Accelerate Your Analytics with GPU Before The Window Closes

Run TCA, backtesting, and risk simulations up to 27× faster on NVIDIA GPUs. Keep your q codebase and move only the operations that benefit from GPU compute.

Schedule a workload assessment Explore Workloads

What would you do with the time you’re currently losing to compute?

Quants & Algo Developers

Backtesting and Model Calibration

Run backtesting sweeps faster so teams can recalibrate models more often.

Full-Resolution Research

Test strategies across full tick histories and larger parameter sweeps without downsampling.

Monte Carlo Risk and Covariance Calculations

Run matrix-heavy Monte Carlo and covariance estimation in q, before committing capital.

Execution Traders & Desk Coverage

Intraday Execution Analysis

Track VWAP cost and spread capture while orders are active so traders can adjust routing sooner.

Same-Day Child-Order Analysis

Review child orders across venues on the trade date instead of relying on sampled T+1 analysis.

Interactive Venue and Liquidity Analysis

Query multiple months of tick data for ad hoc venue and liquidity analysis with lower latency.

Inside GPU-accelerated query execution in KDB-X

Read the technical whitepaper for a detailed look at the KDB-X .gpu module, its CUDA integration, and how it runs q workloads on NVIDIA GPUs.

  • The .gpu architecture and q API
  • GPU computing and execution inside KDB-X
  • NYSE TAQ benchmarks on A100, H100, and H200 GPUs
  • Where GPU execution helps and where the CPU remains faster

Get the KDB-X GPU technical whitepaper

Explore the architecture, .gpu API, CUDA integration, and performance results across A100, H100, and H200 GPUs.

「*」は必須フィールドを示します

このフィールドは入力チェック用です。変更しないでください。

By submitting this form, you will also receive sales and/or marketing communications on KX products, services, news and events. You can unsubscribe from receiving communications by visiting our Privacy Policy. You can find further information on how we collect and use your personal data in our Privacy Policy.

Protected by reCAPTCHA. Google Privacy Policy & Terms of Service apply.












Expand Your Possibilities Without Rebuilding

Service More Customers

Double the number of weekly client deep-dives to support commission growth and retention.

Infrastructure Consolidation

Replace 8–12 CPU servers with a single GPU node to reduce power, cooling, footprint, and cost per query.

Zero Code-Rewrite Risk

Use standard q syntax through native .gpu primitives, without CUDA development, retraining, or loss of existing investment.

Use Cases

KDB-X Accelerates The Most Complex, Time-Consuming Analytics

KDB-X runs CPU and GPU processing in the same environment. Teams can move compute-intensive joins, sorts, aggregations, and matrix calculations to NVIDIA GPUs while other work remains on the CPU.

CPU

Everything Else

Stays exactly where it is.

GPU

VWAP Curve-Tilt Backtesting

9.2× faster

Select a category, then choose a use case to see its before/after CPU-to-GPU comparison.

Execution and TCA VWAP Curve-Tilt Backtesting

36,040 ms

Before · CPU

3,900 ms

After · GPU

9.2×

faster

Quants Recalibrate Execution Schedules Weekly or Daily.

Backtesting sweeps that validate price-prediction signals rerun VWAP across many volume-curve tilts on 60K orders. With KDB-X GPU Acceleration, the sweep runs 9.2× faster end to end on live desk data, allowing schedules to be recalibrated weekly or daily instead of quarterly.

  • 9.2× end-to-end speedup
  • 8–12× compute-kernel speedup
  • Recalibrate weekly or daily instead of quarterly
  • Same q, no CUDA

GPU Workload Assessment

Identify which q workloads are strong candidates for GPU acceleration.

KX analyzes the joins, sorts, grouped calculations, and data-movement patterns in your q code, maps them against measured benchmark patterns, and identifies where GPU acceleration is most likely to help. A KX engineer reviews the findings with you and recommends what to test first.

Estimates are directional. Measured results depend on data volumes, hardware, memory movement, and the proportion of GPU-supported operations.

Schedule a workload assessment

01

Candidate workloads & bottlenecks

Identifies joins, sorts, grouped calculations, and data-movement patterns most likely to benefit from GPU execution.

02

Directional acceleration ranges

Maps findings against measured benchmark patterns to estimate where GPU acceleration is most likely to help.

03

Executive summary

Delivers a report scoped to your codebase — what to migrate, in what order, with ROI estimates attached.

