Case Study: National Taiwan University achieves a record-breaking SVP solution and cost-effective GPU compute with Amazon Web Services

National Taiwan University - Customer Case Study

Fast Crypto Lab at National Taiwan University is a research group that designs and implements algorithms on massively parallel computers. After running Hadoop on a private cloud, the team needed large-scale, reliable GPU compute to break the record for solving the shortest vector problem (SVP) in Euclidean lattices and to compare algorithm performance across consistent cost metrics.

They migrated to AWS, using Amazon EC2 cluster GPU instances, Amazon EMR with Hadoop Streaming, and Amazon CloudWatch to monitor utilization. By running 50 cg1.4xlarge instances (100 Tesla M2050 GPUs) for about 10 hours—costing roughly $2,300—they solved what are now considered the hardest SVP instances to date, reduced maintenance costs, and gained stable, scalable compute; they are now planning to increase their GPU quota and host services like an SVN server on AWS.

National Taiwan University

Chen Mou Cheng

The Principal Investigator, Fast Crypto Lab