Alibaba Cloud Launches M890 Supernode for 10T-Parameter Model Inference
Dr_Singularity · x · 2026-07-22
At WAIC 2026, Alibaba Cloud debuted the Lingjun Zhenwu M890 supernode instance, now open for invite testing. It utilizes ICN Switch 1.0 to scale GPU interconnect from 16 to 64 cards at 800GB/s, supporting FP8/FP4 low-precision compute.
The company claims three times the training throughput of the prior 810E generation, with a single M890 instance capable of handling inference for a 10-trillion-parameter mixture-of-experts model. The underlying HPN 8.0 architecture allows a single cluster to mix up to 130,000 heterogeneous GPUs, scaling to a million-card mesh with 99.7% availability.
More from Infra
- SF Compute founder: buying compute is 'an absolutely awful experience' right now — IgorCarron · 2026-09-11
- SmolVM open-sources persistent computer infrastructure for agents that outlive chat sessions — aniketmaurya · 2026-09-11
- PyTorch Day Korea 2026 launches first offline conf, CFP closes Sept 13 — PyTorch · 2026-09-11
- Local LLM server dilemma: 4x CMP-170HX (price up 53% in 20 days) vs Mac Studio M5 Ultra — rumboll · 2026-09-11
- llama.cpp lands Flash Attention tuning for RDNA4, big prefill gains on AMD — pmttyji · 2026-09-11
- Your p99 latency benchmark may be lying: a deep dive into coordinated omission — Franc0Fernand0 · 2026-09-11