Image created with gemini-3.1-flash-image-preview with claude-sonnet-4-5. Image prompt: Wide static shot of a sparse decommissioned server room in muted gray concrete, half-empty equipment racks casting geometric shadows, a rust-colored horse standing naturally among obsolete hardware under flat fluorescent light, weathered walls, desaturated teal and beige tones, observational realism, large white Chinese poster-style text overlay reading DEEPSEEK, documentary stillness, Jia Zhangke compositional patience, postindustrial decay, human-scale framing.
DeepSeek-V3.2 on GB300: Performance Breakthrough | vLLM Blog https://vllm.ai/blog/gb300-deepseek
People who say “”ohh this model is faster it must be mini”” must have forgotten that DeepSeek boasted of 60 tps when V3 came out. V3, of course, was ≈3x larger and had almost 2x more active params than V2. Then they increased the batch size and tps went back down. it’s weak signal”” https://x.com/teortaxesTex/status/2022255213394948360
Leave a Reply