RTX 3080 in an eGPU discussion

Moving this here from a different thread since it's eGPU related.

Here's someone that is compiling egpu benchmarks using the same CPU I am (i7-8565U) compared to a desktop CPU
It's pretty impressive for what the setup is. There is also results from 9980HK laptop CPU.

https://egpu.io/forums/postid/86664/

I generated a table from the results that were in text. Some of these are a single desktop result, some are desktop average from 3dmark. I don't know under what condition the eGPU was run, but it will be faster on an external monitor hooked to the GPU vs Internal laptop panel.

My 2070 is definitely faster than a 1060 in a desktop.. The higher you go in resolution using an eGPU, the more GPU bound you are, and the less the difference between a desktop build is vs. eGPU build. The difference is more massive at lower resolutions and as well some games just use more PCIe bandwidth than others. Overall I'm very happy even with the 2070 build. I have no idea why this forum software is putting this table so far below this text.. I'm starting the table on the next line.. SHRUG.. sorry..

EDIT: figured it out, forum software doesn't like pretty formatted tables it retains the whitespace, they should fix that..

Benchmark	Desktop CPU 3080	i7-8565U eGPU 3080	% difference
4k Super Position	14,248	13, 391	-6%
1080P Extreme Super Position	10,920	10,389	-5%
Time Spy Extreme (4K)	8,898	8,194	-8%
Port Royal (RTX)	11,229	8,571	-24%
Firestrike (1080p)	42,254	28,000	-34%

So at the end of the day, it looks like if I put a 3080 in my eGPU.. it would be at minimum about the same as a desktop system with a 2080Ti, and it some cases it would beat it by some margin.. which is pretty damn impressive..

@JRE84

 

hey thats pretty cool thanks for posting that...Looks like i'll be getting a 3080 egpu...hopefully consoles don't kill the card with super demanding ports

 

But I don't want to use an external monitor.... I wouldn't mind yo use a TB eGPU with internal screen, but external screen no way. What performance improvement should I get with an external 3080 + internal screen vs a single internal 1080?

 

This matches up with data from my Titan Xp running in AW 13 R3, Area-51m, and desktop. Difference was like 2-3%. 3080 being faster shows a 6-8% difference with the TB3 based eGPU.

I did use Alienware graphics amplifier though. Up to 3090 at 4K should provide minimal bottleneck (~10%?) so should be faster than any laptop gpu.

1440p/1080p is not worth it for 3080+ unless you're running a 1660Ti or other low end gpu in the laptop.

 

i might be running my 3080 for VR and maybe a external monitor and use my laptop like a desktop...funny thing is its a thin and light gaming laptop lol....so incredibly pointless going thin and light with a good gpu

 

Not entirely pointless, since you can still use the internal dGPU with eGPU active, plus you still get that mobility.

 

I use mine without an external monitor, works great

 

What's the performance percentage you get using internal monitor compared to the full desktop version??? A MXM mobile RTX 2080 is around $ 1500 (which is ridiculous) for much lower price I would be able to get a desktop 3080 but I want to be sure that's actually a better deal in terms of performance/$

 

I would do some research on eGPU.io, they have a ton of builds you can look at, most people post benchmarks.

Typical perf loss is around 20% but it varies depending on the game, resolution, etc.. usually the more GPU bound the less the performance loss. TB3 has its limits. I think it works great though, fits my needs perfectly, definitely not for everyone.

 

Man in this review:

the results are not pretty good. At 1080p the difference between internal vs external monitor is not that high, but internal monitor vs full desktop most of the time the eGPU gets half performance. At 4k the gap is smaller, but I'm not going to play at 4k and I rather have high frame rates at 1080p

 

Something smells off about that, Control was running much better for me with RTX on High than he is showing there, and I have a theoretically slower CPU and GPU.. some of his numbers seemed off. I don't really feel like redownloading Control and Doom: Eternal (the only two games I own that he used) to refute those scores.. but something seems off there. Unless I'm just happy with it and not very sensitive to it.

 

This one here seems pretty close to what I'm trying to do:



Though the 3DMark test is pretty good on the eGPU compared to the mobile 1080, while gaming the eGPU RTX 2080 Ti is just not as good most of the time compared to the GTX 1080. Which is pretty sad. How is that even possible?

 

TB3 is slow compared to PCIe x16

 

Oh yeah, so we need to see in game performance of the eGPU 3080 and not being so excited by synthetic benchmarks, which in this case not always seems to translate in to better performance

 

I honestly would look forwards to an eGPU with an 2080 Ti or 2080 due to the lower TDP of 250W.

 

Those cards aren't a big enough jump for me. I just want to try to lock every game at 60 with full details and RTX. That isn't happening right now.

 

just because it says ultra doesnt mean your getting more details.ultra is way way over the top with the fact low medium and some high looks just as good

 

BTW Most laptops and even desktops only use 8 PCIe lanes. In games it all depends on how much data needs to be constantly sent to the GPU. If there is a lot of texture streaming then TB3 could bottleneck, so in that case the 3090 should be the answer with its 24GB of VRAM.

 

I stopped watching after he explained his garbage "methodology".

 

The performance uplift of having the TB controller directly connected to the CPU is quite frankly massive. Ice Lake and Tiger Lake U series have been the only CPUs with this configuration so far. The Intel H35 chips starting to come out now (i7-113xxH series) have this benefit as well. The good news is that it appears based on Intel's recent slides and some articles that nearly all the upcoming 11th Gen H parts are going to have this scenario. The i5-11400H and i7-11800H are the ones I've seen reports of, if that's true I would imagine there will be an embedded TB4 controller up and down the entire H line. I think the 6c/12t i5-11400H is going to be a sweet place to land for eGPU gaming this year.

If I was in the market for build an eGPU system and wanted something a little better than an i7-1165G7 or i7-1185G7 (make sure the notebook is using 28W!), or even i7-11370H (they are all 4c/8t) I would wait until later this year and see what good Rocket Lake S parts are going to appear. The external TB controller for the 10th Gen H parts is a huge performance hinderance.

For some proof, watch this Ice Lake i7-1065G7 (28W 4c/8t) in a Razer Blade Stealth 13 stomp all over a laptop with an i7-10750H (45W 6c/12t) in eGPU gaming with a 3090. It's because of the direct-to-CPU TB controller in Ice Lake U vs external controller in 10th Gen H.

 

any verdict on tb 4