GitHub Community
Why is my Azure D3 VM so slow for AI work, and why do I keep having connection/IP issues? #180707
-
Select Topic AreaQuestion BodyI’m using an Azure D3 VM for some small AI/ML and web dev work. I’m facing two issues: Sometimes I can’t connect to the VM (SSH/RDP) even though I’m using the correct public IP. When I try to run small GPU/ML workloads, it’s extremely slow or fails. I’m a bit confused about whether D3 is good enough for this, and how to make the connection more stable. What’s the best way to: Fix the connection / IP issues Understand the limits of a D3 VM and what I should use instead for GPU work |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
|
A D3 VM isn’t designed for AI/ML work - it has no GPU, so anything involving PyTorch or TensorFlow will run slowly or fail. It’s mainly for basic CPU workloads. If you’re having trouble connecting (SSH/RDP), it’s usually because the VM is using a dynamic public IP, the VM was deallocated, or the NSG rules for ports 22/3389 aren’t open. Switching to a static IP and confirming your NSG rules usually fixes it. For actual GPU work, you’ll need a different VM type such as the NC or ND series, or NCasT4_v3 if you want a cheaper option. These are built specifically for ML workloads and will perform far better. |
Beta Was this translation helpful? Give feedback.
A D3 VM isn’t designed for AI/ML work - it has no GPU, so anything involving PyTorch or TensorFlow will run slowly or fail. It’s mainly for basic CPU workloads.
If you’re having trouble connecting (SSH/RDP), it’s usually because the VM is using a dynamic public IP, the VM was deallocated, or the NSG rules for ports 22/3389 aren’t open. Switching to a static IP and confirming your NSG rules usually fixes it.
For actual GPU work, you’ll need a different VM type such as the NC or ND series, or NCasT4_v3 if you want a cheaper option. These are built specifically for ML workloads and will perform far better.