⚡️ NVIDIA announces Vera rubin:

• Up to 10x lower inference token costs (vs. Blackwell)
• 4x fewer GPUs required for MoE training (vs. Blackwell)
• 5x better energy efficiency (Spectrum-X Photonics)
• 5x longer uptime (Spectrum-X Photonics)
• 10x higher reliability (Ethernet Photonics)
• 18x faster assembly and maintenance (vs. Blackwell)

Available H2 2026

aipost 🏴