裸金属 GPU 规格
适用于任何 AI 工作流的 GPU：从模型训练到推理。
| Model | Location | QTY | Pricing($/Gpu/Hour) | Network |
|---|---|---|---|---|
| H100 | EU | 200 | $1.95 | IB |
| Singapore | 32 | $2.10 | IB | |
| US | 16 | $2.10 | IB | |
| H200 | US | 128 | $2.35 | RoCe |
| Japan | 8 | $2.40 | IB | |
| EU | 16 | $2.40 | IB | |
| Singapore | 8 | $2.40 | IB | |
| US | 8 | $2.40 | IB | |
| GB200 | Malaysia | 8 | $4.5 | IB |
| A100 | US | 64 | $1.35 | / |
H100
LocationEU
QTY200
Pricing$1.95/Gpu/Hour
NetworkIB
LocationSingapore
QTY32
Pricing$2.10/Gpu/Hour
NetworkIB
LocationUS
QTY16
Pricing$2.10/Gpu/Hour
NetworkIB
H200
LocationUS
QTY128
Pricing$2.35/Gpu/Hour
NetworkRoCe
LocationJapan
QTY8
Pricing$2.40/Gpu/Hour
NetworkIB
LocationEU
QTY16
Pricing$2.40/Gpu/Hour
NetworkIB
LocationSingapore
QTY8
Pricing$2.40/Gpu/Hour
NetworkIB
LocationUS
QTY8
Pricing$2.40/Gpu/Hour
NetworkIB
GB200
LocationMalaysia
QTY8
Pricing$4.5/Gpu/Hour
NetworkIB
A100
LocationUS
QTY64
Pricing$1.35/Gpu/Hour
Network/
高效基础设施
全面访问 NVIDIA H100/H200/A100/4090 GPU 和 x86 CPU，无虚拟化带来的性能限制，加速 AI 模型训练。
高性能网络
Atlas 的 NVIDIA HGX H100 和 H200 超算集群采用 NVIDIA InfiniBand 网络，采用轨道优化设计，网络性能损耗极低。
