Local RTX 4080S 32G + modified Qwen3.8-27B-Q4XL
Local RTX 4080S 32G + modified Qwen3.8-27B-Q4XL, driven by pure prompt words copied from DeepSeek Harness online, created a black hole effect in 2 hours, and the effect exceeded expectations. 📊 2 rounds · 90 steps | LLM 125m56s · Tool 1m9s | First token 7.3s · 35 tok/s | Cache hit 97% | Input 12.3M · Output 242K tok…
Goldman Sachs report: Ulanqab has locked 12.5GW of computing power capacity! The end game of AI competition is energy.
Four trump cards: average annual air cooling of 4.3°C, GW-level savings of 500 million, fiber latency of 4.2ms, and DeepSeek layout. The rules have changed, and natural endowments have been repriced by AI! #AI #computing power 🎬 Full version synced to YouTube:
A Tsinghua University lab just put a project on GitHub that replaces a $400,000 H100 rack with a single 24GB graphics card.
It's called ktransformers, and the trick is almost stupidly simple: the experts you actually use stay on the gpu, the ones you don't sit on the cpu until
Kevin Carter of EMQQ Global explains AI's impact on emerging markets (EM), noting that $TSM
Samsung and $SKHY have driven a substantial share of EM returns. He also discusses his firm's products, including the China AI Tigers ETF, which focuses on open-weight LLMs such as
You have reached the end of the archive
All of deepseek