Blog

Engineering notes

Engineering notes on number formats, LLM inference with vLLM and fine-tuning large models, with the memory math and the commands.

All Engineering articles

Engineering
FP8 vs FP4 vs BF16: number formats for AI, explained
Bit layouts, NVFP4 and MXFP4 block scaling, which GPUs support each format, the memory saved and the accuracy you trade. Oct 22, 20268 min read
Engineering
How to fine-tune a 70B model: GPU requirements and workflow
QLoRA fits one 80 GB GPU, BF16 LoRA about two, full fine-tuning over 1 TB (8x H200 class). Memory math and setup. Oct 22, 20269 min read
Engineering
How to self-host LLM inference with vLLM
Size the GPU, run an OpenAI-compatible server, tune FP8, prefix caching and parallelism, and secure it for production. Oct 22, 20269 min read

Size it yourself

Deploy a GPU server

Get your GPU server this month.

Choose a configuration, create your account in the same form and pay the first month from your balance. Your firm delivery date is shown before you pay, or you get a full refund.

Firm delivery date before you pay
Cancel any month
Pay with crypto, up to +50% bonus

Configure a server See all prices

Engineering notes

All Engineering articles

FP8 vs FP4 vs BF16: number formats for AI, explained

How to fine-tune a 70B model: GPU requirements and workflow

How to self-host LLM inference with vLLM

Size it yourself

Get your GPU server this month.