Skip to main content

Iluvatar BI-150/BI-V150

Vendor: Iluvatar

Category: GPU Graphics Processor

Architecture: Tianguai Architecture

Introduction

Iluvatar domestic GPU. BI-150 is a cloud training-inference card, BI-V150 is the next-gen product. Supports CUDA-compatible ecosystem, 2025 revenue 1.034 billion RMB, listed on Hong Kong Stock Exchange.

Specifications

ModelComputeMemoryInterfaceTDPProcess
Tianguai 150 (BI-V150)256 TFLOPS (FP16) / 384 (INT8)64GB HBM2e (1.6 TB/s)OAM350W7nm
Tianguai 100 (BI-V100)147 TFLOPS (FP16) / 295 (INT8)32GB HBM2 (1.2 TB/s)PCIe 4.0250W7nm
Zhikai 100200 TOPS (INT8)32GB GDDR6PCIe 4.0150W12nm

Official Website

Visit Official Website

Driver Downloads

Linux

OS Support

WindowsLinuxmacOSAndroid

Version History

VersionRelease DateDescription
BI-V 2.02023PyTorch compatibility layer

Performance Benchmarks

ModelTaskPerformance Metric
Tianguai 100BF16 TrainingNear A100 efficiency
Tianguai 100Inference ThroughputINT8 optimized inference
Zhikai 100Inference/ComputeLow-power inference scenarios

Pricing

ModelReference PriceNotes
Tianguai 100Contact vendorEnterprise AI accelerator
Zhikai 100Contact vendorInference-dedicated card

Quick Installation

Linux

# 1. Install Iluvatar driver
sudo rpm -ivh iluvatar-driver-*.rpm
# 2. Install SDK
tar -xzf iluvatar-sdk-*.tar.gz && sudo ./install.sh

# 3. Verify
iluvatar-smi

Code Examples

Python (Iluvatar BI-ACC)

import torch
# Tianguai GPU supports CUDA-compatible mode
assert torch.cuda.is_available()
print(f"GPU: {torch.cuda.get_device_name(0)}")

Architecture Highlights

  • Tianguai Architecture: Iluvatar proprietary GPGPU architecture, CUDA programming model compatible
  • Zhikai Series: Compact accelerator card for low-power inference scenarios
  • Domestic Ecosystem: Supports mainstream domestic OS (Kylin / UOS)

Model Compatibility

Model/FrameworkSupportNotes
PyTorch✅ CompatibleCUDA-compatible backend
PaddlePaddle⚠️Under adaptation
Large Model Inference⚠️Ecosystem developing

If you're evaluating alternatives, the following products may also fit your scenario: