Houmo M50

Headlines · 3

1. 22h
   Houmo AI unveils dual 2D/3D CIM roadmap; first 3D CIM chip A20 in production

   At its first ecosystem partner conference on Oct. 14, Houmo AI announced two compute-in-memory paths built on SRAM-CIM: a 2D CIM line for compute- and capacity-heavy workloads such as image/video generation and MoE, and a 3D CIM line for bandwidth-hungry dense models; the company says its 3D approach upgrades the underlying logic die to an SRAM-based CIM rather than using conventional 3D stacking alone. Its roadmap lists the already mass-produced 2D CIM chips M30 and M50, the newly in-production first 3D CIM chip A20 — claimed to raise compute utilization by 50% over the previous generation and support up to 16 fully interconnected dies — and the in-development M6X, claimed to reach 3–6x M50's overall capability. Houmo says M50 has won more than 150 product design-ins and serves over 100 customers within a year of mass production; a Lenovo AI host P7 based on M50 is said to run local models up to 122B parameters (190 TOPS, up to 128K context, about 50 tokens/s offline), with details due at an Oct. 24 launch, and its AI ecosystem software has been open-sourced.

   智东西 · 🔥 5
2. 2d
   Houmo AI: agent boom pushes edge AI toward a 'singularity moment'

   At an October 14 conference on large models for smart terminals and Houmo AI's ecosystem partner event, founder and CEO Wu Qiang said the surge in agents has driven token consumption up by two to three orders of magnitude, while the company's M50 compute-in-memory edge chip has won more than 150 product design-ins and over 100 customers less than a year after mass production. Fudan University scholar Jiang Changjian cited Amazon's blocking of Meta's personal agent Muse as an example of the technical-permission and business-rule questions edge AI raises, arguing for open and fair interface governance. Several speakers said edge AI could approach a tipping point in one to two years as models and chips co-evolve, with Alibaba Cloud's Xu Chao predicting edge models could cover many work needs by mid-2027, though limited scenarios and slow scaled deployment remain the main bottlenecks.

   钛媒体 · 🔥 3
3. 3d
   Houmo AI open-sources full software stack for its M50 edge chip

   At an ecosystem partner conference on October 14, Houmo AI said it will fully open-source its AI software stack, covering four layers — model quantization and compilation, inference frameworks and operator libraries, domain kits and middleware, and deployment/operations tools — in three batches, with 15 projects due by January, including adaptations of llama.cpp and vLLM; postmo-modelzoo, postmo-quantflow and vllm-postmo are already on GitHub. The company says its M50 chip, less than a year into mass production, has been picked for more than 150 terminal products by over 100 customers, including Lenovo's 300g AI host P7 (up to 48GB LPDDR5/5X, a 122-billion-parameter model running offline at 50 tokens/s), Great Wall's N90 Pro laptop running a 35B model fully offline on Kylin V11, and UGREEN's HomeAgent HA100 Pro. Houmo also claims its SRAM compute-in-memory design cuts chip area by about 70% and power to roughly one-tenth, and that M50 is China's only high-compute edge AI chip compatible with mainstream operating systems and CPU architectures.

   新智元 · 🔥 0

Experience and discussion from the community

Share my Houmo M50 experienceAsk about Houmo M50