RunPod multi-GPU pods fail to establish SSH

Problem

When launching multi-GPU pods on RunPod using runpod-dev, SSH connections fail to establish. Single-GPU pods work fine, but 2+ GPU configurations never become SSH-accessible.

Environment

• runpod-dev version: latest
• Tested GPUs: H100 (2x), RTX 4090 (2x)
• Pod image: runpod/pytorch:1.0.3-cu1290-torch260-ubuntu2404

Steps to Reproduce

# This works (1 GPU):
runpod-dev launch test --gpu H100 -n 1

# This fails (2 GPUs):
runpod-dev launch test --gpu RTX4090 -n 2

Observed Behavior

1. Pod is created successfully (API returns pod ID)
2. Pod shows RUNNING status in runpod-dev list
3. SSH column shows - (no SSH info available)
4. runpod-dev connect <name> returns "No SSH info available"
5. Direct gateway SSH (ssh <podHostId>@ssh.runpod.io) returns "Your SSH client doesn't support PTY"
6. Pod never becomes "ready" within timeout (300s)

Expected Behavior

Multi-GPU pods should have SSH access like single-GPU pods.

Workaround Needed

We need multi-GPU SSH access to test tensor parallelism across multiple GPUs.

Additional Context

This is blocking our tensor parallelism testing for BitNet training. Single-GPU pods with SSH work correctly - the issue appears specific to multi-GPU configurations on RunPod's platform.