WHERE GPU ENGINEERS WRITE

Stories from the bleeding edge.

Deep-dive articles, technical postmortems, tutorials, and research from the engineers building NovAI’s serverless GPU platform. Written by builders, for builders.

In-depth guides

From setup to scale

Real pricing data

From live provider scans

Practical analysis

Written by practitioners

ENGINEERING

How we hit 340ms cold starts on H100 GPUs

A breakdown of every millisecond — from container scheduling to CUDA context init and VRAM prefetching.

Reading progress 0%
POPULAR
  • Total reads0
  • Avg read time0
  • Subscribers0
ENGINEERING

Fine-tune Llama 3 with LoRA in 30 min

LoRA QLoRA Serverless
RESEARCH

Speculative decoding: 3x throughput