According to IT Home, AMD 11 9 launched the enterprise-grade AI-assisted coding platform Instinct Coder 355 in partnership with Spectro Cloud, supporting the deployment of Zhipu GLM-5.3 inference and claiming to achieve 98% of the performance of frontier models locally. The system core consists of 8 Instinct MI355X accelerator cards, paired with 2 64-core EPYC 9575F CPUs and 3TB of memory, available in two versions from Dell and Supermicro; each node serves up to 85 developers (with 50 concurrent users). Spectro Cloud's accompanying PaletteAI Inference Launchpad management software defaults to invoking local compute resources rather than cloud APIs.
Why it matters Overseas hardware vendors are using domestic open-source models as the inference models for local coding solutions, defaulting to local compute resources rather than cloud APIs.
Credibility: Fairly credible · Source IT Home · Event time 11-09 (Beijing Time)
Reasoning
- IT Home reported the hardware configuration and number of users served
Usage
- Targeted at enterprise procurement, with two versions from Dell and Supermicro
Risks
- "98% of the performance of frontier models" is AMD's claim, without specifying the compared models or testing methods
- Pricing has not been disclosed