If you didn’t have a card that required Auxiliary power, I would say that there is the RocketStor 6661A Thunderbolt 3 enclosure. which is made primarily for NVMe PCIe expansion and other stuff, but I’ve already done testing using it’s older brother the 6361A Thunderbolt 2 enclosure with a single slot card that’s under 75W and it seemed to work alright.
I feel you. The Razer Core is big, stupid, and heavy. Currently in the middle of testing this one, listed in the Top 5 on egpu.io Even though it needs some heavy Dell charging brick, the whole setup fits in a backpack which is all I need
Seems promising - what GPU are you running with it? I have a spare 1070 laying around. And I’d love a metal case for it since it would primarily be for travel (LAN, hotel room, etc.)
Using a 1060. Yeah another downside is that there’s no case for it since it’s just your GPU sitting in the open air on your desk. I bought some large anti-static bags to transport everything in travel as my own workaround
Good review, honestly. You mentioned many important points about eGPUs.
I would have liked to maybe hear some advice about properly dimensioning power delivery with eGPU setups. I feel like many make the mistake to only look at the average power consumption of GPU and forget to consider power spikes when buying PSUs for their eGPU setup. This can cause the over current protecting to trigger very fast when a 10ms+ power spikes comes that is 50% over the average TDP of the GPU.
With the F1 2020 frametimes i feel like the bandwidth issue was the bottleneck and not the cpu. Otherwise the iGPUs frametimes should be lower than the frametimes of the eGPU.
This is just meant as some constructive critic and advice, i liked the video overall
The implementation of the Ego game engine used by F1 2020 is largely physics based, so with the relatively easily rendered basic textures, and short draw distances the game is very CPU bound. This is why the 1% lows dropped so significantly with the eGPU. The GTX 1660, even with it’s reduced bandwidth could render frames very quickly, but had to wait on frame calculations from the CPU.
Still i’ve never seen this kind of behaviour before. Normally a cpu bottleneck causes the fps to not go higher at some point, no matter how much GPU performance you have. But here we actually see a decrease in frametimes for some reason instead. It might just be a bad programmed engine
Hey all! Thanks for all the great info and input. So glad this thread could become that and I love seeing all that has been discussed. You’ve all certainly helped me out!
I am running a Sonnet Breakaway Box 750ex with a RTX 3060ti. I have a HP Omen 27i 1440p monitor connected to the eGPU directly. I am able to play modern titles like Battlefield V and Battlefield 2042 with reasonable frame rates (~60-75 fps) on the laptop though it absolutely pegs the CPU. I use this setup for my docking station at an apartment I work out of from time to time. Other great features I have found is that this also provides support for NVidia Broadcast on my work laptop which is great for conferencing.
I have found one game (Far Cry 5) really uses the i7 in my framework laptop and lowering setting and adding additional cooling (cooling pad, lid open) has had no effect whatsoever. I have determined that this is due to DRM (anti-cheat) software running with that game.
The laptop and eGPU (same Breakaway box by the way, good choice!) runs games just fine but the anti-cheat component must not be optimized very well. I mention this because I believe both those Battlefield games use anti-cheat.
Interesting. Right now i’m waiting on a response from @FrameworkSupport on a CPU cooler issue but am looking forward to getting back to some eGPU testing.
Is there a full bandwidth low latency PCIe dongle for the frame.work? Would be great to be able to use the best eGPUs. Not sure what the ideal dongle / interface is, but the new ASUS ROG laptop claims to have it.
@John, such interfaces would require exposing some of the PCIe lanes to the outside world. This has been tried for over a decade with limited results. The problem has always been that the cable and connectors are just to expensive.
@Shawn_Lewis Based on your other comments in this thread, I don’t think you have the expertise to conclude that. This seems like an ideal opportunity to get that right. Especially considering the modules architecture of the frame.work. For the next iteration of the laptop an actual PCIe socket could be exposed inside behind a dongle socket that could be exposed outside. I’m sure someone could think of a better way to do it than myself - and it would defaultly be more foolproof than all other current manufacturers.
@John, fair enough, however, I’ve been in IT for a long time and I’ve seen multiple attempts at external PCIe solutions. It’s usually the cable that ends up being difficult to make and quality control that has limited such advancement over the past 20 years. It was only when LightPeak was developed, that even the possibility of an external GPU was relized.
A cable stopped advancement for 20 years? Amazing they were able to make a connection with PCIe internally yet not externally, despite the fact that you can extend PCIe slots using cables!
Has anyone actually got ‘usable’ performance with an eGPU on Linux on the fw? I’m really struggling with the setup I’ve got - the desktop grinds along at about 10fps and games are just unplayable at 1fps. (Considerably worse than the iGPU)