Initial RTX 30 Series eGPU Benchmarks Look Promising for Bottleneck Concerns
Last Updated: 9/24/20 - 08:25AM EST
I compiled some 3DMark data on generational differences of Nvidia cards (desktop counterpart vs in an eGPU setup). Previous gens were all compared to my current MBP (i9 9980HK). I was only able to find one 3DMark user running an eGPU setup so far, so take this with a grain of salt, but this user was running a relatively underpowered CPU so I'm quite hopeful for higher end CPU specs. Also, TS is only 1440p so TSE @4k would most likely close the gap even further. Take a look:
Time Spy (1440p) Graphics Score
Average of All Desktop
9980HK MBP w/eGPU
Edit 1: lol I should have known! The 'anonymous' 3DMark user I was using to scrape the 3DMark eGPU scores from has posted his build to these forums: https://egpu.io/forums/postid/86633/
Additional Benchmark Comparisons:
Super Position (4k Optimized)
Desktop 3080: 14,248 ( https://tinyurl.com/y4bm7zaw)
eGPU 3080: 13, 391 ( https://egpu.io/forums/postid/86633/)
% Difference: -6%
Super Position (1080P Extreme)
Desktop 3080: 10,920
eGPU 3080: 10,389
% Difference: -5%
Time Spy Extreme (4K)
Desktop 3080 Average Graphics: 8,868
eGPU 3080: 8,234
% Difference: -7%
Port Royal (Ray Tracing)
Desktop 3080: 11,330
eGPU 3089: 8,873
% Difference: -22%
Desktop 3080: 42,124
eGPU 3080: 28,000
% Difference -33%
Firestrike Extreme (1440p)
Desktop 3080: 21,002
eGPU 3080: 15,213
% Difference: -28%
Firestrike Ultra (4K)
Desktop 3080: 10,775
eGPU 3080: 9,496
% Difference: -12%
I'm continuing to update these numbers as new scores come in to 3DMark (see last updated time at top of post). As I predicted, the average desktop scores are slowly decreasing as the numbers get populated by non-enthusiast early adopters. Except, notably, the Port Royal score which went up after Gamers Nexus used LN2 to beat the all-time Port Royal score the other night, after which a Brazilian team further raised the world record. To be transparent, I am cherry-picking the 'top' scores for the eGPU numbers while using the average for desktop, as I want to give a best case scenario while also being fair by not picking the 'worst' desktop scores. I have no doubt we'll see the eGPU scores continue to trend up as more people get their hands on the 3080 and optimize their setups, and the desktop scores to come down, and so I expect the % differences to reduce in the coming days and weeks.
@arizor, Can't wait to see them! I'm honestly astounded at these initial results considering they were all done on a 15w TDP i7 8565U. What CPU will you be testing on?
@bmagnien, I ran the RTX 2080 Ti Gaming Box in Time Spy Extreme, Fire Strike Ultra, and Port Royal in some of my builds. If you need to compile more data point, i7-1065G7 and i7-8569U numbers are in this build. At 4K and higher resolutions, eGPU performance loss is in single digit. This is a test in FHD, QHD, 4K, and 5K when I was comparing AGA x4 PCIe 3.0 vs TB3 x4 PCIe 3.0 [build link].
|AW15R3 + RTX2080||AGA FHD||TB3 FHD||AGA QHD||TB3 QHD||AGA 4K||TB3 4K||AGA 5K||TB3 5K|
|3DMark Time Spy||15,846||13,564||10,545||9,416||5,167||4,879||3,029||2,946|
|3DMark Fire Strike||25,428||20,452||18,342||13,675||9,898||7,694||5,129||4,785|
|Tomb Raider 2013||228.5 FPS||187.7 FPS||152.7 FPS||135.1 FPS||77.2 FPS||74.7 FPS||44.2 FPS||43.9 FPS|
|Shadow of Mordor||161.6 FPS||127.3 FPS||121.7 FPS||99.4 FPS||71.0 FPS||62.6 FPS||45.7 FPS||42.7 FPS|
|Dirt Rally||100.3 FPS||119.25 FPS||99.3 FPS||106.0 FPS||73.9 FPS||74.9 FPS||52.5 FPS||53.3 FPS|
|Hitman||86.3 FPS||81.7 FPS||77.1 FPS||68.5 FPS||38.6 FPS||35.8 FPS||21.5 FPS||21.1 FPS|
Hello I see my benchmarks: D Great
my "slower" i7 8565U with TDP 15W; draws between 25-35 watts until the keyboard burns ...
therefore, without much better cooling, there is no enormous increase in performance for GPU-intensive operations; do you agree with me?
[email protected] and yes my laptop only has HD so all > have been made as 1920x1080 on my 4k television.
@arizor, I'm really excited for your results next week. I have the same MacBook and I get a TimeSpy graphics score of 13100 with my 2080 TI. It's good to see the eGPU bottleneck hasn't fully caught up yet.
@w7_basel, hey man I hope my 'slower' comment wasn't taken the wrong way. Compared to some of the 'gaming' laptops and of course desktop CPUs, it's at a disadvantage on paper. But that being said, your results are nothing short of impressive, so thank you for sharing. Even though you ran the tests on a 1080p internal and 4k external, most of these synthetic benchmarks run independently of the native resolution, so no worries there, your results are just as useful for us. Do you mind running a TimeSpy Extreme, a Port Royal, a FireStrike Extreme, and a FireStrike Ultra? All on an external display, in Full Screen mode?
@itsage, these are awesome man! I think what will be most enlightening moving forward are 3080/3080 desktops vs eGPU enclosure hard data. I'm still struggling to find a lot of that to draw better conclusions from. 3dMark only has 3080 data so far, and the people that are running those are probably mostly enthusiast early adopters so I would imagine the average will slowly come down from where it's at now as it's probably skewed high from overclockers paired with high-end CPU/Mem internals. And of course we only have the one data point on @w7_basel 's Gigabyte GeForce RTX 3080 GAMING OC eGPU. If you happen to come across any 3080 desktop 8k superposition scores let me know!
@arizor, that'll perform like a champ i bet. can't wait! I know @itsage has access to 10th gen intels so if he gets his hands on one of these new cards we'll see if that unlocks more potential from the 30 series. I'm hoping to be able to get a 3090 this coming week to test with my i9 9980hk.
@w7_basel, I can back up your point that the slower processor won't make a difference in GPU tasks.
I get the same TimeSpy graphics score on my i9-9880h even if I limit my CPU's TDP to 10W.
@w7_basel, hey I pulled your Time Spy Extreme, Port Royal, and Firestrike scores off 3DMark and added them to the main post. Could you confirm those scores are accurate and performed on an external display in full screen? And would you be able to add Firestrike Extreme and Firestrike Ultra when you get a chance? Thank you!