Products · GPU产品 · GPU

How much power per chip and per rack, and at what coolant temperature?每颗芯片、每个机柜多大功率,冷却液多少度?

Power per AI chip heads past 1.4 kW; NVIDIA designs racks for 45 °C coolant.单颗 AI 芯片功率迈过 1.4 kW;NVIDIA 按 45 °C 冷却液设计机柜。

Power boundaries differ between points, so read the direction, not the ratio.各点的功率口径不同,看方向,不要算倍数。

Official sources checked 2026-10-07.官方来源核查 2026-10-07。

Chip and rack power ratings: design envelopes, not measured load芯片与机柜额定功率:设计上限,不是实测负载
202420252026202720282029 Chip or module power, W芯片或模块功率,W05001,0001,5002,000H200 SXM 700 WH200 SXM 700 WB200 1,000 WB200 1,000 WGB300 GPU 1,400 WGB300 GPU 1,400 WMI300X 750 WMI300X 750 WMI325X 1,000 WMI325X 1,000 WMI355X 1,400 WMI355X 1,400 WMaia 200 750 WMaia 200 750 WMTIA 300 800 WMTIA 300 800 WMTIA 400 1,200 WMTIA 400 1,200 WMTIA 450 1,400 WMTIA 450 1,400 WMTIA 500 1,700 WMTIA 500 1,700 WGaudi 3 HL-325L 900 WGaudi 3 HL-325L 900 W 8-GPU server, maximum system power (AC input), kW (NVIDIA)8-GPU 服务器最大系统功率(交流输入),kW(NVIDIA)05101520DGX H200 10.2 kWDGX H200 10.2 kWDGX B200 14.3 kWDGX B200 14.3 kWDGX B300 15 kWDGX B300 15 kW Designed rack power and cabinet design envelope, kW (NVIDIA): do not read 135 → 330 kW as growth between generations设计机柜功率与机柜设计包络,kW(NVIDIA):不要把 135 → 330 kW 读成代际增长0100200300400GB200 NVL72 120 kWGB200 NVL72 120 kWGB300 NVL72 135 kWGB300 NVL72 135 kWVera Rubin NVL72 330 kW (DSX design envelope)Vera Rubin NVL72 330 kW(DSX 设计包络)

Chip or module power芯片或模块功率

  • H200 SXM 700 W GPU TDP, up toGPU TDP,最高
  • MI300X 750 W OAM module, maximum TBPOAM 模块,最大 TBP
  • B200 (DGX B200) 1,000 W GPU TGP, maximum (default cap 700 W)GPU TGP,最大值(默认上限 700 W)
  • Gaudi 3 HL-325L 900 W OAM module TDPOAM 模块 TDP
  • MI325X 1,000 W OAM module TBP, peakOAM 模块 TBP,峰值
  • GB300 GPU 1,400 W GPU maximum in the NVL72 rack budgetNVL72 机柜功率预算中的单 GPU 最大值
  • MI355X 1,400 W OAM module TBPOAM 模块 TBP
  • Maia 200 750 W SoC TDP envelopeSoC TDP 上限
  • MTIA 300 800 W Module TDP模块 TDP
  • MTIA 400 1,200 W Module TDP模块 TDP
  • MTIA 450 1,400 W Module TDP, planned模块 TDP,计划
  • MTIA 500 1,700 W Module TDP, planned模块 TDP,计划

8-GPU server, maximum system power (NVIDIA)8-GPU 服务器最大系统功率(NVIDIA)

  • DGX H200 10.2 kW 8-GPU server, maximum system power (AC input)8 GPU 服务器,系统最大功率(交流输入)
  • DGX B200 14.3 kW 8-GPU server, maximum system power (AC input)8 GPU 服务器,系统最大功率(交流输入)
  • DGX B300 15 kW 8-GPU server, maximum system power (AC input); NVIDIA also gives about 14 kW on the product page and 14.5 kW power consumption in the user guide8 GPU 服务器,系统最大功率(交流输入);NVIDIA 产品页另写约 14 kW,用户指南另列功耗 14.5 kW

Designed rack power and cabinet design envelope (NVIDIA)设计机柜功率与机柜设计包络(NVIDIA)

