Cool setup. I’m also running an eGPU but with an enclosure I’ve crammed a little water-cooling setup into, am using the USB-C expansion card, and quite a long cable running from on top of a desk to under a desk.
The Iris iGPU seems to interfere when I choose to only display on an external monitor (and/or close the laptop lid). I then get stuttering on the Win10 desktop and some issues with full-screen applications.
I tried disabling the iGPU in device manager and it fixed it - but the integrated audio device (“High Definition Audio Device”) also dissapears so I’m left without audio.
I was thinking to use it as an add on encoding card, something like how you use a capture card/coprocessor, rather than for display purposes, so not sure if it can be done.
I’m assuming this means use the eGPU without a display attached, so not displaying anything on a monitor from the GPU but still having it available for compute tasks. This is absolutely possible; once you install the drivers the card will show up regardless of whether it has a monitor attached. In the program you’re using for encoding/decoding/rendering/etc, you can typically select what GPU you want the task to run on, and that’s where you’ll select the eGPU.
You can also make the eGPU offload specific programs. In Linux this is “PRIME render offloading,” not sure what the windows equivalent is (I’d assume using the nvidia control panel to specify which card to run it with).
This comes with the caveat that you have to go back and forth on a single cable, so your performance is theoretically half of what you’d get on an external monitor (in practice it’s not that dramatic a drop, see benchmarks in this thread)
Another funny question here, has anyone used this eGPU with a standard ATX power supply?
If so, how is the power connectors connected? From some photos, it seems like the 24pin and 4 pin motherboard connectors are connected to the R43G board, then from the R43G board, you run the included PCIe splitter cable to the GPU?
It should, since it’s effectively a Thunderbolt PCIe bridge
How exactly said PCIe device behave when being attached via such a bridge, not enturely sure.
Depend on the device you might be able to get a USB-PCIe bridge really cheap instead.
They are usually in the form of NVMe drive … things. So you will need some adapter-finagling.
It might also be cheaper to just get a USB capture card. a Thunderbolt PCIe bridge is really expensive ($150), because Intel’s small volume
Unless you have multiple diferent PCIe cards and would like to swap them around.
Yup, I would get the Thunderbolt 4 version of the R43SG, probably connect it to a Thunderbolt 4 NVMe drive enclosure.
I am not capturing video, just using hardware accelerated video encoding so I would need a GPU for it (i.e. Intel A380 AV1 Quicksync).
The question I have is, how does the ADT-Link power connectors work with an ATX power supply?
I saw on the sellers site that the ADT-Link is powered by the 24-pin+4pin connector of the ATX power supply and the ADT-Link handles the 12V pass through to the GPU via the provided 8pin PCIe power cable with splitter.
Is that correct? Or does the GPU require its own dedicated 8pin PCIe power cable? I do not think that the ADT-Link requires that much power (at most 75W for the PCIe slot power), the 24pin+4pin is used to power the mainboard and CPU (which can be as high as a GPU in power draw).
Basically, the Thunderbolt can only support 45W (varies) to any downstream device. So usually they will require their own power supply.
In addition to the PCIe bridge you will also need to plug in any PCIe ports found on your GPU. You can do that via an external power supply (a spare desktop PSU is fine; just make sure to bridge the “sense” pin so the power supply will turn on).
From the above pictures of the ADT link you are to plug in an entire external ATX 24, as well as CPU 4+8 pin. The 8 pin from the PCB then go to your graphics card (?)
Alternatively just plug in ATX power from your external supply direct.
Bumping this thread to ask for recommendations on a m.2 to usb-A/C converter. I am trying to use the same setup as OP on a linux machine (running ubuntu 22.04). running dmesg when plugging into usb-c via this connection shows a mass-storage device plugged in instead of a gpu, and so nvidia-smi does not register even when drivers are installed.
Well it does use a controller explicitly for mass storage. You’ll need something with actual pcie tunneling if you want to plug a gpu into it but you might as well use a dedicated egpu board instead.
