Qwen3.8 FP8: does the cheap card really save?Cost of the Answer — issue #2. Issue #1 closed on a promise. The FP8 checkpoint fits a 32 GB consumer card that BF16 cannot rent, and the consumer card costs about half as much per hour as the 96 GB wAug 19 · 12 min readkrauncher.hashnode.dev
Qwen3.8: is FP8 worth it under transformers?Cost of the Answer — issue #1. Qwen3.8-27B ships in two precisions on the HuggingFace hub. The FP8 checkpoint is roughly half the byte count of the BF16 one, and I wanted to know what that halving buyAug 17 · 9 min readkrauncher.hashnode.dev
Most people who quit machine learning never got to itDisclosure: I build Krauncher, a service that prices GPU tasks before they run and places them on rented hardware. This article compares it honestly with the alternatives, including the ones I use mysAug 5 · 17 min readkrauncher.hashnode.dev
The Messy Real World, or Why Probability Decides What You PayResearchers pick GPUs the way a spec sheet invites them to: choose a card, read a number, pay the bill. But the real world doesn't behave like a number. It behaves like a cloud — and that cloud is thrJul 8 · 5 min readkrauncher.hashnode.dev
There are two ways to rent a GPU. I think there's a third.Disclosure: I build Krauncher, a service that prices and provisions GPU hosts per task automatically. What follows is the reasoning the project was built on — and why the unit on the price list is notJun 30 · 14 min readkrauncher.hashnode.dev
