OpenAI's Astra gets 3x throughput on NVIDIA Vera Rubin, plus 2x more in 72 hours
MickeySteamboat · x · 2026-09-17
NVIDIA AI Infra announced that OpenAI moved its Astra code, previously optimized for Blackwell, to the Vera Rubin platform, delivering 3x higher throughput out of the box. Within 72 hours, AI-driven optimization unlocked an additional 2x gain. NVIDIA claims the right hardware-plus-tooling combo can compress weeks or months of optimization into a single weekend. Full conversation with Bloomberg's Dina Bass at AI Infra Summit 2026.
More from Infra
- OpenAI's Jalapeño chip wasn't AI-designed: 100+ ex-Google TPU engineers and Broadcom were — JFPuget · 2026-09-17
- S3 is supposed to span AZs — devs question cheapest-tier AWS data loss claim — zetalyrae · 2026-09-17
- Build the agent setup first, pick the model second: a Linux + Tailscale + llama.cpp stack guide — max_paperclips · 2026-09-17
- Huawei unveils Ascend 960 chips for 2027 and UnifiedBus linking one million processors — mark_k · 2026-09-17
- Leak: CXMT Supplies TSV Embedded DRAM Dies for Chinese cHBM, Stacking Done In-House or via Local OSAT — zephyr_z9 · 2026-09-17
- Baseten adds server-side web search for open models, 15% lower latency — baseten · 2026-09-17