DeepSeek reportedly runs inference on NVIDIA gaming GPUs, 70% of $1B budget goes to training
teortaxesTex · x · 2026-09-24
A thread compiling DeepSeek details: a $1B compute budget with 70% allocated to training deemed logical, and a notable claim that DeepSeek uses NVIDIA gaming GPUs to run inference on smaller models. Commenters see Jev-style small models or a custom Qwen 35B as an obvious efficiency play; another quips about Liang's open-source pledge covering the strongest model but asking about the fastest.
Related event: DeepSeek Hits $1B ARR, Targets $70B Raise and STAR Market IPO(7 posts)→
More from Infra
- Oracle issues force-majeure notices on data center project, fueling delay speculation — ns123abc · 2026-09-25
- Custom ComfyUI Node Boosts Qwen-Image 2.1 Prompt Enhancer 1.4-1.7x via MTP Head — mozophe · 2026-09-25
- Elon Musk's Old Energy Warning Comes True as AI Power Crunch Hits — XFreeze · 2026-09-25
- Nvidia to buy $1.5B more SB Energy shares pre-IPO, lifting stake to $3B — Beth_Kindig · 2026-09-25
- Celesto Launches Real Computers for AI Agents With 500ms-Boot MicroVMs — aniketmaurya · 2026-09-24
- Report: Google to Launch Space TPUs on Falcon 9 Next Week to Test Orbital AI Data Centers — ns123abc · 2026-09-24