  • GB200 NVL72 120 kW Designed rack power (Mission Control); the DGX GB user guide gives about 120 kW for the GB rack family设计机柜功率(Mission Control);DGX GB 用户指南对 GB 机柜系列给出约 120 kW
  • GB300 NVL72 135 kW Designed rack power (NVIDIA's reference architecture gives up to 142 kW for the full rack)设计机柜功率(NVIDIA 参考架构给出整柜最高 142 kW)
  • Vera Rubin NVL72 (DSX) 330 kW Cabinet TDP of the DSX facility reference design (a design envelope)DSX 设施参考设计的机柜 TDP(设计包络)

Shapes group similar power boundaries: 形状按相近的功率口径分组:GPU TDP, TGP or per-GPU budget maximum (NVIDIA)GPU TDP、TGP 或单 GPU 预算最大值(NVIDIA) OAM module board power (AMD) or TDP (Intel)OAM 模块板卡功率(AMD)或 TDP(Intel) SoC or module TDP (Microsoft, Meta)SoC 或模块 TDP(Microsoft、Meta) designed rack power (GB200, GB300)设计机柜功率(GB200、GB300) 8-GPU server, maximum system power, AC input (DGX H200, B200, B300)8 GPU 服务器,系统最大功率,交流输入(DGX H200、B200、B300) cabinet TDP of a facility reference design (Vera Rubin)设施参考设计的机柜 TDP(Vera Rubin)

Filled: disclosed specification. Hollow: official plan. No line joins the points because their boundaries differ. Each point is placed at the product's announcement or launch date; the value is today's official specification. Meta MTIA 450 and 500 sit at the start of their planned deployment year (early 2027; 2027), not at an announcement date. Hover or focus a point for details.实心:已公布规格。空心:官方计划。各点口径不同,所以不连线。每个点放在产品发布或上市日期;数值取当前官方规格。Meta MTIA 450 和 500 放在计划部署年份的起点(2027 年初;2027 年),不是发布日期。悬停或聚焦数据点可查看详情。

Chip and rack power ratings: design envelopes, not measured load芯片与机柜额定功率:设计上限,不是实测负载

Table表格

Date日期Company公司Product产品Power功率Power boundary功率口径Status状态Sources来源
2023-11NVIDIAH200 SXM700 WGPU TDP, up toGPU TDP,最高Disclosed已公布A1 A2
2024-03NVIDIAB200 (DGX B200)1,000 WGPU TGP, maximum (default cap 700 W)GPU TGP,最大值(默认上限 700 W)Disclosed已公布A3 A10
2025-03NVIDIAGB300 GPU1,400 WGPU maximum in the NVL72 rack budgetNVL72 机柜功率预算中的单 GPU 最大值Disclosed已公布A4 A8
2023-12AMDMI300X750 WOAM module, maximum TBPOAM 模块,最大 TBPDisclosed已公布A11 A12
2024-10AMDMI325X1,000 WOAM module TBP, peakOAM 模块 TBP,峰值Disclosed已公布A13
2025-06AMDMI355X1,400 WOAM module TBPOAM 模块 TBPDisclosed已公布A14
2026-01MicrosoftMaia 200750 WSoC TDP envelopeSoC TDP 上限Disclosed已公布A25
2026-03MetaMTIA 300800 WModule TDP模块 TDPDisclosed已公布A26
2026-03MetaMTIA 4001,200 WModule TDP模块 TDPDisclosed已公布A26
2027 earlyMetaMTIA 4501,400 WModule TDP, planned模块 TDP,计划Planned计划A26
2027MetaMTIA 5001,700 WModule TDP, planned模块 TDP,计划Planned计划A26
2024-04IntelGaudi 3 HL-325L900 WOAM module TDPOAM 模块 TDPDisclosed已公布A28 A29
2024-03NVIDIAGB200 NVL72120 kWDesigned rack power (Mission Control); the DGX GB user guide gives about 120 kW for the GB rack family设计机柜功率(Mission Control);DGX GB 用户指南对 GB 机柜系列给出约 120 kWDisclosed已公布A3 A34 A8
2025-03NVIDIAGB300 NVL72135 kWDesigned rack power (NVIDIA's reference architecture gives up to 142 kW for the full rack)设计机柜功率(NVIDIA 参考架构给出整柜最高 142 kW)Disclosed已公布A4 A8 A30
2026-01NVIDIAVera Rubin NVL72 (DSX)330 kWCabinet TDP of the DSX facility reference design (a design envelope)DSX 设施参考设计的机柜 TDP(设计包络)Disclosed已公布A5 A9
2023-11NVIDIADGX H20010.2 kW8-GPU server, maximum system power (AC input)8 GPU 服务器,系统最大功率(交流输入)Disclosed已公布A1 A31 A37
2024-03NVIDIADGX B20014.3 kW8-GPU server, maximum system power (AC input)8 GPU 服务器,系统最大功率(交流输入)Disclosed已公布A3 A32 A38
2025-03NVIDIADGX B30015 kW8-GPU server, maximum system power (AC input); NVIDIA also gives about 14 kW on the product page and 14.5 kW power consumption in the user guide8 GPU 服务器,系统最大功率(交流输入);NVIDIA 产品页另写约 14 kW,用户指南另列功耗 14.5 kWDisclosed已公布A4 A33 A35 A36

