AWS Trainium3AWS Trainium3
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「AWS 自研训练芯片最新一代给什么」。Trainium3 是 AWS 首颗 3nm AI 芯片,单颗 2.52 PF FP8、144GB HBM3e、4.9TB/s;Trn3 UltraServer 可扩到 144 颗,合计 362 PF、20.7TB HBM3e、706TB/s,官方称比 Trn2 UltraServer 最多 4.4 倍性能、4 倍每瓦性能。只能通过 EC2 / Neuron SDK 使用,不是通用 GPU;选型时要连实例价格与网络一起算。Answers what AWS’s latest in-house training chip offers. Trainium3 is AWS’s first 3nm AI chip: 2.52 PF at FP8, 144GB of HBM3e and 4.9TB/s per chip; a Trn3 UltraServer scales to 144 chips for 362 PF, 20.7TB of HBM3e and 706TB/s, and AWS claims up to 4.4x the performance and 4x the performance-per-watt of Trn2 UltraServers. It is reachable only through EC2 and the Neuron SDK, not as a general-purpose GPU; compare instance pricing and networking together with the specs.
关键参数Key specs
- process
- 3 nm
- FP8 compute
- 2.52 PFLOPS per chip
- memory
- 144 GB HBM3e per chip
- memory bandwidth
- 4.9 TB/s per chip
- interconnect
- NeuronSwitch-v1 / NeuronLink-v4, 2 TB/s per chip
- UltraServer
- up to 144 chips, 362 PFLOPS MXFP8, 20.7 TB HBM3e, 706 TB/s
- availability
- Amazon EC2 Trn3 UltraServers
- software
- AWS Neuron SDK
官方原文怎么说What the source says
144 GB HBM3e per chip, 4.9 TB/s bandwidth - 1.7x higher than Trainium2
Up to 2x MXFP8 compute throughput compared to Trainium2
Trn3 UltraServers scale up to 144 Trainium chips, delivering up to 362 MXFP8 PFLOPs, 20.7 TB of HBM3e, and 706 TB/s of aggregate memory bandwidth
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://aws.amazon.com/ai/machine-learning/trainium/
- 类目Category
- 云算力Cloud compute
- 状态Status
- 已发布 / 未出货Announced
- 适合Use cases
- 按租用Rent · 训练Training
- 价格口径Price basis
- 未公布单价No public price · not-published