Skip to main content

Enflame Blaze T20/T21

Vendor: Enflame Technology

Category: ASIC Dedicated Accelerator

Architecture: Enflame DTU

Introduction

Enflame Technology focuses on cloud AI chips with four generations of proprietary architectures and five chips. Blaze T20/T21 are training and inference products covering chips, accelerator cards, intelligent computing clusters, and a complete software platform. In January 2026, the company received STAR Market IPO acceptance.

Specifications

ModelComputeMemoryInterfaceTDPProcess
T21128 TFLOPS (FP16) / 256 (INT8)32GB HBM2E (1.6 TB/s)OAM300W7nm
T20134.4 TFLOPS (FP16) / 268.8 (INT8)32GB HBM2E (1.6 TB/s)PCIe 4.0300W7nm

Official Website

Visit Official Website

Driver Downloads

Linux

OS Support

WindowsLinuxmacOSAndroid

Version History

VersionRelease DateDescription
SDK 2.02024Training + inference dual scenarios

Performance Benchmarks

ModelTaskPerformance Metric
T21Training/InferenceNear A100 level (official data)
T20InferenceINT8 optimized

Pricing

ModelReference PriceNotes
T21Contact vendorEnterprise customers
T20Contact vendorIn mass production

Quick Installation

Linux

# Install Enflame driver and SDK
sudo rpm -ivh enflame-driver-*.rpm

Code Examples

Python (Enflame)

import torch
# Enflame GPU provides CUDA-compatible backend
assert torch.cuda.is_available()

Architecture Highlights

  • DTU Architecture: Enflame proprietary AI accelerator architecture designed for unified training and inference
  • Complete Software Stack: Deep learning compiler + runtime + operator library

Model Compatibility

Model/FrameworkSupportNotes
PyTorch⚠️ CompatibleCUDA-compatible backend
General AI⚠️Ecosystem developing

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