DIHUBMT HPCMT
The DiHubMT HPC is a high-performance computing environment engineered to handle large-scale computational and data-driven workloads. It enables users to accelerate scientific discovery, machine learning and engineering innovation.
Ready to get started?
HPC - INFRASTRUCTURE
Compute
-
GPU node (1x GPU Server):
-
CPUs: 2 x 56-Core 4th Gen Intel Xeon Scalable Processors.
-
GPUs: 8 x NVIDIA H100 SXM5 GPUs with 80GB RAM each, providing ~ 2 PFlops of FP16 performance per GPU / ~15.8 PFLOPS in total.
-
System RAM: 1024 GB of high-speed DDR5 memory.
-
Storage: Multi-terabyte NVMe SSD storage for ultra-fast data access.
-
-
General Compute Cluster (3x Compute Nodes):
-
CPUs: 2 x 32-Core CPUs per node.
-
System RAM: 768 GB of DDR5 memory per node.
-
-
Large Memory Node (1x Fat Memory Node):
-
CPUs: 4 x 32-Core CPUs.
-
System RAM: A massive 4096 GB (4TB) of DDR5 memory.
-
Storage
-
Fast Tier: Two all-flash metadata nodes equipped with a total of 36 x 3.84TB NVMe SSDs for extreme I/O performance.
-
High-Capacity Tier: Two storage arrays providing over 640TB of raw capacity using 160 x 4TB SAS drives, accelerated by an SSD cache. The arrays feature dual controllers for high availability and support auto-tiering, snapshots, and thin provisioning.
-
Staging Area: Two dedicated, all-flash NAS arrays provide a high-speed staging environment for data ingestion and preparation.
Networking
The entire cluster is interconnected via a Quantum 2 NDR Infiniband switch, ensuring ultra-low latency and high-bandwidth communication between all nodes, which is critical for distributed computing tasks.
Redundant
25Gb/s Ethernet fabric for production and management traffic.