atlas
p GPU training
6 Generation 6
-b200 B200
.
48xlarge 192 vCPU · 2 TiB

$137/hr

On-demand in US-East

192 vCPU

Xeon Emerald Rapids, 3.2 GHz

2 TiB

10.7 GiB per vCPU

31 TB NVMe

Local storage

3200 Gbps

Network bandwidth

100 Gbps

EBS bandwidth

x86_64

Architecture

Every way to pay

Prices in US-East. Bars are relative to on-demand.

PlanMonthlyYearlySaves

On-demand

Pay by the second, no commitment

$99,805$1,197,662

Reserved, 1 yr

No upfront, this size and region

Not sold in this region

Reserved, 3 yr

No upfront, this size and region

Not sold in this region

What will it cost me?

Enter how you plan to run it.

× p6-b200.48xlarge, on-demand

/mo

a year

You run these most of the time. A 1-year reservation bills every hour, and would still cost /mo.

Same usage on each planMonthlyYearly
On-demand

Reservations bill every hour of the month, whether the instance runs or not.

Other sizes in this family

Price roughly doubles with each step. Pick a size to open it.

Price in every region

Cheapest in US East (N. Virginia). Moving from US-East would save $199,610 a year per instance.

Not offered in N. California , Central , Calgary , Central , São Paulo , Ireland , London , Frankfurt , Zurich , Milan , Spain , Hyderabad , Singapore , Sydney , Jakarta , Melbourne , Malaysia , New Zealand , Thailand , Tokyo , Seoul , Osaka , Hong Kong , Taipei , UAE , Bahrain , Tel Aviv , Cape Town .

Specifications

API name
p6-b200.48xlarge
Processor
Intel Xeon Scalable (Emerald Rapids)
Local storage
8 × 3.8 TB NVMe
Family
GPU instance
Clock speed
3.2 GHz
Network
3200 Gbps
Generation
Gen 6
Architecture
x86_64
EBS bandwidth
100 Gbps
vCPUs
192
Memory
2 TiB
GPUs
8 · 1432 GB HBM3e