Inside KDB-X GPU Acceleration: How GPU Acceleration Speeds Core q Workloads

Capital markets teams are processing larger datasets inside fixed research, risk, and reporting windows. For workloads built around large joins, sorts, aggregations, and matrix calculations, adding more CPU capacity may not provide the required improvement.

In this technical webinar, KX CTO Paul Colgan explains how KDB-X GPU Acceleration uses CPU and GPU processing within the same q environment. You’ll see how teams can selectively move suitable data and operations to the GPU while the rest of the workflow continues to run on the CPU.

The session includes two transaction cost analysis (TCA) demonstrations. The first compares CPU and GPU query performance across 20 million trades and 200 million quotes. The second shows CPU and GPU processing side by side as a simulated market dataset grows throughout the trading day.

What You’ll Learn

• Which q workloads are strong candidates for GPU acceleration
• Why joins, sorts, aggregations, and matrix calculations can benefit from parallel processing
• How the GPU namespace supports GPU-enabled q and qSQL operations
• How selective GPU residency can reduce unnecessary data-transfer overhead
• Where CPU processing may remain more efficient
• How data volume, cardinality, hardware, and transfer patterns affect end-to-end performance
• How KX assesses existing q workloads against measured GPU benchmarks

See CPU and GPU Performance Side by Side

The webinar demonstrates a T+1 TCA workload running on 48 Intel Xeon CPU cores and a single NVIDIA L40 GPU. In the demonstrated configuration, the GPU query completed in 0.684 seconds, compared with 10.79 seconds for the equivalent CPU query.

You’ll also see a streaming TCA example that continuously recalculates execution metrics as the simulated dataset grows. This shows how workload size and GPU residency affect the relative performance of CPU and GPU processing.

Results depend on the operation, data volume, cardinality, data-transfer pattern, and hardware used. The webinar also examines cases where moving work to the GPU may provide little or no end-to-end benefit.

Who Should Watch?

This webinar is relevant to:

• q and kdb+ developers
• Quant developers and research teams
• Execution analytics and TCA teams
• Risk analytics teams
• Data, platform, and infrastructure leaders
• Teams assessing GPU infrastructure for capital markets workloads

Speakers