FastFlowLM is the latest acquisition by AMD as the chip giant continues to look for ways to improve performance across its stack.
FastFlowLM was formed from a collaboration of academic researchers, software engineers, and community maintainers. They built a runtime environment to run AI models locally on AMD hardware.
FastFlowLM is supported by Iron, an open-source, near-metal Python API tailored for AMD Neural Processing Units (NPUs), and ships without drivers, so users can simply run the installer, pull the models they want, and go.
AMD confirmed that the FastFlowLM team has joined the company’s AI group and their expertise aims to “accelerate” AI software stacks for AMD’s clients and workstations.
The team behind FastFlowLM includes Professors Tao Wei and Ken Qing Yang from the University of Rhode Island, and Professor Zhenyu (Alfred) Xu from Clemson University.
“What started as a bold idea to make AI inference faster and more efficient is now entering an exciting new chapter,” Wei said in a LinkedIn post, whose profile notes his role as an AMD fellow.
In announcing the deal, AMD said, “We remain committed to investing in this open ecosystem and are excited to build the future of on-device AI together.”
FastFlowLM joins Mext, which joined AMD in mid-June. Mext is a storage software startup whose predictive memory solution tricks the operating system into thinking flash is DRAM, expanding the system’s usable memory capacity with cheaper, higher-density options.
However, FastFlowLM is part of AMD’s efforts to promote open alternatives to break Nvidia’s proprietary stack lock-in. Its flagship Radeon Open Compute (ROCm) software stack provides developers and enterprises with the means to run AI and high performance computing (HPC) workloads using popular open frameworks such as PyTorch, TensorFlow, and JAX upstream, as well as open Python-first GPU compilers in the form of Triton, developed by OpenAI.
Financial details were not disclosed.
