跳到主要内容Skip to main content

← 返回硬件算力← Back to Compute

整机Systems未公布单价No public price

NVIDIA GB300 NVL72NVIDIA GB300 NVL72

未公布单价No public price未公布单价No public price · 官方页未给公开单价;这类产品通常按整机、云实例或询价成交,不拿经销商价冒充官方价。No public unit price on the official page; these usually sell as systems, cloud instances or on request, so dealer quotes are not presented as official.

回答「Blackwell Ultra 的机柜级形态给多少显存、带宽和 FP4」。GB300 NVL72 把 72 颗 B300 与 36 颗 Grace CPU 做进一个液冷机柜:20TB HBM3e、576TB/s 聚合带宽、NVLink 5 域内 130TB/s,官方 FP4 表列 1440 / 1080 PFLOPS 两档。定位是 test-time scaling 与推理,不是单卡;买整柜和买 8 卡服务器的账要分开算。Answers what the rack-scale Blackwell Ultra form delivers in memory, bandwidth and FP4. GB300 NVL72 puts 72 B300 GPUs and 36 Grace CPUs into one liquid-cooled rack: 20TB of HBM3e, up to 576TB/s of aggregate bandwidth and 130TB/s of NVLink 5 domain bandwidth, with the official FP4 table listing 1,440 / 1,080 PFLOPS. It is positioned for test-time scaling and inference, not as a single GPU; the economics of a full rack and an 8-GPU server are different purchases.

关键参数Key specs

configuration
72 × Blackwell Ultra B300 + 36 × Grace CPU
NVLink bandwidth
130 TB/s
fast memory
37 TB
GPU memory
20 TB HBM3e
GPU bandwidth
up to 576 TB/s
CPU memory
17 TB LPDDR5X / 14 TB/s
CPU cores
2,592 Arm Neoverse V2
FP4 Tensor Core
1,440 / 1,080 PFLOPS (official table)
FP8/FP6
720 PFLOPS
FP16/BF16
360 PFLOPS
cooling
fully liquid-cooled

官方原文怎么说What the source says

72 NVIDIA Blackwell Ultra GPUs, 36 NVIDIA Grace CPUs
1 https://www.nvidia.com/en-eu/data-center/gb300-nvl72/
GPU Memory | Bandwidth 20 TB | Up to 576 TB/s
2 https://www.nvidia.com/en-eu/data-center/gb300-nvl72/
FP4 Tensor Core 1440 | 1080 PFLOPS
3 https://www.nvidia.com/en-eu/data-center/gb300-nvl72/

资料与许可Source and licence

资料时间Source checked
2026-11-19
许可 / 类型Licence / kind
official-spec-page
许可说明Licence evidence
https://www.nvidia.com/en-eu/data-center/gb300-nvl72/
类目Category
整机Systems
状态Status
已出货Shipping
适合Use cases
集群Cluster
价格口径Price basis
未公布单价No public price · not-published

同类目还有More in this category

下一步Next step

去官网看原页Open the official page →

← 返回硬件算力← Back to Compute