Sources来源

  1. A1 NVIDIA: H200 announcementA1 NVIDIA:H200 announcement 2023-11-13
  2. A2 NVIDIA: H200 official product specificationsA2 NVIDIA:H200 official product specifications
  3. A3 NVIDIA: Blackwell announcementA3 NVIDIA:Blackwell announcement 2024-03-18
  4. A4 NVIDIA: Blackwell Ultra announcementA4 NVIDIA:Blackwell Ultra announcement 2025-03-18
  5. A5 NVIDIA: Rubin launch and production statementA5 NVIDIA:Rubin launch and production statement 2026-01-05
  6. A8 NVIDIA: Mission Control 2.3.0 power budgeting FAQA8 NVIDIA:Mission Control 2.3.0 power budgeting FAQ 2026-10-12
  7. A9 NVIDIA: DSX Facilities Infrastructure Reference Design OverviewA9 NVIDIA:DSX Facilities Infrastructure Reference Design Overview 2026-08-19
  8. A10 NVIDIA: DGX B200 GPU power cappingA10 NVIDIA:DGX B200 GPU power capping 2026-09-08
  9. A11 AMD: MI300X availability announcementA11 AMD:MI300X availability announcement 2023-12-06
  10. A12 AMD: MI300X datasheetA12 AMD:MI300X datasheet
  11. A13 AMD: MI325X product specificationsA13 AMD:MI325X product specifications
  12. A14 AMD: MI355X product specificationsA14 AMD:MI355X product specifications
  13. A25 Microsoft: Maia 200 deployment and engineeringA25 Microsoft:Maia 200 deployment and engineering 2026-01-26
  14. A26 Meta: MTIA 300–500 roadmap and official specifications imageA26 Meta:MTIA 300–500 roadmap and official specifications image 2026-03-11
  15. A28 Intel: Gaudi 3 announcementA28 Intel:Gaudi 3 announcement 2024-04-09
  16. A29 Intel: Gaudi 3 technical overview version 2.6A29 Intel:Gaudi 3 technical overview version 2.6
  17. A30 NVIDIA: Enterprise Reference Architecture with GB300 NVL72 and Spectrum-X networking (2026 March)A30 NVIDIA:Enterprise Reference Architecture with GB300 NVL72 and Spectrum-X networking (2026 March) 2026-03-26
  18. A31 NVIDIA: DGX H100/H200 user guide: power specificationsA31 NVIDIA:DGX H100/H200 user guide: power specifications
  19. A32 NVIDIA: DGX B200 user guide: power specificationsA32 NVIDIA:DGX B200 user guide: power specifications
  20. A33 NVIDIA: DGX B300 user guide: power specificationsA33 NVIDIA:DGX B300 user guide: power specifications
  21. A34 NVIDIA: DGX GB rack-scale systems user guide: hardware (covers GB200 and GB300 racks)A34 NVIDIA:DGX GB rack-scale systems user guide: hardware (covers GB200 and GB300 racks) 2026-09-17
  22. A35 NVIDIA: DGX B300 product pageA35 NVIDIA:DGX B300 product page
  23. A36 NVIDIA: DGX SuperPOD reference architecture with DGX B300 (XDR)A36 NVIDIA:DGX SuperPOD reference architecture with DGX B300 (XDR)
  24. A37 NVIDIA: DGX SuperPOD reference architecture with DGX H200: componentsA37 NVIDIA:DGX SuperPOD reference architecture with DGX H200: components
  25. A38 NVIDIA: DGX SuperPOD reference architecture with DGX B200: componentsA38 NVIDIA:DGX SuperPOD reference architecture with DGX B200: components
  26. NVIDIA: Maximizing AI Factory Performance per Watt with NVIDIA DSX MaxLPS (provisioned rack power, cited in the note)NVIDIA:Maximizing AI Factory Performance per Watt with NVIDIA DSX MaxLPS(配置机柜功率,见图注) 2026-08-24

