IndustrySelected

1.3 billion sandboxes in 4 weeks: frontier RL compute math

Why we picked it

The arithmetic turns a vague scale claim into concrete GPU counts, giving a rare back-of-envelope view of what frontier RL training infrastructure costs. Practitioners benchmarking their own training setups, or investors sizing compute demand, can use the per-unit figures as reference points. The DeepSeek comparison also shows how Chinese labs may reach similar throughput with tighter hardware budgets.

A post on X analyses a reported figure of 1.3 billion training sandboxes run over 4 weeks, equal to roughly 48.

Body

46 million sandboxes per day.The author benchmarks this against DeepSeek, where one DSec scale unit processes 3 million sandboxes, implying about 16 such units would be needed.

That translates to roughly 656 GB300 GPUs per unit, though the author notes DeepSeek likely operates with 128 GB300-equivalents or fewer per unit.

Original · Teortaxes

1.3 billion sandboxes over 4 weeks, huh. That's 48,46M sandboxes/day. At DeepSeek, one DSec scale unit does 3M, so would need 16 units, or roughly 656 GB300s/unit (DeepSeek likely uses ≤128 GB300-equivalents though). A big project. A hint of how it's done at the frontier. https://t.co/CaQgrLlfAQ

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Industry" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include DeepSeek, GB300, reinforcement learning, compute.

Source attribution: curated from Teortaxes (@teortaxesTex).