inference/research
verifiedhigh · 0.95updated 2026-10-03

Iluvatar CoreX Tiangai 100 (BI-V100) training accelerator card product page

spec-sheet

Fields

id
sources/cn-iluvatar-tianggai100
Publisher
Iluvatar CoreX (天数智芯)
Kind
spec-sheet
Accessed
2026-10-03
notes
PRIMARY VENDOR SOURCE. Model BI-V100: 7 nm general-GPU architecture, 32 GB HBM2, 250 W board power, PCIe Gen4 x16, 64 GB/s bidirectional host bandwidth AND 64 GB/s device-to-device interconnect bandwidth, full-height full-length dual-slot passive card, 169.5 mm class. Released March 2021 after four years of development. Vendor claims support for 200+ AI models and competitive average performance across 100+ benchmark models, but publishes NO throughput number. Software is DeepSpark.

Notes

PRIMARY VENDOR SOURCE. Model BI-V100: 7 nm general-GPU architecture, 32 GB HBM2, 250 W board power, PCIe Gen4 x16, 64 GB/s bidirectional host bandwidth AND 64 GB/s device-to-device interconnect bandwidth, full-height full-length dual-slot passive card, 169.5 mm class. Released March 2021 after four years of development. Vendor claims support for 200+ AI models and competitive average performance across 100+ benchmark models, but publishes NO throughput number. Software is DeepSpark.

Cross-references

This record references no other records.