GPU installed inside of desktop vs laptop/eGPU performance test

[pangaea]

Used an AMD 7950 GPU and Windows 10 for both setups

 

Desktop used:

-AMD FX-6 CPU

 

-12GB DDR3 RAM

 

-GPU installed internally on a cheap microatx motherboard

 

Laptop:

 

-Intel i7-4600M vPro CPU

 

-16GB DDR3 RAM

 

-GPU installed via an expresscard adapter

 

I ran a benchmark with uningine heaven and got the following results (higher is better)

 

Desktop

 

FPS: 29

 

Score: 730

 

Min FPS: 7.9

 

Max FPS: 62.9

 

eGPU

 

FPS: 25

 

Score: 630

 

Min FPS: 14.6

 

Max FPS: 50

 

I will also note that the eGPU ran ~20C cooler than installing it internally in the desktop due to it being in open air rather than inside of a case. For both tests I used DX11/ultra quality/extreme tessellation. I used the same 1080p monitor for both tests however I recommend watching out for the benchmark not "bleeding over" onto the laptop's integrated display as this will cause a giant unfair performance hit. When I first did the test I made this mistake with a small sliver showing on the laptop's integrated screen and it got a score of only 300.

 

In conclusion the internal install performed better but not by a big enough margin to make or break your average gaming experience.

[Ross Siggers]

This is interesting, I just made myself an eGPU setup, and was wondering how much performance I was losing by having expresscard as my interface. If you delve into the egpu forums there's quite a bit of discussion about bandwidth blah blah. But my laptop still actually moves places, so hardwiring something internally would have been less than elegant.

 

Question though, you're running an external monitor as well as the laptop's integrated screen, is that not bottlenecking your performance by sending a return signal to the laptop? I understnad the benchmark wasn't running on the internal display, but do you think you could liberate performance by simply disabling it in Windows? Admittedly this is not an ideal option for everyone. I have a Lenovo Thinkpad with a docking station, so when I'm at home, it lives there with the lid closed, and outputs via the gpu to a pair of 1080p screens.

Using a spare 2Gb HD6850, I managed to raise my Unigine: Superposition score from 572 to a surprising 2502 points. Somewhat respectable when you consider both the card and laptop are from 2012...Next step is to dig out a 4Gb GTX680 I have...somewhere...and build an enclosure. Given the specs of the laptop itself, I feel like any newer card would be edging into the territory of 'overkill' for the rest of the hardware;

 

Lenovo X230t

i5 3320m

Intel HD4000 graphics

16Gb ram

500Gb ssd

[pangaea]

50 minutes ago, Ross Siggers said:

This is interesting, I just made myself an eGPU setup, and was wondering how much performance I was losing by having expresscard as my interface. If you delve into the egpu forums there's quite a bit of discussion about bandwidth blah blah. But my laptop still actually moves places, so hardwiring something internally would have been less than elegant.

 

Question though, you're running an external monitor as well as the laptop's integrated screen, is that not bottlenecking your performance by sending a return signal to the laptop? I understnad the benchmark wasn't running on the internal display, but do you think you could liberate performance by simply disabling it in Windows? Admittedly this is not an ideal option for everyone. I have a Lenovo Thinkpad with a docking station, so when I'm at home, it lives there with the lid closed, and outputs via the gpu to a pair of 1080p screens.

Using a spare 2Gb HD6850, I managed to raise my Unigine: Superposition score from 572 to a surprising 2502 points. Somewhat respectable when you consider both the card and laptop are from 2012...Next step is to dig out a 4Gb GTX680 I have...somewhere...and build an enclosure. Given the specs of the laptop itself, I feel like any newer card would be edging into the territory of 'overkill' for the rest of the hardware;

 

Lenovo X230t

i5 3320m

Intel HD4000 graphics

16Gb ram

500Gb ssd

I did a few tests and it does seem to use the eGPU for all tasks regardless of what screen I run them on. The simple solution to that is to not use the laptop's screen for any GPU-intensive tasks but I would prefer that tasks be handled by the GPU connected to that screen for enhanced multitasking (though I suppose such a thing would ruin the seamless experience you get, being able to just drag things across screens). Many people will have a laptop with an internal GPU that's good enough for Youtube videos but not good enough to run their favorite game at 4K with ultra settings. There is definitely a huge performance hit if I run anything even remotely GPU intensive on the laptop's screen, as I said in my test the benchmark's screen having a sliver showing on the laptop's screen completely ruined performance all the way down to 1/7 of what I got on with the card being used normally on the desktop. After correcting this issue and using only the monitor connected to the eGPU, the performance hit was only ~13%, possibly a lot less due to the nature of the fact that I was using the laptop's CPU instead of the 6 core FX.