2018 12" Lenovo ThinkPad X280 [8th,4C,U] + RTX 3090 @ 32Gbps-TB3 (Razer Core X) + Linux Ubuntu 20.04 [lessse]
- 2018 12.5'' Lenovo ThinkPad X280
- Intel Core i7-8650U 1.9- 4.2Ghz 4c/8t
- 16GB DDR4 2400
- Intel HD 620 iGPU
- Lenovo ThinkVision panel
- 500GB SSD
- Dual boot: Ubuntu 20.04 + Windows 10 (for now eGPU just working on Ubuntu 20.04)
- Razer Core X - eGPU box
- Graphics card: PNY GeForce RTX 3090 24GB XLR8 Gaming REVEL EPIC-X RGB Triple Fan Edition
1. Add nvidia repo
2. Install nvidia-460 drivers
3. Disable secure boot and TB3 security in BIOS
5. To check if everything is ok, run
(For TensorFlow with eGPU)
8. Use TensorFlow 2.4.1
Q1: I'm getting "ERROR: NVIDIA driver is not loaded" error, what should I do?
A: Check if your thunderbolt port is ok (Q6) and drivers are loaded (Q2).
Q2: How to know if my Nvidia Drivers are loaded and how to load them manually?
Q3: I'm getting "modprobe nvidia throws error: could not insert 'off'" error when doing Q2, what should I do?
Q4: I'm not sure I can reach the GPU?
A: To check if you see you GPU in the result of any of the following commands:
Q5: I'm not sure my usb-c supports TB3?
Q6: My laptop support TB3 but I don't know if its working?
A: Try plugin it in and out and check if the output of the following command changes
Q7: I'm getting "Could not load dynamic library 'libcudart.so.11.0'" error or cudnn library error, what should I do?
A: Check if you installed correctly the cuda and cudnn libraries and check os.environ inside python to check if your LD_LIBRARY_PATH is set to the directories at run time.
- update your OS
- when installing and adding repos to apt, check if your repos in apt are ok (some of apt functionality such as adding keys don't work as explained around the web if you have unsigned repos);
- check your firmware (fwupdmgr get-devices);
- create a bug report (nvidia-bug-report.sh) and publish it on Nvidia forum;
No benchmark for now 🙁 Sorry! I'll update these later
I'm using it for deep learning training, so if you're looking for a solution like this for training your models, it works.
@luis_espirito_santo, What an excellent build and clear troubleshooting steps to get the eGPU working in Linux!
@luis_espirito_santo, I have a similar setup having just secured an EVGA 3090 XC3 for my deep learning projects. I have a peculiar issue whereby everything works great (card detected, external display, and tensorflow) until I shut the computer down. When the computer is off or asleep and the Razer Core is connected to my laptop the GPU fans spin at max. This happens during Windows and Linux boots. Do you have any similar issues?
@drsteve, I had no such issue. My box (fans and lights) just stops whenever I disconnect it from the laptop. It does exactly the same thing when I shut it down or when I suspend it.
TBH I have no clue on how to help you.
Do you have any other hint? Did you follow the same steps as I did?
Thanks @luis_espirito_santo. My steps are slightly different (but more or less identical)--I use Docker + Tensorflow to avoid the headache of dealing with CUDA updates. After further online searching it does seem specific to EVGA cards (there is a subreddit about it) and there is no known work around. I am hoping a critical mass of eGPU + EVGA cards will compel EVGA to issue a firmware update.