DSDaya Shankarinnextgengpu.hashnode.dev·Aug 3 · 6 min readHosted LLM API vs Self-Hosted LLM on a Cloud GPUA hosted LLM API lets you send prompts to a provider’s model. Self-hosting means renting a cloud GPU, deploying an open-source model, and operating the inference server yourself. For most small teams,00
DSDaya Shankarinnextgengpu.hashnode.dev·Aug 3 · 7 min readWhen CPU Inference Is Enough and When You Need a GPU You can run an LLM without a GPU. The harder question is whether it will run well enough for your use case. A CPU may be perfectly suitable for local experiments, occasional document processing, or a 00
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
