Technical details about Puhti
Puhti has a total of 682 CPU nodes, with a theoretical peak performance of 1,8 petaflops. Each node is equipped with two Intel Xeon processors, code name Cascade Lake, with 20 cores each running at 2,1 GHz. The cores support AVX-512 vector instructions and VNNI instructions for AI inference workloads. The interconnect is based on Mellanox HDR InfiniBand. The nodes are connected with a 100 Gbps HDR100 link, and the topology is a fat tree with a blocking factor of approximately 2:1.
The Puhti AI artificial intelligence partition has a total of 80 GPU nodes with a total peak performance of 2,7 petaflops. Each node has two latest generation Intel Xeon processors, code name Cascade Lake, with 20 cores each running at 2,1 GHz. They also have four Nvidia Volta V100 GPUs with 32 GB of memory each. The nodes are equipped with 384 GB of main memory and 3,6 TB of fast local storage. This partition is engineered to allow GPU-intensive workloads to scale well across multiple nodes. The interconnect is based on a dual-rail HDR100 interconnect network connectivity providing 200 Gbps of aggregate bandwidth in a non-blocking fat-tree topology.
|Name||Number of nodes||Compute||Cores||Memory||Local disk|
|M||532||Xeon Gold 6230||2 x 20 cores @ 2,1 GHz||192 GiB|
|L||92||Xeon Gold 6230||2 x 20 cores @ 2,1 GHz||384 GiB|
|IO||40||Xeon Gold 6230||2 x 20 cores @ 2,1 GHz||384 GiB||3600 GiB|
|XL||12||Xeon Gold 6230||2 x 20 cores @ 2,1 GHz||768 GiB|
|BM||12||Xeon Gold 6230||2 x 20 cores @ 2,1 GHz||1,5 TiB|
|GPU||80||Xeon Gold 6230
|2 x 20 cores @ 2,1 GHz
4 GPUs connected with NVLink
4 x 32 GB
In addition to the compute nodes above, Puhti has two login nodes with 40 cores and 2900 GiB local disk each.
Currrnet Lustre configuration for Puhti is:
|Storage area||# OSTs||# MDTs|
Please see Lustre documentation for the terminology.
All the OSTs and MDTs are shared across the storage areas, thus the performance should be similar between them. The peak I/O performance in Puhti is around ~50 GB/s with 64 compute nodes under dedicated access to the system.