FP8
An 8-bit floating-point format that roughly doubles throughput and halves memory versus 16-bit for AI workloads.
Current numbers
216 GBHBM3E per Microsoft Maia 200 @ 7 TB/s; >10 PFLOPS FP4 / 5 PFLOPS FP8, 750W, +30% perf/$
2.52 PFLOPSFP8 per Trainium3 chip; 144 GB HBM3E @ 4.9 TB/s; first 3nm AWS accelerator
144 chipsper Trn3 UltraServer = 362 FP8 PFLOPS, 20.7 TB HBM, 4x perf/watt vs Trn2