Key features of IBM Bayesian Optimization Accelerator

Task parallelization

Make the most of infrastructure by parallelizing efforts and spending less CPU and wall clock time getting to the ultimate answer.

Scale to orders of magnitude more dimensions

Unlike open source Bayesian libraries, you get the ability to scale to orders of magnitude more dimensions, allowing you to tackle real world problems instead of academic ones.

Determine design points with fewer samples

Use batch sampling

Unlike search methods such as Grid or Random search, you are able to determine design points with much fewer samples required, which gets to results faster and cheaper.

Ensure traceability

Ensure traceability to your models to build trust in your model design methodology through interrogation of the optimizer.

Easy to add to existing clusters

Add easily and quickly to any existing cluster without needing any Bayesian priors.

Improved throughput

Get more from your existing investment by improving the throughput of your existing infrastructure.

Technical details

Software requirements

  • RHEL 7.7, CUDA 10.2, ESSL 6.2

Hardware requirements

  • AC922 with Minimum Configuration (included)
  • 256 GB Memory
  • Dual CPU, 16 cores per CPU
  • (Qty. 2) Nvidia V100 GPUs
  • (Qty. 2) 1.6 TB NVMe
  • 2x 1 Gb Ethernet ports

Gain agility and flexibility

IBM flexible payment plans help align infrastructure investments with workload needs.