Method and notes方法和说明

Official sources checked 2026-10-07; rack note rechecked 2026-10-17. Boundaries differ (GPU TDP or TGP, OAM board power, SoC or module TDP; rack power, 8-GPU server maximum system power or cabinet TDP), so read the direction, not exact ratios between points; none is measured load. DGX H200, B200 and B300 are single 8-GPU servers at the maximum system power (AC input) in NVIDIA's user guides (the 10.2 kW row is shared by DGX H100 and H200); they sit at the date of the HGX platform announcement, which names the 8-GPU boards rather than the DGX systems. For cooling, the same guides give heat output of 38,557 BTU/hr (about 11.3 kW) for DGX H200, 48,794 BTU/hr (14.3 kW) for B200 and 49,476 BTU/hr (14.5 kW) for B300. How many servers go in a rack is a design choice: NVIDIA's DGX SuperPOD references put two DGX H200 or B200 systems in a rack, and four DGX B300 in a rack at about 56 kW. For DGX B300, NVIDIA also gives about 14 kW (product page) and 14.5 kW power consumption (user guide); the chart uses the guide's 15 kW system maximum, the same basis as H200 and B200. GB200 NVL72 120 kW and GB300 NVL72 135 kW are the designed rack power in the same NVIDIA Mission Control table. NVIDIA's GB300 reference architecture also says a full rack requires "up to 142 kW", a different boundary. The Vera Rubin 330 kW is the cabinet TDP of NVIDIA's DSX facility reference design, not a rack rating comparable with the GB200 and GB300 rack power. In a separate NVIDIA test on inference workloads (different models), provisioned rack power before NVIDIA's MaxLPS power management was 136 kW for Vera Rubin NVL72 and 125 kW for GB200 NVL72. Do not read 135 → 330 kW as growth between generations. We found no official absolute chip or rack power for Google, AWS or Huawei, and no verified official value yet for the Rubin GPU, the AMD Helios rack or later NVIDIA racks; missing is not zero.官方来源核查 2026-10-07;机柜说明于 2026-10-17 复核。各点口径不同(GPU TDP 或 TGP、OAM 板卡功率、SoC 或模块 TDP;机柜功率、8 GPU 服务器系统最大功率或机柜 TDP),只宜看方向,不宜计算点与点之间的精确倍数;都不是实测负载。DGX H200、B200 和 B300 是单台 8 GPU 服务器,数值取 NVIDIA 用户指南中的系统最大功率(交流输入),其中 10.2 kW 为 DGX H100 与 H200 共用;它们放在 HGX 平台的发布日期,该发布提到的是 8 GPU 主板而不是 DGX 整机。用于冷却设计时,同一批指南给出的散热量为:DGX H200 38,557 BTU/hr(约 11.3 kW),B200 48,794 BTU/hr(14.3 kW),B300 49,476 BTU/hr(14.5 kW)。一个机柜放几台由设计决定:NVIDIA DGX SuperPOD 参考设计每柜放 2 台 DGX H200 或 B200,DGX B300 每柜放 4 台、约 56 kW。DGX B300 另有两个 NVIDIA 数值:产品页约 14 kW,用户指南功耗 14.5 kW;图中取指南的 15 kW 系统最大值,与 H200、B200 口径相同。GB200 NVL72 的 120 kW 和 GB300 NVL72 的 135 kW 取自 NVIDIA Mission Control 同一张表中的设计机柜功率。NVIDIA GB300 参考架构另写整柜需求“最高 142 kW”,口径不同。Vera Rubin 的 330 kW 是 NVIDIA DSX 设施参考设计的机柜 TDP,不能与 GB200、GB300 的机柜功率直接比较。NVIDIA 另一项推理负载测试(所用模型不同)中,启用 MaxLPS 功率管理前的配置机柜功率为 Vera Rubin NVL72 136 kW、GB200 NVL72 125 kW。不要把 135 → 330 kW 理解为代际增长。未找到 Google、AWS 和华为官方公布的芯片或机柜绝对功率;Rubin GPU、AMD Helios 机柜和 NVIDIA 后续机柜暂无经核实的官方数值;缺失不等于零。

