Swordfish Inference Kernel Released
AlpinDale · x · 2026-07-13
Introduces the Swordfish Kernel: a weight-only inference kernel designed for the Blackwell SM100 architecture. The author positions it as the successor to Marlin and Machete, aiming to maximize throughput on data center-grade Blackwell GPUs.
Two core results are highlighted:
- On the B200, performing INT4 inline dequantization achieves roughly 90% of cuBLAS BF16 performance.
- On Jetson Thor, it reaches approximately 95% of memory bandwidth utilization.
The author notes this is the result of their personal research into the Blackwell architecture, and both a related blog post and the source code have been made public.
Related event: Swordfish Inference Kernel for Blackwell Open Sourced(2 posts)→
More from Infra
- Meta approves nonvoting Class C shares as 2026 capex guidance hits $135B–$145B — thione · 2026-10-26
- Unsloth Ships MiniMax-M3.1-Flash Quants: 2-bit Retains 91.5% Accuracy, Runs in 150GB RAM — danielhanchen · 2026-10-26
- Engineer with thousands of fabricated devices pours cold water on photonic crystal modulator yields — jwt0625 · 2026-10-26
- Anthropic S-1 reveals 48 cents of compute cost per dollar of Q2 revenue — nmasc_ · 2026-10-26
- Persistent AI joins UR network to route private AI inference with encrypted, IP-hidden traffic — 0xSammy · 2026-10-26
- Anthropic's disclosed compute commitments: $114.5B Google, $161.2B Broadcom top the list — BenBajarin · 2026-10-26