以下部分引用 AI 总结，现阶段 AI 仍然有幻觉。请以内容的原文为准。

数据新鲜度群聊 07-14 ✓卖方 07-14 ✓Wrap 断档15天 ✕News 07-15 ✓X 断档12天 ✕页面生成 07-15

LineShine Armv9 CPU Supercomputer

7 atoms · 跨 1 天 · 首见 2026-05-12 · 最近 2026-05-12

三色: 🟦 fact 7 · 🟥 take 0 · stance ▲0/▼0/◆7

来源: X 7

时态: fresh:7

标签: 好数字:7

2026-05-12 · sustained performance, 6B model, 20480 nodes
sustaining 1.54 EFLOP/s at 20480 nodes with a weak scaling efficiency of 76.0%
tags: 好数字 · → daily
value: qty=1.54 EFLOP/s · date=
2026-05-12 · peak FLOPs, 6B model, full-machine scale
Peak FLOPs reaches 2.16 EFLOP/s
tags: 好数字 · → daily
value: qty=2.16 EFLOP/s · date=
2026-05-12 · weak scaling efficiency, 6B model, 20480 nodes
weak scaling efficiency of 76.0%
tags: 好数字 · → daily
value: qty=76.0% · date=
2026-05-12 · sustained performance, 3B model, 4096 nodes
For the 3B and 8B models, the system scales to 4096 nodes and achieves 275 PFLOP/s and 295 PFLOP/s sustained performance, with corresponding weak scaling efficiencies of 80.4% and 70.4%, respectively.
tags: 好数字 · → daily
value: qty=275 PFLOP/s · date=
2026-05-12 · sustained performance, 8B model, 4096 nodes
295 PFLOP/s sustained performance
tags: 好数字 · → daily
value: qty=295 PFLOP/s · date=
2026-05-12 · weak scaling efficiency, 3B model, 4096 nodes
weak scaling efficiencies of 80.4%
tags: 好数字 · → daily
value: qty=80.4% · date=
2026-05-12 · weak scaling efficiency, 8B model, 4096 nodes
corresponding weak scaling efficiencies of 80.4% and 70.4%
tags: 好数字 · → daily
value: qty=70.4% · date=