Chip roadmap: AI chips and platforms from eight vendors, 2023 Q4 to 2029 Q3芯片路线图:八家厂商的 AI 芯片与平台,2023 年第四季度至 2029 年第三季度
202420252026202720282029 Checked 2026-10-12核查 2026-10-12 NVIDIAH200H200Blackwell GB200Blackwell GB200Blackwell Ultra GB300Blackwell Ultra GB300Vera RubinVera RubinVera Rubin NVL72Vera Rubin NVL72Rubin partner productsRubin partner products800 V DC racks800 V DC racksRubin UltraRubin UltraFeynmanFeynman AMDMI300XMI300XMI325XMI325XMI355XMI355XMI455X / HeliosMI455X / HeliosMI500MI500MI600MI600 GoogleTrillium (TPU v6)Trillium (TPU v6)Ironwood (TPU v7)Ironwood (TPU v7)TPU 8t / 8iTPU 8t / 8i AWSTrainium2Trainium2Trainium3Trainium3Trainium4Trainium4 MicrosoftMaia 100Maia 100Maia 200Maia 200 MetaMTIA 300MTIA 300MTIA 400MTIA 400MTIA 450MTIA 450MTIA 500MTIA 500 HuaweiAtlas 950 SuperPoDAtlas 950 SuperPoDAscend 960DTAscend 960DTAscend 960PRAscend 960PRAscend 970Ascend 970Ascend 980Ascend 980 IntelGaudi 3Gaudi 3Crescent IslandCrescent Island
  1. NVIDIA
    • H200H200
    • Blackwell GB200Blackwell GB200
    • Blackwell Ultra GB300Blackwell Ultra GB300
    • Vera RubinVera Rubin
    • Vera Rubin NVL72Vera Rubin NVL72 (plan)(计划)
    • Rubin partner productsRubin partner products (plan)(计划)
    • 800 V DC racks800 V DC racks (plan)(计划)
    • Rubin UltraRubin Ultra (plan)(计划)
    • FeynmanFeynman (plan)(计划)
  2. AMD
    • MI300XMI300X
    • MI325XMI325X
    • MI355XMI355X
    • MI455X / HeliosMI455X / Helios
    • MI500MI500 (plan)(计划)
    • MI600MI600 (plan)(计划)
  3. Google
    • Trillium (TPU v6)Trillium (TPU v6)
    • Ironwood (TPU v7)Ironwood (TPU v7)
    • TPU 8t / 8iTPU 8t / 8i
  4. AWS
    • Trainium2Trainium2
    • Trainium3Trainium3
    • Trainium4Trainium4 (plan)(计划)
  5. Microsoft
    • Maia 100Maia 100
    • Maia 200Maia 200
  6. Meta
    • MTIA 300MTIA 300
    • MTIA 400MTIA 400
    • MTIA 450MTIA 450 (plan)(计划)
    • MTIA 500MTIA 500 (plan)(计划)
  7. Huawei
    • Atlas 950 SuperPoDAtlas 950 SuperPoD
    • Ascend 960DTAscend 960DT (plan)(计划)
    • Ascend 960PRAscend 960PR (plan)(计划)
    • Ascend 970Ascend 970 (plan)(计划)
    • Ascend 980Ascend 980 (plan)(计划)
  8. Intel
    • Gaudi 3Gaudi 3
    • Crescent IslandCrescent Island (plan)(计划)

Solid capsule: happened (announcement, availability or production, as the event says). Dashed capsule: a stated plan, placed at the start of its window. Hover or focus a capsule for the event and its date precision.实线胶囊:已发生(发布、供货或量产,以事件描述为准)。虚线胶囊:公司公布的计划,放在其时间窗口起点。悬停或聚焦胶囊可查看事件及日期精度。

Chip roadmap: AI chips and platforms from eight vendors, 2023 Q4 to 2029 Q3芯片路线图:八家厂商的 AI 芯片与平台,2023 年第四季度至 2029 年第三季度

