Cloud • AI • Architecture
Cloud and AI. No fluff. Practical insights for architects, engineers, and leaders navigating the modern tech landscape.
The nbd-vram daemon got a thread pool, ~4x concurrent 4K IOPS, and the swap-pressure freeze is finally gone. Fresh benchmarks against NVMe.
Most RAG accuracy problems are retrieval misses, not model failures. Here is how to measure and fix the part that actually breaks.
CUDA Unified Memory can extend effective VRAM for LLM inference. The catch is PCIe bandwidth, which makes it slower than just splitting layers to CPU for any meaningful overflow.
New posts delivered to your inbox. No spam.