E
Enflame Tech Stack
AI chip design and software stack for data center inference
Enflame designs hardware and software for AI inference at scale, built on a deep compiler and systems stack (LLVM, CUDA, TensorFlow, PyTorch, NCCL, DeepSpeed). The project list—chip interconnect, quantization tools, operator development, and chiplet architecture—reveals a vertically integrated play from silicon through ML framework adaptation. Heavy senior engineering hiring (20 of 28 roles) against active work on bandwidth efficiency and power consumption points to a company solving hard performance and cost problems in domestic AI infrastructure.
Tech Stack 46 technologies
Core StackLinux C++ Python TensorFlow PyTorch LLVM GCC C/C++ CUDA RISC-V OpenCL OpenMP clang-tidy CDC C Caffe MXNet PaddlePaddle RDMA TensorRT MPI Open MPI NCCL DeepSpeed VTune Bash DPDK SPDK PCIe Intel+16 more
What Enflame Is Building
◆Challenges
- Ai model sparsification design
- Improving bandwidth efficiency
- Optimizing inference performance on domestic ai chips
- Next-gen ai chip design
- Balancing business and r&d priorities
- Ai model performance optimization
- Performance bottlenecks
- Ai system competitiveness
- Reducing power consumption
- Aligning market demands with chip development
▲Active Projects
- Ai chip interconnect design
- Build low-bit hardware-aware quantization tools
- Ai model sparsification design
- Quantization operator development
- Ai operator development on custom ai chip
- D2d interconnect architecture design
- Develop day-0 adaptation toolchain for gcu
- Pd separation architecture implementation on gcu
- Distributed communication component development
- Multi-die/chiplet architecture for ai server chip
Hiring Activity
30 roles · 25 in 30d
Department
Seniority
Company intelligence
