DSDaya Shankarinnextgengpu.hashnode.dev·Mar 6 · 6 min readMulti-GPU scaling: when NVLink matters and when it doesn'tMulti-GPU scaling stops being linear when GPU-to-GPU communication grows faster than per-GPU compute in each step. This guide helps you decide when NVLink is worth paying for and when PCIe is the rig00
DSDaya Shankarinnextgengpu.hashnode.dev·Mar 6 · 8 min readHow to Add Quantum-Safe Encryption to Disaster Recovery Workflows for Ransomware ResilienceDisaster recovery (DR) is supposed to be your clean exit when ransomware hits. But attackers increasingly target the systems you rely on to recover, especially backup infrastructure and administrative00
DSDaya Shankarinnextgengpu.hashnode.dev·Feb 9 · 5 min readTraining vs inference on cloud GPUs: two very different infrastructure designsAI training and AI inference sit next to each other in the ML lifecycle, yet they pull GPU infrastructure in the cloud in opposite directions. Training is the development phase where you curate data, run experiments and update model weights repeate...00
DSDaya Shankarinnextgengpu.hashnode.dev·Dec 8, 2025 · 4 min readWhat NVIDIA Blackwell GPUs Mean for AI Training in 2026I will try to give you a clear picture before you reshuffle roadmaps, just to try NVIDIA Blackwell for AI training. So, Blackwell GPUs center on higher throughput, faster on-package memory, and a tighter interconnect that links many GPUs like one lar...00
DSDaya Shankarinnextgengpu.hashnode.dev·Dec 8, 2025 · 5 min readNVIDIA Blackwell GPUs (B100 B200) vs HopperWhen NVIDIA first unveiled Blackwell, my cloud GPU brain immediately went to one question: what does this actually change compared to Hopper in a production cluster? Hype is cheap. Power, cooling and utilization charts are not. After a few months of...00
