GPU Memory Physics: DRAM vs SRAM, Kernel Fusion and FlashAttention (Series Part 3)
GPU memory physics: DRAM vs SRAM latency and bandwidth, tiling, kernel fusion, and why FlashAttention made LLM inference faster. Series finale.
GPU Kernel Engineering: Why C = A + B Is Not Enough (Series Part 2)
GPU kernel engineering explained: thread indexing, launch configuration, coalescing, divergence, and why a naive C = A + B kernel starves on memory.
GPU Parallelism Explained: SM, Block, Warp and Thread Hierarchy (Series Part 1)
GPU parallelism from first principles: CPU vs GPU, the chip-SM-block-warp-thread hierarchy, and how an iPhone 120 FPS camera pipeline maps onto it.
SGLang 0.5.18 vs 0.5.15: What Changed and How to Upgrade
SGLang 0.5.16-0.5.18 add DSpark speculative decoding, a Rust server and Breakable CUDA Graph. Breaking flags, cache moves and a step-by-step upgrade path.
vLLM 0.28 to 0.30 Migration: Model Runner V2 Default, Breaking Changes
Upgrading vLLM 0.28 to 0.30: Model Runner V2 is now default, MRV1 removal targeted for v0.32, new flags, removed env vars and a safe rollout checklist.
Kubernetes 1.37 Gang Scheduling vs Volcano, Kueue and YuniKorn
Kubernetes 1.37 promotes PodGroup and gang scheduling to beta and adds CompositePodGroup. Do you still need Volcano, Kueue or YuniKorn?
TensorRT 11 vs TensorRT 10: Porting IPluginV2 to IPluginV3 Before Your Build Breaks
TensorRT 11 removed the entire IPluginV2 family, weak-typing builder flags, and implicit quantization. What breaks in your engine build and how to port it.
Karpenter vs Cluster Autoscaler for GPU Nodes: 2026 Cost Guide
Karpenter vs Cluster Autoscaler for GPU node scaling in 2026: bin-packing, spot, cold-start, consolidation and the real cost difference for ML clusters.
Kubernetes Cost Optimization and GPU Rightsizing (2026)
A deep dive into Kubernetes cost optimization in 2026: bin-packing, fractional GPUs, Karpenter, requests/limits tuning, and FinOps guardrails.