Deployment

Simple Machines’ Mozart Platform is available now as a cloud service, or on-premise via a plug-and-play upgrade to existing infrastructure.

Availability and Evaluation Paths

Available Now

Cloud service trials
  • SMI cloud stack
  • Running on fast emulator on SMI cloud
  • Customer managed with SimpleMachine field engineering support
Early access evaluation kit
  • FPGA based system
  • Full SMI software development kit
Performance Benchmarking
  • Customer provided models and data sets
  • SMI runs performance metrics

Available July/August

Mozart Cloud Service
  • Quantization and Inference service
  • Docker, VM, bare-metal access
PCIe Card + OEM System Evaluation Kit
  • PCIe card with system integration
  • Standalone PCIe card also available

CLOUD SERVICE

SMI Cloud Service

  • Docker container with SMI software development kit and Tensorflow
  • Library of pre-trained models: Resnet50, Mask-RCNN, SSD-Resnet34, BERT-Base, BERT-Squad, GNMT, DeepSpeech, DeepSpeech2
  • Deployed on Equinix Cloud

Networking and Storage

  • Native backedup storage
  • Leading Colo Provider provides fast access to AWS, GCP, Azure, Oracle Cloud
  • Virtual Private Cloud for customer private network

Kubernetes cluster management

Inference Service

SMI Quantizer

  • Available as RESTful API
  • Customer provides FP32 pre-trained model, calibration/tuning dataset

SMI Inference Service

  • Models reside in SMI cloud
  • Inference API calls

SMI Analytics Service

  • Offline performance analysis
  • Upload full dataset / large number of batches
  • Run in batch made to get performance and get TCO analysis

Training ServicE

Quantized SMI-Native training

SMI Training

  • Uploaded training data, command line access to S3/Google cloud
  • Use SMI hardware for training

SMI Analytics Service

  • Offline performance analysis
  • Upload full dataset / large number of batches
  • Run in batch made to get performance and get TCO analysis

On-Premise Solutions

PCIe Cards in OEM System

System Specification

  • Available with tier-1 System OEMs
  • 1U and 4U Intel Xeon-based systems
  • Low Power and Multiple Form Factors
  • Specifications
  • S300
  • S100
  • System Interface
  • PCIe Gen 3 x16
  • Memory
  • 24 GB
  • 8 GB
  • Power (Max.)
  • 220W
  • 75W
  • Form Factor
  • Full Height
    & Length
  • Half Height
    & Length

Framework and OS Support

Deep Learning

tensoflow
onnx
pytorch

Machine Learning & Accelerated Applications

xgboost
spark
gatk

Operating Systems

ubuntu
suse
fedora
opensuse