← All instances

inf2.48xlarge

⚡ 12x GPU • 384.0 GB

Part of the INF2 Accelerated Computing family — Hardware accelerators (GPUs, FPGAs) for specialized compute.

Name:infAWS Inferentia Acceleratorsinf = AWS Inferentia AcceleratorsPurpose-built AWS Inferentia silicon designed for ultra-low latency, high-throughput machine learning inference.22nd Generation2 = 2nd GenerationAWS 2nd generation hardware platform with updated CPU/system architecture.48xlarge48 Extra Large48xlarge = 48 Extra Large192 vCPUs • 768.0 GiB RAM • EBS only • 100 Gigabit Network • 12x GPU (384.0 GB VRAM)
Price (Linux)
$12.9813/ hr
~$9,476.33/mo
Compute
192vCPUs
96 cores • x86_64
Memory
768.0GiB
4.0 GiB / vCPU
Value
$0.0676/ vCPU-hr
$0.0169/GiB-hr
Network
100Gbps
Baseline: 100 Gbps
Storage & EBS
7,500MB/s
Max: 240,000 IOPS

Technical Specifications & Limits

Hardware specs, container thresholds, and system limits

Compute & Processor

ProcessorAMD EPYC 7R13 Processor
Architecturex86_64 (x86_64)
vCPUs / Threads192 vCPUs
Physical Cores96
Clock Speed2.95 GHz
Intel AVX / AVX2No
Intel Turbo BoostNo
Bare MetalNo (Virtualized)

Performance & Benchmarks

CoreMark ScoreNot benchmarked
CoreMark / vCPU—
FFmpeg Transcoding—
CoreMark per Dollar—
Cost per vCPU-hr$0.0676
Cost per GiB-hr$0.0169

Containers & Kubernetes

EKS Max Pods737 pods
Max ECS Tasks120 tasks
Max ENIs15
IPv4/v6 per ENI50
Total Addressable IPs750 total IPs
CNI Prefix DelegationSupported (High Density)

Networking & Bandwidth

Network Performance100 Gigabit
Baseline Bandwidth100 Gbps
Burst / Peak Bandwidth100 Gbps
Enhanced NetworkingYes
EFA SupportNo
IPv6 SupportNative Dual-Stack Supported

Storage & EBS Optimization

Storage ConfigurationEBS only
Local NVMe DiskNone (EBS only)
EBS Baseline Throughput7500 MB/s
EBS Max Throughput7500 MB/s
EBS Baseline IOPS240,000
EBS Max / Burst IOPS240,000
Dedicated EBS Throughput60 Gbps

GPU & Hardware Acceleration

GPU ModelNVIDIA Accelerator
Accelerator Count12x GPU(s)
Total GPU Memory384.0 GB
Memory ArchitectureGDDR / HBM
CUDA Compute Capability—
Acceleration RuntimesCUDA, TensorRT, cuDNN, PyTorch, PyJAX

Platform & Architecture Details

Hypervisor / SystemAWS Nitro System (PCIe Offload)
Nitro EnclavesSupported (Isolated security enclaves)
Normalization Factor17.12116466
Generation StatusCurrent Generation (Recommended)
Date Introduced2023-04-13

Regional Pricing Matrix

Live on-demand rates across all global AWS cloud regions
Lowest Price Region
$12.9813/ hr
US East (N. Virginia) (us-east-1)
Highest Price Region
$22.0681/ hr
South America (Sao Paulo) (sa-east-1)
Global Fleet Average
$16.9256/ hr
Across all active regions
Regional Coverage
13regions
Available worldwide
Linux — Available in all 13 regions
Region NameRegion CodeOn-Demand Price
Asia Pacific (Mumbai)
ap-south-1$16.8757
Asia Pacific (Singapore)
ap-southeast-1$18.1738
Asia Pacific (Sydney)
ap-southeast-2$16.8757
Asia Pacific (Tokyo)
ap-northeast-1$19.4719
EU (Frankfurt)
eu-central-1$19.4719
EU (Ireland)
eu-west-1$16.2266
EU (London)
eu-west-2$19.4719
EU (Paris)
eu-west-3$18.1738
EU (Stockholm)
eu-north-1$14.2794
South America (Sao Paulo)Highest
sa-east-1$22.0681
US East (N. Virginia)Lowest
us-east-1$12.9813
US East (Ohio)Lowest
us-east-2$12.9813
US West (Oregon)Lowest
us-west-2$12.9813