Once You Turn On GPU

How GPU Acceleration Changes Analytics Workflows

Six shifts your quants, traders, and platform owners feel from the first GPU-offloaded job through to the post-trade review.

01

Broader Backtesting and Research

Run broader universes, longer histories, and larger parameter sweeps without reducing data resolution.

04

More Throughput per Node

Process the same data volumes with fewer nodes when the workload is a strong fit for GPU execution.

02

Faster Execution Analysis

Return execution insight sooner, leaving more time to act within the relevant trading window.

05

End EOD Overruns

Reduce the runtime of the large joins, sorts, and aggregations that determine whether reports finish during the overnight window.

03

q-Native GPU Operations

Use the .gpu namespace to run compute-intensive joins, sorts, and aggregations on NVIDIA GPUs from q.

06

Larger Histories and Parameter Sweeps

Recalculate rolling windows, test more parameters, and validate models against higher-resolution histories within the existing research cycle.

Related Resources

By The Numbers

Performance Results From Four KDB-X GPU Deployments

These results come from production deployments using existing q code and schemas, with selected operations moved to the GPU.

Monte Carlo Risk & Tail-Risk Hedging

Global Hedge Fund

27.3×

CPU

GPU

Custom simulations across 25,000 securities for tail-risk hedging, replacing external CUDA platforms.

Execution Analytics Modernization

Tier-1 Global Investment Bank

30×

CPU

GPU

Replaces a legacy execution analytics stack entirely – OneTick migration plus GPU boost.

Real-Time TCA & Desk Cannibalization

Global Prop Trading Firm

35×

CPU

GPU

Query latency from 5s to 600ms – prevents adverse venue selection during live trading.

Child Order Spread Capture

Tier-1 Global Investment Bank

11.1×

CPU

GPU

100% order coverage, no statistical sampling – real-time liquidity capture on live flow.

Deployment and Setup

Add GPU Acceleration to an Existing KDB-X Deployment

KDB-X GPU Acceleration runs as a q-native module in supported on-premises, cloud, and containerized deployments.

Use Native .gpu Operations

Call GPU operations from q through the .gpu namespace.

Keep Development in q

Move selected joins, sorts, aggregations, and other supported operations to NVIDIA GPUs without low-level CUDA development.

Scope the Change With a Workload Assessment

The GPU Workload Assessment identifies which parts of the current deployment are suitable for benchmarking and what needs to change.

KDB-X GPU Acceleration FAQs

What Existing Customers Ask First

KDB-X GPU Acceleration is a q-native module designed to work with existing KDB-X deployments. A workload assessment confirms the supported deployment model, hardware, and any required configuration changes.

No. Developers call supported .gpu operations from q. The module handles CUDA execution beneath the q interface.

KDB-X campaign and technical materials cover NVIDIA A100, H100, H200, L40, and B200 GPUs. The supported models depend on the KDB-X release and deployment.

Large joins, sorts, windowed aggregations, matrix calculations, and other parallel workloads are the strongest candidates. Smaller workloads, operations that depend on CPU index shortcuts, and queries with high data-transfer overhead may remain faster on the CPU.

No. Teams can move selected operations and required columns to the GPU while the rest of the workload continues to run on the CPU.

Workload assessment and execution

How KDB-X Routes q Workloads to the GPU

KDB-X GPU Acceleration runs selected q operations on NVIDIA GPUs and controls which data moves between CPU and GPU memory.

Selective GPU residency
Control which data moves to the GPU and when. This limits PCIe transfers to the data required for each computation.

Accelerate the ops that matter
Use q-native operations for time-series joins, sorting, filtering, aggregation, matrix computation, and controlled data movement between CPU and GPU memory.

.gpu.gpu.aj.gpu.select / .gpu.xasc.gpu.gemm.gpu.to / .gpu.xto.gpu.fromcuFileuse

Move required columns to the GPU
Move the columns required by an operation, such as join keys and sort keys. Keep other payload columns on the CPU.

GPU workload assessment

Assess Your q Workloads for GPU Acceleration

KX reviews joins, sorts, grouped calculations, and data-movement patterns in your existing q code to identify suitable candidates for GPU execution.

  • Workloads that match current GPU performance patterns
  • Joins, sorts, and aggregations that dominate runtime
  • Indicative performance ranges based on measured benchmarks
  • Recommended operations to benchmark first, with estimated ROI

Schedule a workload assessment

// social // social