inference/research
draftlow · 0.30updated 2026-10-11

Iluvatar CoreX BI-V150 (Tiangai 150) — product overview and estimates

review
This record is a draft. Not yet verified. Provisional. Treat every number on this page as provisional.

Fields

id
sources/cn-iluvatar-bi-v150
Publisher
mirrorfrog.com (aggregating Chinese sources)
Kind
review
Accessed
2026-10-11
notes
SECONDARY, and the source itself warns that 'Detailed specifications of TianGai 150 are not fully disclosed by official sources.' The BI-V150 is Iluvatar's second-generation cloud GPGPU training accelerator, released 2023, using the ivcore11 architecture on 7nm with 2.5D CoWoS packaging. Estimates (explicitly labeled as such by the source): FP16 150-200 TFLOPS, INT8 300-400 TOPS, memory 32-64 GB HBM2/HBM3, TDP 275 W (estimated) or 350 W (per moark.com). The BI-V150 participated in Zhiyuan Research Institute's Aquila2-70B-Expr hybrid training (120-node BI-V100 + 8-node BI-V150 cluster, 85.3% of theoretical peak). The moark.com page states 350 W peak power. Iluvatar's own website publishes no specifications for this part. 2026-10-10: set vendor_silence=aggregator-only. The record's own basis/notes already establish the reason; the field only names the case the schema now distinguishes. No number changed. 2026-10-11: Re-read mirrorfrog.com/en/docs/cards/others/iluvatar-tg150/; confirms aggregator-only mechanism (source explicitly notes 'Detailed specifications of TianGai 150 are not fully disclosed by official sources' and Iluvatar publishes no official spec sheet).
vendor silence
aggregator-only

Notes

SECONDARY, and the source itself warns that 'Detailed specifications of TianGai 150 are not fully disclosed by official sources.' The BI-V150 is Iluvatar's second-generation cloud GPGPU training accelerator, released 2023, using the ivcore11 architecture on 7nm with 2.5D CoWoS packaging. Estimates (explicitly labeled as such by the source): FP16 150-200 TFLOPS, INT8 300-400 TOPS, memory 32-64 GB HBM2/HBM3, TDP 275 W (estimated) or 350 W (per moark.com). The BI-V150 participated in Zhiyuan Research Institute's Aquila2-70B-Expr hybrid training (120-node BI-V100 + 8-node BI-V150 cluster, 85.3% of theoretical peak). The moark.com page states 350 W peak power. Iluvatar's own website publishes no specifications for this part.

2026-10-10: set vendor_silence=aggregator-only. The record's own basis/notes already establish the reason; the field only names the case the schema now distinguishes. No number changed.

2026-10-11: Re-read mirrorfrog.com/en/docs/cards/others/iluvatar-tg150/; confirms aggregator-only mechanism (source explicitly notes 'Detailed specifications of TianGai 150 are not fully disclosed by official sources' and Iluvatar publishes no official spec sheet).

Cross-references

This record references no other records.