Table表格

Date or window日期或窗口Company公司Product产品Event事件Status状态Source来源
2023-11-132023-11-13NVIDIAH200H200NVIDIA announces the H200 GPUNVIDIA 发布 H200 GPUHappened已发生link链接
2023-11-152023-11-15MicrosoftMaia 100Maia 100Microsoft announces its first Azure Maia AI acceleratorMicrosoft 发布首代 Azure Maia AI 加速器Happened已发生link链接
2023-12-062023-12-06AMDMI300XMI300XAMD announces the Instinct MI300X is availableAMD 宣布 Instinct MI300X 可供货Happened已发生link链接
2024-03-182024-03-18NVIDIABlackwell GB200Blackwell GB200NVIDIA announces the Blackwell platform and the GB200 superchipNVIDIA 发布 Blackwell 平台和 GB200 超级芯片Happened已发生link链接
2024-04-092024-04-09IntelGaudi 3Gaudi 3Intel unveils Gaudi 3; the announcement is not its general availability dateIntel 发布 Gaudi 3;发布不等于正式供货Happened已发生link链接
2024-10-102024-10-10AMDMI325XMI325XAMD Instinct MI325X launch dateAMD Instinct MI325X 产品发布日Happened已发生link链接
2024-12-032024-12-03AWSTrainium2Trainium2AWS makes EC2 Trn2 instances with Trainium2 availableAWS 推出搭载 Trainium2 的 EC2 Trn2 实例Happened已发生link链接
2024-12-112024-12-11GoogleTrillium (TPU v6)Trillium (TPU v6)Google Cloud makes Trillium, its sixth-generation TPU, generally availableGoogle Cloud 第六代 TPU Trillium 正式商用(GA)Happened已发生link链接
2025-03-182025-03-18NVIDIABlackwell Ultra GB300Blackwell Ultra GB300NVIDIA announces Blackwell Ultra and the GB300 NVL72NVIDIA 发布 Blackwell Ultra 和 GB300 NVL72Happened已发生link链接
2025-04-092025-04-09GoogleIronwood (TPU v7)Ironwood (TPU v7)Google announces Ironwood, its seventh-generation TPU; general availability came laterGoogle 发布第七代 TPU Ironwood;正式商用在之后Happened已发生link链接
2025-06-122025-06-12AMDMI355XMI355XAMD Instinct MI355X launch dateAMD Instinct MI355X 产品发布日Happened已发生link链接
2025-12-022025-12-02AWSTrainium3Trainium3AWS makes Trainium3 UltraServers generally availableAWS Trainium3 UltraServer 正式商用(GA)Happened已发生link链接
2026-01-052026-01-05NVIDIAVera RubinVera RubinNVIDIA announces the Rubin platform and says it is in full production; partner products are a separate planNVIDIA 发布 Rubin 平台并称已全面投产;合作伙伴产品另有计划Happened已发生link链接
2026-01-262026-01-26MicrosoftMaia 200Maia 200Microsoft introduces Maia 200, already deployed in its US Central regionMicrosoft 推出 Maia 200,已部署在 US Central 区域Happened已发生link链接
2026-03-112026-03-11MetaMTIA 300MTIA 300Meta says MTIA 300 is in production (start date not disclosed)Meta 称 MTIA 300 已在生产(开始日期未披露)Happened已发生link链接
2026-03-112026-03-11MetaMTIA 400MTIA 400Meta says MTIA 400 has finished lab testing and is on the path to deploymentMeta 称 MTIA 400 已完成实验室测试,正在走向部署Happened已发生link链接
2026-04-222026-04-22GoogleTPU 8t / 8iTPU 8t / 8iGoogle introduces TPU 8t for training and TPU 8i for inference; general availability targeted later in 2026Google 推出训练用 TPU 8t 和推理用 TPU 8i;目标 2026 年稍后正式商用Happened已发生link链接
2026 H22026 年下半年IntelCrescent IslandCrescent IslandIntel expects customer sampling of the Crescent Island GPU in 2026 H2; sampling is not a commercial releaseIntel 预计 2026 年下半年向客户送样 Crescent Island GPU;送样不是商业发布Planned (announced 2025-10-14)计划(2025-10-14 公布)link链接
