chore(specs): set real RTX PRO 6000 all-in power and cost tiers / 设置 RTX PRO 6000 的正式整机功耗与成本档位

[functionstackx]

Summary

Replaces the 9.99 placeholder values for the NVIDIA RTX PRO 6000 Blackwell Server Edition in HW_REGISTRY (packages/constants/src/gpu-keys.ts) with the official values from the SemiAnalysis AI Cloud TCO model:

Field	Meaning	Old	New
power	All-in kW per GPU	9.99	0.975
costh	Owning – Hyperscaler ($/GPU/hr)	9.99	0.43
costn	Owning – Neocloud Giant ($/GPU/hr)	9.99	0.676
costr	3 Year Rental ($/GPU/hr)	9.99	0.52

With real numbers in place, $/token and per-MW metrics for this SKU are now meaningful, so the placeholder TODO comment is removed. TDP stays at the datasheet 600 W max board power.

No presentation/UI code changes — HW_REGISTRY is the single source of truth and all cost/power metrics derive from it at render time, for both official runs and ?unofficialrun= overlays (the overlay path reads the same registry, so no separate overlay work applies).

Testing

• pnpm typecheck, pnpm lint, pnpm fmt — pass
• pnpm test:unit — 2810 tests pass (no test pinned the placeholder values)

中文说明

将 HW_REGISTRY（packages/constants/src/gpu-keys.ts）中 NVIDIA RTX PRO 6000 Blackwell Server Edition 的 9.99 占位值替换为 SemiAnalysis AI Cloud TCO 模型的正式数值：

字段	含义	旧值	新值
power	每 GPU 整机功耗 (kW)	9.99	0.975
costh	自有 – Hyperscaler ($/GPU/hr)	9.99	0.43
costn	自有 – Neocloud Giant ($/GPU/hr)	9.99	0.676
costr	3 年租赁 ($/GPU/hr)	9.99	0.52

正式数值就位后，该 SKU 的 $/token 与每 MW 指标随之有效，因此移除了占位 TODO 注释。TDP 保持数据手册标称的 600 W 最大板卡功耗。

不涉及展示层/UI 代码改动 —— HW_REGISTRY 是唯一数据源，所有成本/功耗指标（包括官方运行与 ?unofficialrun= 覆盖层）都在渲染时从同一注册表推导，无需单独的覆盖层适配。

测试：pnpm typecheck、pnpm lint、pnpm fmt 通过；pnpm test:unit 2810 个测试全部通过（无测试固定占位值）。

🤖 Generated with Claude Code

---

Note

Low Risk
Single-registry data update with no logic changes; only RTX PRO 6000 derived metrics change.

Overview
Updates rtx6000pro in HW_REGISTRY so RTX PRO 6000 Blackwell metrics use SemiAnalysis AI Cloud TCO values instead of 9.99 placeholders.

power is now 0.975 kW/GPU (all-in); costh, costn, and costr are 0.43, 0.676, and 0.52 $/GPU/hr respectively. tdp stays 600 W. The TODO about pending TCO numbers is removed and the block comment now states that power and costs come from the TCO model.

Because everything that derives $/token and per-MW figures reads this registry, RTX PRO 6000 comparisons and overlays pick up the new economics without separate UI changes.

^{Reviewed by Cursor Bugbot for commit ca59df2. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
inferencemax-app	Building	Preview, Comment	Jul 27, 2026 2:17am