🔍 MLPerf Configuration Finder (ongoing preliminary work)

Find the optimal configurations for your AI workloads by specifying your model and constraints. Results are ranked by performance and include both real benchmark data and AI-generated predictions.

All configurations include a ±10% tolerance for continuous features like model size, memory capacity, etc.

Ready to search. Enter your criteria and click 'Search Configurations'.
Architecture

Model architecture type
Weight Data Type

Precision format for model weights
6 671
Vendor

Hardware manufacturer
Model

Specific accelerator model
24 1024
24 1024
Interconnect

GPU-to-GPU connection type
1 72
1 72
CPU Vendor

CPU manufacturer
CPU Model

Specific CPU model
Number of Nodes

Number of physical servers in the system
64 8192
64 8192
Operating System

Host operating system
cuda

Select cuda framework version
jetpack

Select jetpack framework version
pytorch

Select pytorch framework version
rocm

Select rocm framework version
tensorrt

Select tensorrt framework version
vllm

Select vllm framework version

When enabled, AI will predict performance for configurations not in the benchmark database
Optimization Target

Choose whether to optimize for highest performance or lowest cost per token

Enter your requirements and click 'Search Configurations' to find suitable hardware.

Configuration Details

Configuration Details

Top Configurations Comparison
1 100
Authors: Daniel Altunay and Grigori Fursin (FCS Labs)