I tried launching a GPU instance with NVIDIA Brev CLI
This page has been translated by machine translation. View original
Introduction
There are many things you want to try using a GPU, like local LLMs and image generation, right?
...But you can't just buy a dedicated machine right away, and shared machines are fully booked.
In that case, let's easily try out a GPU environment in the cloud!
This time, I'd like to get started with the NVIDIA Brev CLI.
Referencing the official Quickstart, I'll try creating a GPU instance and getting to the point where I can see the GPU with nvidia-smi.
Environment
| Item | Details |
|---|---|
| OS | macOS 26.5.1 |
| Package Manager | Homebrew |
| Brev CLI | v0.6.328 |
Steps
1. Sign up for Brev
Create a Brev account from the following link.
2. Install the Brev CLI
Install the Brev CLI with the following command.
brew install brevdev/homebrew-brev/brev
Let's check the version.
% brev --version
Current Version: v0.6.328
Log in.
% brev login
▸ Starting Login
You can log in by entering the email address registered with your account.
3. Search for available GPU instance types
With Brev, you can search for available GPU instance types using the brev search command.
This time, since I want to use something with a short startup time and low cost, I'll specify the following options.
brev search --max-boot-time 3 --sort price
--max-boot-time 3 → Narrow down to instances with an estimated boot time of 3 minutes or less
--sort price → Sort the search results by price
Let's look at the search results.
% brev search --max-boot-time 3 --sort price
TYPE PROVIDER GPU COUNT VRAM/GPU TOTAL VRAM CAPABILITY DISK $/GB/MO BOOT FEATURES VCPUS $/HR
verda_V100 verda:shadeform V100 1 16 GB 16 GB 7.0 250GB - 2m30s - 6 $0.47
hyperstack_A6000 hyperstack:shadeform A6000 1 48 GB 48 GB 8.6 100GB - 3m - 28 $0.60
vultr_A16 vultr:shadeform A16 1 16 GB 16 GB 8.6 350GB - 3m - 6 $0.61
massedcompute_A6000_plus massedcompute:shadeform A6000 1 48 GB 48 GB 8.6 256GB - 2m30s - 12 $0.68
massedcompute_A6000 massedcompute:shadeform A6000 1 48 GB 48 GB 8.6 256GB - 3m - 6 $0.68
excesssupply_RTX4090 excesssupply:shadeform RTX4090 1 24 GB 24 GB 8.9 850GB - 2m30s - 12 $0.72
excesssupply_RTX5090 excesssupply:shadeform RTX5090 1 32 GB 32 GB 10.0 900GB - 2m - 12 $0.78
massedcompute_L40S massedcompute:shadeform L40S 1 48 GB 48 GB 8.9 625GB - 2m30s - 12 $1.06
scaleway_L4 scaleway:shadeform L4 1 24 GB 24 GB 8.9 500GB - 2m - 8 $1.14
massedcompute_RTX6000Ada massedcompute:shadeform RTX6000Ada 1 48 GB 48 GB 8.9 350GB - 2m30s - 12 $1.16
hyperstack_A6000x2 hyperstack:shadeform A6000 2 48 GB 96 GB 8.6 300GB - 3m - 60 $1.20
verda_A6000 verda:shadeform A6000 1 48 GB 48 GB 8.6 250GB - 3m - 10 $1.30
Since I'm only confirming that the GPU instance starts up this time, I'd like to launch the low-cost, short-boot-time TYPE verda_V100.
4. Create and launch a GPU instance
The Quickstart introduces the % brev create my-instance --gpu "nebius.l40sx1.pcie" command, but the --gpu option was not available in the local Brev CLI v0.6.328.
% brev create my-instance --gpu "verda_V100"
unknown flag: --gpu
So, I'll check brev search --help and try using the --type option.
% brev create my-instance --type "verda_V100"
Attempting to create 1 instance(s) with 1 parallel attempts
Instance types to try: verda_V100
Trying verda_V100 for 1 instance(s)...
[Worker 1] Trying verda_V100 for instance 'my-instance'...
[Worker 1] verda_V100 Success! Created instance 'my-instance'
Waiting for instance(s) to be ready...
You can safely ctrl+c to exit
my-instance: Ready
Successfully created 1 instance(s)!
Instance: my-instance
ID: nrn519qmd
Type: verda_V100
Connect:
brev open my-instance
brev shell my-instance
Although the BOOT column showed 2 minutes 30 seconds, it actually completed in about 3 minutes 30 seconds.
5. Connect to the GPU instance
Let's connect to the launched instance using the shell command.
% brev shell my-instance
⣟ waiting for SSH connection to be available Agent pid 18795
shadeform@brev-nrn519qmd:~$ nvidia-smi
Fri Jul 3 05:12:52 2026
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.126.09 Driver Version: 580.126.09 CUDA Version: 13.0 |
+-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 Tesla V100-SXM2-16GB On | 00000000:05:00.0 Off | Off |
| N/A 30C P0 19W / 300W | 0MiB / 16384MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| No running processes found |
+-----------------------------------------------------------------------------------------+
I was able to connect!
Now that I've confirmed it, I'll disconnect using the exit command.
$ exit
logout
Shared connection to xxx.xxx.xxx.xxx closed.
You can use the ls command to check running instances.
% brev ls
You have 1 instances in Org classmethod
NAME STATUS BUILD SHELL ID MACHINE GPU
my-instance RUNNING COMPLETED READY nrn519qmd verda_V100 -
6. Cleanup
Once an instance is created, you will be billed even if you are not connected to it.
When you are done with what you wanted to do, make sure to stop or delete it.
Since the instance I selected does not support brev stop, I will delete it.
% brev stop my-instance
1 error occurred:
* instance "my-instance" does not support stop.
% brev delete my-instance
Deleting instance my-instance. This can take a few minutes. Run 'brev ls' to check status
The instance was deleted in about 3 minutes.
% brev ls
No instances in org <ORG_NAME>
See teammates' instances:
brev ls --all
Impressions
Since you can easily try out GPU instances, I think it's useful for research and verification before purchasing a device.
Next, I'd like to try connecting via SSH from VSCode and experimenting with the GPU environment.