p5e.48xlarge pricing and specs in US West (Oregon)
192 vCPU · 2048 GiB · 3200 Gigabit · AMD EPYC 7R13 Processor
$63.2960/hour on-demand · $46,206/month (Linux, Shared)
p5e.48xlarge is a GPU machine-learning EC2 instance with 192 vCPU and 2048 GiB, generation 5 of the p family. On-demand it costs $63.2960 an hour, about $46,206 a month if it runs all month. A one-year commitment with nothing up front works out at $39.8765 an hour and pays for itself above roughly 459 hours a month; below that, on-demand is cheaper. The three-year options go down to $27.3439 an hour, 57% below on-demand. The newer p5.48xlarge offers the same size for $55.0400 an hour, 13% less. Across regions the cheapest place to run p5e.48xlarge is US East (Ohio) (us-east-2) at $63.2960 an hour, 0% below US West (Oregon).
Pricing by OS (hourly, USD, Shared)
| Option | Linux | Windows | RHEL | SUSE | Ubuntu Pro |
|---|---|---|---|---|---|
| On-Demand | $63.2960 | — | $65.1392 | $63.4210 | $63.6320 |
| Spot | $36.3360 | — | — | — | — |
| 1y Compute SP NU | $49.8456 | — | $51.6888 | $49.9055 | $50.1144 |
| 3y Compute SP NU | $46.5249 | — | $48.3681 | $46.5789 | $46.7438 |
| 1y EC2 SP NU | $39.8765 | — | $41.7197 | $39.9364 | $40.1453 |
| 3y EC2 SP NU | $27.3439 | — | $29.1871 | $27.3979 | $27.5628 |
Spot by AZ (7-day p50, Linux): us-west-2c $36.3360
Newer generations of the same size
Same family letter and size in a newer generation, ranked by price. “vCPU per $” uses vCPU as the performance proxy — newer chips usually add more, so it is a floor.
No Graviton equivalent at this size| Type | $/h | Price Δ | vCPU per $ Δ | Saves / mo |
|---|---|---|---|---|
| p5.48xlarge | $55.0400 | -13% | +15% | $6,027 |
| p5en.48xlarge | $63.2960 | 0.0% | 0.0% | $0 |
Cheapest regions for p5e.48xlarge
| Region | On-Demand $/h | vs us-west-2 | Saves / mo |
|---|---|---|---|
| US East (Ohio) us-east-2 | $63.2960 | 0.0% | $0 |
| US West (Oregon) us-west-2 | $63.2960 | current | — |
| Asia Pacific (Sydney) ap-southeast-2 | $82.2848 | +30% | $-13,862 |
| EU (London) eu-west-2 | $82.2848 | +30% | $-13,862 |
Caveat: instance price is only part of the bill. Cross-region and internet data transfer, latency to your users, data-residency rules and service availability can outweigh a few percent on compute.
Price history (us-west-2)
Daily snapshots since the catalog started tracking this type; a row is added whenever a headline price changes.
| Date | OD Linux | OD Windows | 1y RI | 3y RI | Spot p50 |
|---|---|---|---|---|---|
| 2026-11-06 | $63.2960 | — | — | — | $36.3360 |
| 2026-11-05 | $63.2960 | — | — | — | $36.4221 |
| 2026-11-04 | $63.2960 | — | — | — | $36.5021 |
All attributes
- vCPU
- 192
- Cores / threads per core
- 96 / 2
- Memory
- 2048 GiB
- Memory per vCPU
- 10.67 GiB
- Processor
- AMD EPYC 7R13 Processor
- Architecture
- x86_64
- Sustained clock
- 3.6 GHz
- Hypervisor
- nitro
- Generation
- Current
- Burstable
- No
- Bare metal
- No
- Free tier eligible
- No
- Network performance
- 3200 Gigabit
- Network baseline / burst
- 100.0 / 100.0 Gbps
- Max ENIs × IPv4 per ENI
- 64 × 50
- IPv6 / ENA / EFA
- IPv6 / ENA / EFA
- EBS optimized
- Yes
- EBS bandwidth (baseline / max)
- 80000 / 80000 Mbps
- EBS IOPS (baseline / max)
- 260,000 / 260000
- EBS NVMe
- required
- Instance storage
- 8 × 3800 GB NVMe SSD
- GPU
- 8 × NVIDIA H200 (1128.0 GiB total)
- Placement groups
- cluster, partition, spread
- Hibernation / Nitro Enclaves
- No / Yes
- Dedicated hosts
- —
- Boot modes
- legacy-bios, uefi
- Spot supported
- Yes
Public AWS list prices (USD), refreshed nightly (catalog 2026-11-06). RI and Savings Plan rates are effective hourly (upfront amortised over the term) and are charged for every hour of the term. Excludes taxes, data transfer and storage. How these numbers are sourced and checked. Not affiliated with Amazon Web Services.
List prices are where it starts. Your bill is where it matters.
TruCost reads your actual AWS usage and tells you which of these commitments you should really buy, what your idle capacity costs, and where every dollar of the bill goes. Thirty minutes on your own data.