Introduction to High Performance Computing on Google Cloud Platform
summary
Google Compute Engine exposes custom machine types (1–160 cores, ~4 TB RAM), Skylake CPUs, V100 GPUs, local SSDs and POSIX/cloud storage, plus preemptible VMs discounted up to ~80%. Slurm, HTCondor and PBScloud.io integrate autoscaling and federation for batch/HPC. Real cites: Fermilab bursts 160k→320k cores for ~$1,400/hr, an MIT job reaches ~580,000 cores using preemptibles, Schlumberger taps ~35 PFLOPS and 10 PB, and a drug-discovery run uses 16k cores for 3M core‑hours at ~$30k.
AI-written summary · about
chapters (27)
- 0:00Intro
- 1:03High Performance Computing
- 1:51Two types of workloads
- 3:50HPC Infrastructure Requirements
- 4:30On-premise HPC Clusters
- 5:18HPC Clusters in the Cloud
- 6:35Compute Engine
- 9:35Intel Skylake CPUs
- 10:15NVIDIA GPUs
- 10:47Block Storage Options
- 12:37Cloud Storage for objects
- 13:25Google Networking
- 14:22Global Google Network
- 15:03Better Economics
- 16:07Preemptible VMS
- 18:20Cluster Management Software
- 19:58Altair's PBScloud.io
- 21:33Slurm Workload Manager
- 22:52Cloud Auto-Scaling
- 25:00Federate to Cloud
- 26:39HTCondor with Google Cloud
- 27:20Fermilab & HEPCloud
- 29:36MIT math research with 580.000 cores
- 31:40Schlumberger
- 32:40Silicon Therapeutics
- 34:07Customer Use Cases
- 34:20More talks on HPC at Next
related talks
16mA novel Cloud High Performance Computing infrastructure for user
22mLisa Spelman | Google Next 2017
34mHPC on AWS: Innovation without infrastructure constraints
42mBuilding a green supercomputer in the cloud
39mWhat would you do with 10^6 cores? High Performance Computing in the AWS Cloud
01h 13mEngineering the Cloud for Tomorrow: Performance, Speed and Scale for AI Workloads
17mOptimizing Cloud Compute Resources with Spare Cores
47mCloud Native and Sustainable, Reproducible Scientific Computing