gpu

Hardware coverage for the people actually running the workloads: VRAM budgets, multi-GPU rigs, CUDA kernel profiling, and the gap between raw specs and real throughput. Expect hands-on notes on optimization flags, memory bottlenecks, and what consumer cards like the RTX 3090 can and cannot do for local inference and training. The focus stays on cost tradeoffs, where local compute pays off, and where a cloud API still wins on practical grounds.

local-llms, gpu, self-hosting, ai-development

Four GPUs, Two Weeks, and the Uncomfortable Truth About Local LLMs

What happens when you throw 96 GB of VRAM at open-source models, optimize every last flag, profile CUDA kernels, redesign system prompts, and still end up reaching for a cloud API. It started with a simple premise: why pay per token when you have the hardware? I have four