How It Works
From gpu-dev reserve to SSH-ready GPU pod in ~15 seconds
HTTPS triggers read/write creates pods SSH keys volumes mount
>_
CLI + SDK
pip install gpu-dev
API Gateway
REST API
SQS
reservation queue
λ
Reserve
create pods + attach disks
DynamoDB
state + history
λ
Expiry
cleanup + snapshot
⏰ scheduled
λ
Availability
GPU counts + queue
⏰ scheduled
GitHub
SSH keys + org auth
EBS Snapshots
persistent user disks
EFS
/shared · /cache · 20 TB
⎈ Amazon EKS Cluster
GPU Operator
device plugin + DCGM
Prometheus
+ Grafana
CoreDNS
pod networking
Docker Images
PyTorch · Claude Code · Jupyter · custom
ASG per type
R
Capacity Res.
H100
p5.48xl · 8×80GB · NVLink+EFA
B200
p6-b200 · 8×192GB · NVLink+EFA
H200
p5e · 8×141GB · NVLink+EFA
RTX PRO 6000
g7e · 4×48GB · PCIe
A100/T4/L4
Budget · 1-8 GPUs
CPU
ARM + x86
GPU Pod
🔐
SSH server
🎮
nvidia.com/gpu
💾
EBS /home/dev
📁
EFS /shared
🚀
EFS /cache
📓
Jupyter
🔬
ncu / nsys
SSH Access
podname.devservers.io
Route53
DNS resolution
NodePort
:30000-32767
→ Pod
SSH session ready