Accelerated systems engineering
Open systems,
accelerated.
GPU software adaptation and performance research, published with reproducible evidence.
Selected work ↓Selected work
Research index
01
LAMMPS Build and Performance Validation on Two Hygon DCU Platforms
LAMMPS · HPC · HIP · Kokkos
4.6–6.3×Kokkos 相对测试 GPU 路径
02Hashcat DTK/HIP Adaptation and Cross-platform Validation
Hashcat · HIP · ABI · Performance
5.914 GH/sSHA2-256 验证吞吐
03MinerU Pipeline and Service Pool on Hygon Z100 DCU
MinerU · Z100 · HIP · Deployment
3.9 s/pageDCU 流水线解析速度
04DeepSeek-V4-Flash-0731 Inference Adaptation on Hygon DCU
DeepSeek · DCU · PyTorch · Inference
0.93–1.12 tok/s单序列贪婪解码
05GROMACS 2026.3 Port and Benchmark on Hygon Z100 DCU
GROMACS · Z100 · HIP · Benchmark
3.20×最大 HIP 加速比
Open-source systems
From field notes
to reusable tools.
Operational knowledge for domestic supercomputing clusters, turned into installable agent tooling and ecosystem integrations.
Original · Core
Explore the full project portfolio scnet-hpc
Agent skill and automation for SSH, Slurm, Hygon DCU/DTK, and cluster troubleshooting.
GitHub ↗Original · Integrationdsh-scnet
DeepSeek Harness packaging for the shared SCNet knowledge and tools.
GitHub ↗Collaborative · ContributorOneFLOW
Open-source multiphysics scientific simulation and CFD.
Upstream ↗Build.
Measure.
Explain.
Every result is tied to a stated environment, workload, method, and boundary.
Research & collaboration
luqilinheu@gmail.com ↗