2026 H22026 年下半年NVIDIAVera Rubin NVL72Vera Rubin NVL72Vera Rubin NVL72: 330 kW facility design TDP, 45 °C design point; ships in 2026 H2Vera Rubin NVL72:设施设计 TDP 330 kW,45 °C 设计点;2026 年下半年出货Planned (announced 2026-03)计划(2026-03 公布)link链接
2026 H22026 年下半年NVIDIARubin partner productsRubin partner productsNVIDIA targets Rubin-based products from partners in 2026 H2NVIDIA 目标 2026 年下半年由合作伙伴推出基于 Rubin 的产品Planned (announced 2026-01-05)计划(2026-01-05 公布)link链接
2026-07-232026-07-23AMDMI455X / HeliosMI455X / HeliosAMD launches the MI455X and the Helios rack and says Helios is in productionAMD 发布 MI455X 和 Helios 机柜,并称 Helios 已在生产Happened已发生link链接
2026-09-182026-09-18HuaweiAtlas 950 SuperPoDAtlas 950 SuperPoDHuawei says Atlas 950 SuperPoD clusters are in commercial build; not the Ascend 950 chip release dateHuawei 称 Atlas 950 超节点集群正在商业建设;不是昇腾 950 芯片上市日Happened已发生link链接
20272027 年AMDMI500MI500AMD Instinct MI500 Series: stated as coming in 2027AMD Instinct MI500 系列:官方称 2027 年推出Planned (announced 2026-07-23)计划(2026-07-23 公布)link链接
20272027 年AWSTrainium4Trainium4Amazon expects Trainium4 deliveries to start in 2027Amazon 预计 Trainium4 于 2027 年开始交付Planned (announced 2026-02-05)计划(2026-02-05 公布)link链接
2027 Q12027 年第一季度HuaweiAscend 960DTAscend 960DTHuawei plans Ascend 960DT for 2027 Q1 (three quarters earlier than before)Huawei 计划 2027 年第一季度推出昇腾 960DT(比此前提前三个季度)Planned (announced 2026-09-18)计划(2026-09-18 公布)link链接
early 20272027 年初MetaMTIA 450MTIA 450Meta plans mass deployment of MTIA 450 in early 2027Meta 计划 2027 年初大规模部署 MTIA 450Planned (announced 2026-03-11)计划(2026-03-11 公布)link链接
20272027 年MetaMTIA 500MTIA 500Meta plans MTIA 500 for 2027Meta 计划 2027 年推出 MTIA 500Planned (announced 2026-03-11)计划(2026-03-11 公布)link链接
20272027 年NVIDIA800 V DC racks800 V DC racksNVIDIA 800 VDC architecture for 1 MW racks and beyond (2027)NVIDIA 800 VDC 架构,面向 1 MW 及以上机柜(2027 年)Planned (announced 2025-05)计划(2025-05 公布)link链接
2027 Q32027 年第三季度HuaweiAscend 960PRAscend 960PRHuawei plans Ascend 960PR for 2027 Q3 (one quarter earlier than before)Huawei 计划 2027 年第三季度推出昇腾 960PR(比此前提前一个季度)Planned (announced 2026-09-18)计划(2026-09-18 公布)link链接
2027 H22027 年下半年NVIDIARubin UltraRubin UltraNVIDIA plans Rubin Ultra systems for 2027 H2NVIDIA 计划 2027 年下半年推出 Rubin Ultra 系统Planned (announced 2025-03)计划(2025-03 公布)link链接
20282028 年AMDMI600MI600AMD Instinct MI600 Series: stated as coming in 2028AMD Instinct MI600 系列:官方称 2028 年推出Planned (announced 2026-07-23)计划(2026-07-23 公布)link链接
20282028 年HuaweiAscend 970Ascend 970Huawei plans Ascend 970 for 2028Huawei 计划 2028 年推出昇腾 970Planned (announced 2026-09-18)计划(2026-09-18 公布)link链接
20282028 年NVIDIAFeynmanFeynmanFeynman (2028); no rack power or water temperature publishedFeynman(2028 年);未公布机柜功率和水温Planned (announced 2025-03)计划(2025-03 公布)link链接
20292029 年HuaweiAscend 980Ascend 980Huawei plans Ascend 980 for 2029Huawei 计划 2029 年推出昇腾 980Planned (announced 2026-09-18)计划(2026-09-18 公布)link链接

