eGPU performance delta is a complex issue.

This pretty little list of resolutions vs. GPUs shouldn't be touted as some definitive reference of real-world eGPU gaming performance, for a few reasons:

1. You are not using a real-world gaming test, playing actual games for the majority of your testing data, therefore you can't draw a complete conclusion without looking at the whole story. Gaming with modern, large texture-set games utilizes the PCIe bus in ways and amounts that vary significantly compared to Heaven, and all your list does is tell us that we can sit there and watch Heaven all day at those resolutions. In real-world gaming scenarios, performance loss via eGPU is a more complicated issue.
2. You aren't listing refresh rate, which changes computational requirements significantly, and also can affect the amount of which the TB bottleneck is responsible for the eGPU performance delta.
3. This list is basically preaching to the choir since the relative performance of the listed cards is already a known factor.

For example, playing The Division (4K 60Hz, maxed settings, vsync adaptive), there are situations in which my Titan is:

• Not bottlenecked at all - simple scene, 60fps is being achieved, and as a result the card is not at full utilization
• TB bottlenecked - moderately complex scenes, but traveling around the game world from point A to point B. Card utilization isn't near 99%, meaning the GPU is spending a lot of time idle waiting for textures or data to be sent to it for processing
• CPU bottlenecked - complex scenes (e.g. gunfights in wide open spaces) with a lot of particles, AI, and physics effects - GPU utilization in the 90's, moderate bus utilization, but CPU melting as it tries to set up these complex scenes with a lot of game overhead
• GPU bottlenecked - very complex scenes (lots of volumetric fog, foliage with subsurface scattering and shadows) with little to no AI/particles/physics - GPU utilization at 99%, bus and CPU usage being moderate. This is when the performance delta is the least, compared to desktop-installed equivalent cards. This also happens to be the workload in a lot of GPU benchmarks (3DMark graphics, Heaven, Valley, SP)

A 10-20% performance loss with an eGPU is a best-case scenario, not an average as far as gaming performance is concerned. The answer to someone asking "which GPU should I put in my eGPU" should always be "what are you looking to do with it?"

The resolution someone plays at, the framerate they are targeting, their Thunderbolt version, their CPU, and even the specific games that they play can have a huge impact on which GPU they should be looking at and what kind of performance they will experience. 

---

The reason I am so interested in this topic is because I've seen the 10-20% number being thrown around since I first got my eGPU setup running (around june 2016), and in that time, I have learned a lot about the different bottlenecks in a laptop + eGPU system with how it relates to real-world gaming performance, because, well... I play games with my computer, not solely use it for benchmarks. I've ran a 980, a 980 Ti in my eGPU setup before. Overclocked GPUs, BIOS mods, thermal paste replacements, underclocking & undervolting the CPU for thermal headroom...I love tinkering and if there's a setting that could potentially benefit gaming performance, you can bet I've touched it in my quest to understand how our unique eGPU-based systems deal with the workloads of modern games.

The problem with the "oh cool the GPU will basically be 10-20% slower than desktop-installed" conclusion is this: I was fortunate enough to upgrade to a TITAN Xp in my eGPU. You'd figure "10-20% loss" from the most powerful GPU out right now would be totally fine to handle 1080p Ultra gaming, right? 

In DOOM, I've seen FPS dips down into the 60s and 70s, and I play at 144Hz since it's a twitch-style shooter. ROTTR was barely better of an experience than my old 980 Ti. VR (90fps) in Elite Dangerous & DIRT Rally is a mixed bag. The Division (60Hz), as outlined above, is also a mixed bag...but Elite Dangerous on the 1080p display runs fine at 4K DSR 60Hz with barely any dips below 60fps. GTA V has huge framerate swings and dips, but DSR 4K helps mitigate that. 

A modern game is a much different beast from a focused benchmark and that's why I caution against drawing a general conclusion about gaming performance when the majority of data that we are drawing the conclusion from is benchmark-based.

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. 🙂

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).

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)!

@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 🙂

Pictures from the video:

.... AKiTiO Node is very very BIG wow 🙂

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... 

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

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. 🙂

   

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:

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! 

enjoy 🙂

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).

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?

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?

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!

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. 

UPDATE: Again thanks to @itsage we have complete benchmarks for 1080p (15% performance drop), 1440p (8% performance drop) and 4k (5% performance drop) with the same CPU in the same Desktop PC with attached to the motherboard Thunderbolt 3 port eGPU (Razer Core with GTX 980Ti 6GB Hybrid)

Hello How are you all doing, 

Just wanted to ask can someone Compare the GTX 1080 and the GTX 1070 in Thunderbolt 2 in 4K , Thnaks.