All You Need to Know to Build Your GPU Machine Learning Cloud [B]
summary
Built a private GPU ML cloud that provisions containerized jobs in seconds and raises utilization by moving from shared workstations to Kubernetes + Docker with GPU awareness, host networking and a portal wizard that spins up distributed TensorFlow runs (example: 2 servers, 3 GPUs, 1 parameter server + 3 workers). Data served from Ceph RADOS exposed as volumes and an S3 API via RADOS + Minio, Jupyter for notebooks and TensorFlow Serving for model deployment; it works but faces kernel deadlock and TF‑serving restart issues.
AI-written summary · about
chapters (11)
- 0:00Start
- 2:54Google Driverless Car
- 5:00Hinton and yann LeCun
- 5:40The reason of rising
- 6:47Deep Leaming Accelerator - GPU
- 7:26Nvidia is eye-catching on the stock market
- 13:33First Phase Goal
- 15:04Tensorflow Community vs others
- 21:08Jupyter
- 26:19Tips about building your own GPU clouds
- 27:48More tips about building your own GPU deuts
related talks
32mGPU Machine Learning From Laptop to Cloud
31mKubernetes For GPU Powered Machine Learning Workloads In Academia
34mGPU as a Service Over K8s: Drive Productivity and Increase Utilization
24mDocker for Machine Learning!
06mKubernetes Love Machine Learning, Even on Private Cloud
37mBuilding Distributed TensorFlow Using Both GPU and CPU on Kubernetes [I]
34mDeep Customized Kubernetes for Machine Learning in Tencent
15mTransparent GPU Sharing in Container Clouds for Deep Learning Workloads