Gowtham Ramu
Author, Trainer and Advisor
Long-form engineering notes
Newsletter
Newsletters below are published on dcops.ai and LinkedIn. Click any post to read the full article.
Required for Express Interest
Subscribe once to unlock Express Interest (or register for any event).
Newsletter #7: The New AI Memory Wall - How NVIDIA ICMS and BlueField-4 Are Rewriting Datacenter Architecture
• Feb 27, 2026 • dcops.aiAI systems are becoming memory-bound. NVIDIA ICMS and BlueField-4 introduce a new context tier that changes GPU datacenter design.
Read on dcops.ai ->
Newsletter #6: The Databricks-cuDF Evolution and the Communication Layer Shift
• Feb 19, 2026 • dcops.aiThe core change is not just faster cuDF. Distributed GPU data systems are converging toward GPU-native collective communication.
Read on dcops.ai ->
Newsletter #05: Containers vs KubeVirt for GPU Workloads
• Feb 19, 2026 • dcops.aiHow to choose the right abstraction for AI infrastructure: container-native GPU stacks vs KubeVirt VM isolation.
Read on dcops.ai ->
Newsletter #04: AI Infrastructure Job Market (HPC, Slurm, InfiniBand)
• Feb 11, 2026 • dcops.aiDemand, salary ranges, and skill priorities for AI infrastructure roles across HPC-style GPU clusters.
Read on dcops.ai ->
AWS P6e-GB200 UltraServer on EKS: An Engineer's 24-Hour Train -> Retrain Runbook
• Jan 27, 2026 • LinkedInA practical runbook for executing train->retrain inside a 24-hour AWS P6e UltraServer capacity block using EKS, DRA/IMEX, and EFA.
Read on LinkedIn ->
Kubernetes Enhancements - Enabled by Vera Rubin NVL72
• Jan 17, 2026 • LinkedInHow Rubin/NVL72 shifts Kubernetes realities: topology-aware scheduling, NCCL stability, DPU offload, CPU isolation, and multi-tenancy.
Read on LinkedIn ->
Beyond FLOPS: The Engineering Problems NVIDIA Rubin Actually Solves.
• Jan 6, 2026 • LinkedInA system-architecture view: cabling, firmware lifecycle, scheduler blind spots, deterministic fabrics, DPU offload, and failure domains.
Read on LinkedIn ->
