Self-Hosted HPC

GPU Infrastructure

42 GPU dies across 5 nodes with InfiniBand 40 Gbps fabric and CUDA-aware MPI.

18
V100 GPUs
24
K80 GPU Dies
42
Total GPU Dies
40 Gbps
InfiniBand
5
Compute Nodes

Cloud GPU Computing for Engineering Simulation

KYAMOS provides cloud-accessible GPU computing infrastructure for high-performance multiphysics simulations, AI-assisted modelling, and engineering optimization workflows. The cluster supports scalable numerical solvers, neural surrogate model training, and fast batch inference for research and industrial applications.

Cluster Availability

The telemetry below provides a live operational view of the KYAMOS self-hosted GPU cluster. It is used to monitor resource availability, GPU utilization, memory allocation, and node status for active simulation and AI workloads.

Loading live GPU telemetry...

Hardware

NVIDIA Tesla V100

NVIDIA Tesla V100

3 nodes × 6 GPUs = 18 V100 GPUs. 16/32 GB HBM2 per GPU. Primary workhorse for deep learning training and CFD simulation.

NVIDIA Tesla K80

NVIDIA Tesla K80

2 nodes × 6 cards × 2 dies = 24 GPU dies. 12 GB GDDR5 per die. Used for batch processing and inference.

InfiniBand Fabric

InfiniBand Fabric

40 Gbps interconnect enabling low-latency GPU-to-GPU communication across nodes.

CUDA-Aware MPI

CUDA-Aware MPI

Direct GPU memory transfers between nodes without staging through CPU memory. Near-linear scaling for multi-GPU simulations.