Topic

Nvfp4

2 entries · Topic

Nvfp4

Latest

12 Used Nvidia V100s Run All DeepSeek Experts at NVFP4 for $9,450

A $9,450 server of 12 used Nvidia V100s now runs all 15,360 DeepSeek V4.1 Flash experts at NVFP4, decoding at up to 184 tokens/s and beating DGX Spark on single requests.