Sources来源

  1. H200: NVIDIA announces the H200 GPUH200:NVIDIA 发布 H200 GPU
  2. Maia 100: Microsoft announces its first Azure Maia AI acceleratorMaia 100:Microsoft 发布首代 Azure Maia AI 加速器
  3. MI300X: AMD announces the Instinct MI300X is availableMI300X:AMD 宣布 Instinct MI300X 可供货
  4. Blackwell GB200: NVIDIA announces the Blackwell platform and the GB200 superchipBlackwell GB200:NVIDIA 发布 Blackwell 平台和 GB200 超级芯片
  5. Gaudi 3: Intel unveils Gaudi 3; the announcement is not its general availability dateGaudi 3:Intel 发布 Gaudi 3;发布不等于正式供货
  6. MI325X: AMD Instinct MI325X launch dateMI325X:AMD Instinct MI325X 产品发布日
  7. Trainium2: AWS makes EC2 Trn2 instances with Trainium2 availableTrainium2:AWS 推出搭载 Trainium2 的 EC2 Trn2 实例
  8. Trillium (TPU v6): Google Cloud makes Trillium, its sixth-generation TPU, generally availableTrillium (TPU v6):Google Cloud 第六代 TPU Trillium 正式商用(GA)
  9. Blackwell Ultra GB300: NVIDIA announces Blackwell Ultra and the GB300 NVL72Blackwell Ultra GB300:NVIDIA 发布 Blackwell Ultra 和 GB300 NVL72
  10. Ironwood (TPU v7): Google announces Ironwood, its seventh-generation TPU; general availability came laterIronwood (TPU v7):Google 发布第七代 TPU Ironwood;正式商用在之后
  11. MI355X: AMD Instinct MI355X launch dateMI355X:AMD Instinct MI355X 产品发布日
  12. Trainium3: AWS makes Trainium3 UltraServers generally availableTrainium3:AWS Trainium3 UltraServer 正式商用(GA)
  13. Rubin partner products: NVIDIA targets Rubin-based products from partners in 2026 H2Rubin partner products:NVIDIA 目标 2026 年下半年由合作伙伴推出基于 Rubin 的产品
  14. Maia 200: Microsoft introduces Maia 200, already deployed in its US Central regionMaia 200:Microsoft 推出 Maia 200,已部署在 US Central 区域
  15. MTIA 500: Meta plans MTIA 500 for 2027MTIA 500:Meta 计划 2027 年推出 MTIA 500
  16. TPU 8t / 8i: Google introduces TPU 8t for training and TPU 8i for inference; general availability targeted later in 2026TPU 8t / 8i:Google 推出训练用 TPU 8t 和推理用 TPU 8i;目标 2026 年稍后正式商用
  17. Crescent Island: Intel expects customer sampling of the Crescent Island GPU in 2026 H2; sampling is not a commercial releaseCrescent Island:Intel 预计 2026 年下半年向客户送样 Crescent Island GPU;送样不是商业发布
  18. Vera Rubin NVL72: Vera Rubin NVL72: 330 kW facility design TDP, 45 °C design point; ships in 2026 H2Vera Rubin NVL72:Vera Rubin NVL72:设施设计 TDP 330 kW,45 °C 设计点;2026 年下半年出货
  19. MI600: AMD Instinct MI600 Series: stated as coming in 2028MI600:AMD Instinct MI600 系列:官方称 2028 年推出
  20. Ascend 980: Huawei plans Ascend 980 for 2029Ascend 980:Huawei 计划 2029 年推出昇腾 980
  21. Trainium4: Amazon expects Trainium4 deliveries to start in 2027Trainium4:Amazon 预计 Trainium4 于 2027 年开始交付
  22. 800 V DC racks: NVIDIA 800 VDC architecture for 1 MW racks and beyond (2027)800 V DC racks:NVIDIA 800 VDC 架构,面向 1 MW 及以上机柜(2027 年)
  23. Rubin Ultra: NVIDIA plans Rubin Ultra systems for 2027 H2Rubin Ultra:NVIDIA 计划 2027 年下半年推出 Rubin Ultra 系统
  24. Feynman: Feynman (2028); no rack power or water temperature publishedFeynman:Feynman(2028 年);未公布机柜功率和水温

Method and notes方法和说明

Drawn from the evidence base each time the site is built; the quarterly roadmap review adds and rechecks events. 34 events, sources checked 2026-10-07 to 2026-10-12. Where a vendor publishes a rack configuration, such as NVIDIA's Vera Rubin NVL72, its rack power and coolant temperature are the design point for the thermal equipment (see Products); most entries here publish neither.每次构建网站时从证据底座生成;季度路线图复核负责新增和复查事件。34 个事件,来源核查 2026-10-07 至 2026-10-12。厂商公布机柜配置时(例如 NVIDIA Vera Rubin NVL72),其机柜功率和冷却液温度就是热管理设备的设计点(见产品页);这里多数条目两者都未公布。