DeepSeek's open-source kernels reveal Huawei Ascend 950's 32-core architecture and frequency gains
teortaxesTex · x · 2026-10-03
- DeepSeek's open-sourced DeepGEMM, FlashMLA, TileKernels and DeepEP let analysts reverse-engineer Huawei's Ascend 950 beyond spec sheets.
- Each AI Core packs 1 Cube Core + 2 Vector Cores: 32 AI Cores, 32 Cube Cores and 64 Vector Cores total. Cube handles dense matmul; Vector handles dequantization, Softmax and more.
- skip-scale optimization cuts FixPipe transfers, vector work and memory movement; DeepSeek notes lower compute/data-movement power lets the chip clock higher, improving performance.
- The analysis traces the full path from matrix compute and memory movement to multi-NPU communication, showing open-source kernels as a new lens on China's AI silicon.
Related event: DeepSeek Open-Source Kernels Reveal Custom Chip and Ascend 950 Details(2 posts)→
More from Infra
- AI Conference Day 2: Agent Inference Costs and Observability Steal the Show — sanjaykalra · 2026-10-03
- Apple M5 Ultra 256GB Context Benchmark Shows Strong Prefill Speed on Quantized Qwen — antirez · 2026-10-03
- Fractile Founder on Winning the AI Chip Market and the Memory Bandwidth Bottleneck — saranormous · 2026-10-03
- SoftBank pays $3.1B for data-center investment manager as buildout runs on debt — YvesMulkers · 2026-10-03
- Cloudflare Open-Sources Its Internal AI 'OS', Hits 10k GitHub Stars — ritakozlov · 2026-10-03
- CoreWeave wants the industry to stop calling it a 'neocloud': 'you can only be new for a short time' — DavidLinthicum · 2026-10-03