The following are the minimum requirements to run Pickle Piper:

• Docker running on a host machine with an NVIDIA GPU
• Kafka Broker for Source and Destination Topics
• Docker Container Specifications:
  • 1024MB Memory
  • 2v CPU
• Standard Docker Image uses NVIDIA Container runtime that interfaces with underlying GPU on host.

Pickle Piper uses CUDA libraries that interface directly with NVIDIA GPU’s to preform inferencing, hence a NVIDIA GPU is a hard requirement for running Pickle Piper.

Supported NVIDIA GPU’s;

• Volta: Tesla V100, Tesla T4, and Titan V feature Volta architecture.
• Turing: GeForce RTX 20 series (RTX 2080 Ti, RTX 2080, RTX 2070), Quadro RTX series, and Tesla T4
• Ampere: GeForce RTX 30 series (RTX 3090, RTX 3080, RTX 3070), Quadro RTX series, and A100 GPUs.

Cloud Service Providers NVIDIA GPU Enabled Instances

Google Cloud Platform

a2-highgpu-1g or above (A100 GPU) n1-highmem-2 or above (P100, K80, T4)

AWS

p4d.3xlarge or above (EC2 P4 Instance Types) p3.2xlarge or above (EC2 P3 Instance Types) g4dn.xlarge or above (EC2 G4 Instance Types)

Azure

Standard_NV24 or above (A100 Tensor Core) Standard_NC12s_v3 or above (V100 Tensor Core) Standard_NV6 or above (T4 or K80 Tensor Core)* *AMD Radeon Instinct MI25 instances are not supported