A Call For Measurements: Isolating the Thunderbolt Effect.  

Sorry for the confusion. I'm talking about the Gigabyte Thunderbolt 3 PCIe ad-on card.

Posted by: nando4

@itsage, nice work getting it all going.    Did you want to do a x2 PCH test as well (16Gbps)?  TB3 is 22Gbps so is slightly faster than x2 3.0 so we'd expect TB3 results to be slightly better by comparison *if* TB3 latency is also not a performance detractor. The way to force the card to run at x2 is to isolate lanes 3-4 using stick tape.  Only need to do it on one side.  Confirm x2 3.0 operation by using GPU-Z's render test "?" question mark next to the Bus Interface .

http://cfile3.uf.tistory.com/image/1668D21E4A793AA1229E75

If x2 3.0 is more reflective of TB3 performance then the card (taped to run x2) can be installed in a CPU attached PCIe slot so can get x2 PCIe-CPU performance.

Maybe is just stupid, but, what about doing it at x1?? Because TB1 is 10Gbps and is should be a good comparative too!!

Cheers

Here are the results of this test bench when ran with an Nvidia GTX 980 Ti. Razer Core was the enclosure for eGPU.

This is excellent work, itsage. Thank you for going through all the headache to get us these results!

Posted by: Yukikaze

This is excellent work, itsage. Thank you for going through all the headache to get us these results!

I'm always up for a good challenge. Thank you again for the fun idea!

Thank you very much for the 1440p and 4k benchmarks, the results confirmed our conclusion that if you up the resolution the performance drop is lower. I think after these results finally the eGPU haters will shut up, because for 1080p we have 15-20% performance drop for 1440p is 8% and for 4k the eGPU performance is the same like Desktop PC eGPU is the future even Apple knows that now! Forza eGPU 😉

Posted by: nando4

@itsage, nice work getting it all going.    Did you want to do a x2 PCH test as well (16Gbps)?  TB3 is 22Gbps so is slightly faster than x2 3.0 so we'd expect TB3 results to be slightly better by comparison *if* TB3 latency is also not a performance detractor. The way to force the card to run at x2 is to isolate lanes 3-4 using stick tape.  Only need to do it on one side.  Confirm x2 3.0 operation by using GPU-Z's render test "?" question mark next to the Bus Interface .

http://cfile3.uf.tistory.com/image/1668D21E4A793AA1229E75

If x2 3.0 is more reflective of TB3 performance then the card (taped to run x2) can be installed in a CPU attached PCIe slot so can get x2 PCIe-CPU performance.

Thank you for these instructions @Nando4. I'm running an RX 580 at x2 3.0 speed in the top PCIe slot of this Test Bench. I will have the results in x16, x4, x2, and x1 in the next few days.

Posted by: enjoy

Thank you very much for the 1440p and 4k benchmarks, the results confirmed our conclusion that if you up the resolution the performance drop is lower. I think after these results finally the eGPU haters will shut up, because for 1080p we have 15-20% performance drop for 1440p is 8% and for 4k the eGPU performance is the same like Desktop PC eGPU is the future even Apple knows that now! Forza eGPU 😉

3840x2160 pixels * 26 FPS = 215 million pixels per second

1920x1080 pixels * 60 FPS = 124 million pixels per second

but the bottleneck is actually happening at the lower pixels/second case! 

so there is more data overhead in a "frame" getting sent over the link than in just the raw number of pixels. Since we are only seeing the difference at high FPS and not so much at low FPS.

which is a pretty great problem to have. it means when you're pushing the performance of your card, you don't lose many FPS from the TB3 link when it really counts. When you are up around 60FPS you can afford to lose 5-10FPS and not notice as much.

I followed Nando's recommendation to tape the GPU contact pins so that I could run it as dGPU on the Z170 Test Bench at different PCIe speeds. It was an interesting experiment. I used GPU-Z to confirm the card was running at the speed I wanted. Invisible tape was very handy and made for an easy cleanup, just in case you want to replicate this test yourself.

Here are the results of the Radeon RX 580 ran as dGPU (top PCIe slot) at x16, x4, x2, and x1. I'm also adding the x4 TB3 eGPU results for quick reference.

Was the eGPU attached to the same system? If yes, that is quite a bit of performance drop for the TB3 when compared to the x4 on a PCIE slot...