Specifications
Flagship Part
BI-V100 (TianGai 100) general-purpose GPGPU
Process Node
7 nm
Packaging
2.5D CoWoS, approximately 24 billion transistors
Compute, FP16 / BF16
147 TFLOPS dense
GPU Memory
32 GB HBM2
Memory Bandwidth
1.2 TB/s
Board Power
250 W TDP
Form Factor
Full-height, full-length, dual-slot PCIe Gen4
Architecture
CoreX GPGPU
Launched
2021 — China's first domestic 7 nm GPGPU
Next-Generation Part
TianGai 150 — higher performance-density member of the line
Second Product Line
Zhikai series, oriented toward inference deployment
Roadmap
2026–28 GPU roadmap targeting NVIDIA H200/B200-class parts
Software
CUDA-migration-friendly toolchain with PyTorch, TensorFlow and ONNX support
Deployment Scale
50,000-unit domestic accelerator procurement programmes; China Telecom and state cloud deployments
Successor Architectures
Tianshu (Hopper-class) and Tianxuan (Blackwell-class) generations
Workloads
Cloud inference, light training, HPC and general-purpose computing
Overview
Iluvatar CoreX's BI-V100, branded TianGai 100, was announced in 2021 as China's first domestically developed 7 nm general-purpose GPGPU. It is a full-height, full-length dual-slot PCIe Gen4 card built with 2.5D CoWoS packaging and roughly 24 billion transistors, carrying 32 GB of HBM2 at 1.2 TB/s and rated at 147 TFLOPS of dense FP16/BF16 throughput within a 250 W board power envelope.
The product strategy prioritises deployment continuity rather than headline peak numbers. The toolchain is designed for straightforward CUDA migration, with the TianGai line positioned for cloud inference and light training alongside the Zhikai series, which targets inference-focused installations. Documented deployments include China Telecom and state-affiliated cloud programmes, and the company has been a beneficiary of large domestic accelerator procurement rounds.
Iluvatar's public roadmap pushes the same architecture family toward higher performance density with the TianGai 150, and its 2026–28 plan targets NVIDIA H200 and B200-class parts. Later generations, Tianshu and Tianxuan, are positioned against Hopper and Blackwell respectively. For buyers building sovereign or supply-chain-diversified AI capacity, the TianGai family is the most established domestic GPGPU option to evaluate alongside MetaX and Cambricon alternatives.
Key Benefits
Domestic 7 nm GPGPU supply that reduces single-source exposure on accelerator procurement; large HBM2 memory at 32 GB with 1.2 TB/s of bandwidth for memory-bound inference; CUDA-migration-friendly software that lowers porting cost from NVIDIA stacks; proven deployment scale in cloud and state-affiliated programmes; and a public multi-year roadmap from TianGai 100/150 through Tianshu and Tianxuan generations.
Applications
Cloud AI inference serving; light model training and fine-tuning; general-purpose GPU computing and HPC; data processing pipelines; sovereign and regulated-cloud AI infrastructure; and mixed domestic/NVIDIA accelerator clusters for supply-chain resilience.
Request a Quote — ILUVATAR COREX TIANGAI 100 & 150 — 7NM DOMESTIC GPGPU ACCELERATORS
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →