Hardware reference

GPU & Hardware to AWS Spot Instance Map

Quickly match GPUs and hardware to AWS EC2 Spot instance types. Ideal for sales calls or technical discussions when someone needs something specific—e.g. "I need H100" or "What Spot instances have A100?"

Perpetual Compute Supported (5 instance types)

These instance types are supported for Perpetual Compute and are available for deploy in the portal where listed. Pricing links go to a dedicated page when available, or otherwise to the general pricing page.

GPU → AWS Spot Instance Mapping

15 instance types shown

P5 (H100)

p5.48xlargeSupported
Pricing →
192 vCPUs1920 GB

8× NVIDIA H100 (80GB HBM3)

LLM training · Large-scale inference · HPC · Scientific computing

P4d (A100)

p4d.24xlarge
96 vCPUs1152 GB

8× NVIDIA A100 (40GB SXM4)

ML training · Deep learning · LLM inference · Distributed training

P4de (A100 80GB)

p4de.24xlarge
96 vCPUs1152 GB

8× NVIDIA A100 (80GB)

Large model training · Memory-intensive ML

P3 (V100)

p3.16xlarge
64 vCPUs488 GB

8× NVIDIA V100 (16GB)

Legacy ML training · Inference workloads

p3.8xlarge
32 vCPUs244 GB

4× NVIDIA V100 (16GB)

Medium-scale training · Inference

p3.2xlarge
8 vCPUs61 GB

1× NVIDIA V100 (16GB)

Small-scale training · Development

G6 (L4)

g6.48xlargeSupported
Pricing →
192 vCPUs768 GB

8× NVIDIA L4 (24GB)

Multi-GPU inference · Batch inference · Light training

g6.2xlargeSupported
Pricing →
8 vCPUs32 GB

1× NVIDIA L4 (24GB)

Inference · Fine-tuning · Development

g6.xlargeSupported
Pricing →
4 vCPUs16 GB

1× NVIDIA L4 (24GB)

Inference · Prototyping · Small models

G6e (L40S)

g6e.xlarge
4 vCPUs16 GB

1× NVIDIA L40S (48GB)

Inference · Rendering · GenAI

G5 (A10G)

g5.12xlarge
48 vCPUs192 GB

4× NVIDIA A10G (24GB)

Multi-GPU inference · Rendering

g5.xlarge
4 vCPUs16 GB

1× NVIDIA A10G (24GB)

Inference · Rendering · Virtual workstations

G4dn (T4)

g4dn.12xlarge
48 vCPUs192 GB

4× NVIDIA T4 (16GB)

Inference · Batch processing

g4dn.xlarge
4 vCPUs16 GB

1× NVIDIA T4 (16GB)

Inference · Development · Cost-effective GPU

T3 (CPU)

t3.mediumSupported
Pricing →
2 vCPUs4 GB

No GPU — 2 vCPU, 4 GB RAM

Dev servers · Light workloads · CI/CD

Full reference table

Instance TypeHardware / GPUvCPUsMemoryUse CasesPerpetual Compute
p5.48xlarge8× NVIDIA H100 (80GB HBM3)1921920 GBLLM training, Large-scale inference, HPC, Scientific computingSupported
p4d.24xlarge8× NVIDIA A100 (40GB SXM4)961152 GBML training, Deep learning, LLM inference, Distributed training
p4de.24xlarge8× NVIDIA A100 (80GB)961152 GBLarge model training, Memory-intensive ML
p3.16xlarge8× NVIDIA V100 (16GB)64488 GBLegacy ML training, Inference workloads
p3.8xlarge4× NVIDIA V100 (16GB)32244 GBMedium-scale training, Inference
p3.2xlarge1× NVIDIA V100 (16GB)861 GBSmall-scale training, Development
g6.48xlarge8× NVIDIA L4 (24GB)192768 GBMulti-GPU inference, Batch inference, Light trainingSupported
g6.2xlarge1× NVIDIA L4 (24GB)832 GBInference, Fine-tuning, DevelopmentSupported
g6.xlarge1× NVIDIA L4 (24GB)416 GBInference, Prototyping, Small modelsSupported
g6e.xlarge1× NVIDIA L40S (48GB)416 GBInference, Rendering, GenAI
g5.12xlarge4× NVIDIA A10G (24GB)48192 GBMulti-GPU inference, Rendering
g5.xlarge1× NVIDIA A10G (24GB)416 GBInference, Rendering, Virtual workstations
g4dn.12xlarge4× NVIDIA T4 (16GB)48192 GBInference, Batch processing
g4dn.xlarge1× NVIDIA T4 (16GB)416 GBInference, Development, Cost-effective GPU
t3.mediumNo GPU — 2 vCPU, 4 GB RAM24 GBDev servers, Light workloads, CI/CDSupported

Need a different instance type? Contact us to discuss adding support for your workload.

View live pricing