Computing-infrastructure R&D
We design and prototype custom compute fabrics purpose-built for AI training and inference workloads — GPU pods, high-throughput networking, and the orchestration software that ties it all together. Every reference architecture ships with a power budget, an observability story, and a swap-in roadmap for next-generation accelerators.
- Reference architectures for H100 / H200 / B-class GPU clusters
- Containerised scheduler extensions for Slurm, k8s and Nomad
- Throughput engineering: NCCL tuning, RoCE fabrics, NUMA pinning
- Joint research engagements with academic and industrial partners