AMD Instinct MI300: detailed architecture and performance analysis

The AMD Instinct MI300 series, based on the CDNA3 architecture, introduces a heterogeneous chiplet design with up to 8 XCD modules and 4 IOD modules, using advanced CoWoS packaging and 3D V-Cache technology. The lineup includes the MI300X accelerator and the MI300A APU with integrated Zen 4 cores, offering up to 304 compute units, HBM3 memory up to 192 GB, and significant performance improvements over the previous generation and NVIDIA H100 in both HPC and AI workloads.

NVIDIA Taiwan Semiconductor Manufacturing Company

New DeepSeek-V3 technical report: how hardware-software co-design enables low-cost training of large models

A new technical paper from the DeepSeek team, with CEO Wenfeng Liang as co-author, explores hardware-aware model co-design to reduce LLM training costs. Using DeepSeek-V3 trained on 2048 NVIDIA H800 GPUs as a case study, the paper details innovations in memory efficiency (MLA), sparse computation (DeepSeekMoE), FP8 training, and interconnect-aware routing.

DeepSeek NVIDIA Alibaba/Qwen Meta