NextGenergy · 液冷仿真台 · Liquid-cooling simulator
把你手上的液冷工况跑一遍，看它替你省掉哪一步
Run the liquid-cooling case you already have and see which step it saves you.
范围：整条热通路的六个边界，逐段建模 · 四道门验收 · Direct-to-Chip / 浸没式
Scope: six boundaries along the whole thermal path, each modelled · four commissioning gates · direct-to-chip and immersion
输入机型、柜数、城市、供液温度和末端形式，10 分钟内给出结温、流量与压降、四道调试门判定、N-1 失效后几秒触限、年电耗与 TCO。每个常数都标出处，每个结论都能用链接复现。
Enter the platform, rack count, city, supply temperature and rejection type; within ten minutes you get junction temperature, flow and pressure drop, the four commissioning gates, seconds-to-limit after an N-1 failure, annual energy and TCO. Every constant carries its source; every result is reproducible from a link.
一个完整案例 · A complete case
NVIDIA GB300 NVL72 × 24 柜，每列 8 柜，多伦多，干冷器，25% 丙二醇，供液 40 °C，ΔT 10 K
NVIDIA GB300 NVL72 × 24 racks, 8 per row, Toronto, dry cooler, 25 % propylene glycol, 40 °C supply, ΔT 10 K
≈ 4,775 L/min (1.49 L/min·kW⁻¹) 冷却功耗 / pPUE（设计点口径，含 8 % 风冷部分的风侧；打开链接后读数栏显示的是运行读数，约 166 kW，口径不同）· Cooling power / pPUE (design-point basis, incl. air side of the 8 % air-cooled share; the reading bar after opening the link shows the running value, about 166 kW, on a different basis)约 162 kW / 1.05
≈ 162 kW / 1.05 断流后最差支路触 95 °C · Worst branch reaches 95 °C after flow loss约 9.8 s（持液 2 L/架、从设计点稳态起算、泵流量降到 0.2 %；按 0.2 s 步长取首次到 95 °C 的那一步，与主站告警同口径）
≈ 9.8 s (liquid hold-up 2 L per rack, from design-point steady state, pump flow down to 0.2 %; first 0.2 s step at or above 95 °C, as in the simulator alert)
以上为集总参数模型在该工况、该假设下的模型值（v1.24.3 复算），用于方案比较与初选，不构成设计依据；正式选型以厂家数据为准。案例参数（如逼近温差 2.5 K）是这个案例的设定，不是仿真台默认值。改两个滑块就是你的工况。
Model values from the lumped-parameter model for this case and these assumptions (recomputed on v1.24.3), for comparing and shortlisting options, not a design basis; final selection follows vendor data. Case parameters (e.g. 2.5 K approach) are set for this case and are not the simulator defaults. Move two sliders and it is your case.
带一个工况来比较 · Bring a case to compare
把你正在比较的工况发给我们（脱敏即可：机型、柜数、每列几柜、城市、供液温度、干冷器还是塔）。我们先跑一遍，再约 20 分钟一起看哪一步能省、哪个数你不信、我们回去查出处。
Send the case you are comparing (anonymised: platform, racks, racks per row, city, supply temperature, dry cooler or tower). We run it first, then spend 20 minutes together on which step it saves and which number you doubt — we check the source and come back.
发一个工况 · Send a case 直接打开仿真台 · Open the simulator 使用说明 · Guide数据与测试状态：常数按 S / D / E 标出处（标准或厂家公开值 / 物理推导 / 估计值），校验脚本可复跑；已完成多轮 AI 辅助独立测试与回归，修复已随 v1.24.x 上线。产品目录按字段记录公开来源、核对日期与核对状态，收录不代表厂家认可或性能认证。本站与 OCP 无隶属关系。
Data and test status: every constant is tagged S / D / E (standard or public vendor value / physical derivation / estimate) and the check scripts can be re-run; several rounds of AI-assisted independent testing and regression are complete, and fixes have shipped with v1.24.x. The product catalogue records public sources, check dates and check status per field; a listing does not imply vendor endorsement or performance certification. This site is not affiliated with OCP.