I suspect this might be a question for the NVIDIA engineers, but will the DGX Spark allow you to connect thunderbolt eGPUs? I have an old enclosure lying around, and I’m thinking it’ll be cool to connect my 5090 to the DGX Spark. If I did that, I could actually retire my current AI box, as everything could just run on the DGX Spark, and I can decide whether to run/train on the internal GPU or the faster 5090. In short, is this possible?
I’m also interested in whether or not this is possible. I have two 5090s, and I’ve been considering moving one out of my system to an external enclosure if this scenario works.
I’d imagine it’s dependent on the driver support on DGX OS, and what generation the USB4 ports on the Spark may be.
Crossing my fingers for good luck.
1 Like
Hi, external GPU enclosures aren’t officially supported for Spark or GeForce. However, if you encounter issues, please share what you’re seeing; we’d like to better understand the errors.
Thanks for responding, to confirm, has someone tried it and is it expected to work? I’m still trying to make the buy/don’t buy decision, and since we only have a couple more days to decide, I was hoping for a more definitive answer to this. It just seems like an obvious thing people would want to do, turning the DGX Spark into their complete home AI solution with their additional GPUs connected it.
So it looks like the 3 USB-C ports in the rear of the Spark are USB-C 3.2, based on the “Getting Started with the NVIDIA DGX Spark” video on the NVIDIA Developer Youtube channel (Getting Started with the NVIDIA DGX Spark).
I guess this means using an eGPU is likely not possible. Oh well…
All the PCIE lanes got redirected towards the ConnectX7 I guess
Biggest bummer is the port selection. No thunderbolt makes fast LAN connection basically only over ethernet.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.