Connecting to GPU and TPU is bugged

[TariqNier]

here is what i do to use it:

Select Kernel
Pick Colab
Pick GPU
Pick T4
Im prompted to enter an alias, i press enter to continue to ignore
It says "Detecting Kernel" , here it should connect
It goes back to saying "Select Kernel" like i did absolutely nothing
but the CPU option works normally

[codeNinja62]

is your vscode up-to-date?

[TariqNier]

is your vscode up-to-date?

yea i just checked. i even tried cursor ide it has the same exact issue

[Priyansh4444]

Same issue here, but unsure if this is an issue with me going past the Free limit's quota.

[czhou578]

Yes. It doesn't clearly say on the UI whether I have exceeded the Free limit's quota.

[kevineger]

We warn when you're unable to assign another server (limit to 3) and need to delete one. We also have warnings when you eclipse your paid and free quota. These all should show as notifications. While it's possible there's some bug there, I'd like to rule out if you're hitting some other issue.

To best help, could you please provide the info as requested in the bug template: all the environment info, consistent repro steps and the Colab & Jupyter logs. If you can, bump the verbosity up for both (in VS Code settings) before repro'ing and collecting them.

[Hansehart]

I have the issue too. I guess I reached the limit. However I get no notification and I worked before. I am on WSL2, Ubuntu 24.04, steps you see on the video.

VSC:
Version: 1.106.3 (user setup)
Commit: bf9252a2fb45be6893dd8870c0bf37e2e1766d61
Date: 2025-11-25T22:28:18.024Z
Electron: 37.7.0
ElectronBuildId: 12781156
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Windows_NT x64 10.0.26100

I dont see any logs from the extensions

[kevineger]

@Hansehart you don't see any logs in Output > Colab nor Output > Jupyter?

[Hansehart]

@kevineger I am wrong! Woow, after years of vscode I found that view the first time. Thanks for that hint. As soon as I get the problem again, I will share my logs. Today it works again.

[Hansehart]

Jupyter:

Visual Studio Code (1.106.3, dev-container, desktop)
Jupyter Extension Version: 2025.9.1.
Python Extension Version: 2025.18.0.
Python Environment Extension Version: 1.12.0.
Pylance Extension Version: 2025.10.3.
Platform: linux (x64).
Home = /home/user
Temp Storage folder ~/.vscode-server/data/User/globalStorage/ms-toolsai.jupyter/version-2025.9.1
Workspace folder ~/projects/dna

Colab:

[2025-12-04T18:59:14.129Z] [Info] Visual Studio Code: 1.106.3
[2025-12-04T18:59:14.129Z] [Info] Remote: dev-container
[2025-12-04T18:59:14.129Z] [Info] App Host: desktop
[2025-12-04T18:59:14.129Z] [Info] Jupyter extension version: 2025.9.1

Behaivior is exactly from the video. CPU works, when using GPU you dont reach the option to choose a kernel.

[timettelt]

Hello, I come along with a similar problem, but I get this notification:

Even if I disconnect my older sessions, CPU is fine but for the GPU I still get the same message.

[kevineger]

@timettelt

Even if I disconnect my older sessions

Disconnecting from the session does not free up your quota. You need to Remove Server (either use the command or from the Colab icon in the toolbar).

---

@Hansehart

CPU works, when using GPU you dont reach the option to choose a kernel

Hmmm, I've not seen that before. Could you please set the log verbosity to Trace (under Colab settings) and pass along whatever's there. Hoping there's some more info to go off of.

[adamhadani]

FWIW I have the same issue. I noticed that if I startup VSCode (running latest version on MacOS), and try off the bat to open a notebook and connect to colab when choosing GPU etc., I get this behavior.

However, if I try first to use the "Auto connect" option, that works. Subsequently, trying to use the GPU option works as well (let me choose Python or julia kernel).

So it seems like something about choosing the "auto connect" (first option) initializes something that makes everything work afterwards, but trying to go for GPU first is missing some initial state perhaps?

Couldn't see any print outs in any output/problem views