>_ imrc@compute-fabric-01: ~/signal-fabric
File Edit View Search Terminal Help
IMRC DISTRIBUTED COMPUTE FABRIC LIVE
imrc@compute-fabric-01:~/signal-fabric$ spark-submit --master cluster --deploy-mode client signal_fabric.jar
>_ solver@hpc-node-07: ~/numerics
File Edit View Search Terminal Help
HIGH-DIMENSIONAL NUMERICAL ENGINE SOLVING
solver@hpc-node-07:~/numerics$ mpirun -np 96 ./adaptive_solver --method newton-krylov --tol 1e-9
>_ feed@market-gw-02: ~/stream-engine
File Edit View Search Terminal Help
REAL-TIME MARKET INTELLIGENCE STREAMING
feed@market-gw-02:~/stream-engine$ ./market_stream --mode=replay --venues consolidated --window 100ms
>_ mlops@model-core-01: ~/production
File Edit View Search Terminal Help
AI MODEL OPERATIONS CORE DEPLOYING
mlops@model-core-01:~/production$ ./mlopsctl deploy --registry production --strategy canary

We are a machine learning research firm working at the intersection of high-performance computing and quantitative finance. We develop end-to-end models that ingest vast amounts of data each day, delivering robust, production-grade insights and truly accomplishing machine learning at scale. Our technology stack includes Spark clusters with thousands of cores and high-performance computing hardware that solve very large machine learning and nonlinear optimization problems in mere seconds.

About

Projects

MetalBoost‍ ‍(Open Source)

An implementation of Gradient Boosted Trees specifically for Apple Silicon.

Machine Learning Based Stock Price Valuation and Prediction (Proprietary)

Using massive parallel computing to analyze petabytes of stock-related data to predict future stock prices with high accuracy.

Models run across custom built HPC clusters that constantly ingest new data and update predictions.

The screen capture below is a real live screen recording from our predicting the world state custom ML solution that ingests PBs of data from multiple markets across the world and simulates prices across these markets and sectors.

(Please note that IMRC does not currently accept funding from the public / outside investors.)

MetalBoost: A Custom Gradient Boosting Library specifically for Apple Silicon

HPC Spark-based Kolmogorov/Fokker-Planck PDE Solvers (Proprietary)

We use Databricks and PySpark to distribute large-scale numerical PDE solving across parallel compute clusters.
In one benchmark, 10,000 forward Kolmogorov equations evolved 6.4 billion grid-time states in just 5.5 seconds.

Machine Learning At Scale

Some of Our Current Infrastructure

CPU Platform Max Cores Max Memory Typical Role
AMD EPYC Zen 5 192 768 GB Spark executors, ML training, HPC
AMD EPYC Zen 4 192 768 GB Distributed compute
Intel Xeon 6 192 3 TB SQL analytics, lakehouse, large JVM workloads
Intel Xeon Ice Lake 64 1 TB NVMe accelerated processing
AWS Graviton4 192 1.5 TB ARM compute cluster
AWS Graviton3E 64 128 GB HPC simulation
Apple Mac Mini Pro M4 5 x 12 = 60 24 GB x 5 = 120 GB ML Prototyping
GPU Platform Memory Architecture Primary Workloads
NVIDIA H100 80 GB Tensor Core Foundation models
NVIDIA A100 40 / 80 GB Tensor Core Deep learning
NVIDIA L40S 48 GB RTX Inference

Where We Are Located

Toronto Office

100 University Avenue, 5th Floor

Toronto, Ontario, M5J 1V6

 

New York Office

135 Madison Ave, 5th Floor

New York, NY 10016