Case Study | 100x Faster Risk Analytics for a Top Brokerage——Powered by DolphinDB Shark and GPU Acceleration

With the rise of high-frequency trading and increasing market volatility, real-time risk monitoring is becoming mission-critical for brokers and trading institutions. What used to be end-of-day calculations must now run intraday — in seconds, not hours.

To meet this challenge, a leading brokerage firm adopted DolphinDB Shark, a CPU-GPU hybrid computing platform, to dramatically accelerate risk metric computations. By leveraging GPU acceleration as the core computing engine, the firm achieved performance gains of 10x to over 100x in key high-frequency trading scenarios.

The Challenge: Real-Time Risk in a Faster Market

In recent years, the rapid evolution of algorithmic trading and complex derivatives has pushed risk management systems to their limits. Today's trading firms are grappling with:

• Increasingly sophisticated risk models (valuation, sensitivities, counterparty risk, full-market exposure)

• Stricter regulatory demands for transparency and responsiveness

• The need to react instantly to intraday market volatility

However, most legacy infrastructures are struggling to keep pace — prompting forward-looking institutions to rethink how they manage risk in real time. This brokerage firm is one such example.

Breaking Point: Where the Legacy Stack Fell Behind

Despite heavy investment in traditional infrastructure, the firm's existing setup faced three major bottlenecks.

Data Movement Bottlenecks

The architecture's separation of storage and compute created a constant data shuffle that became the system's Achilles heel. Moving massive datasets between storage and processing layers generated significant overhead, with performance degrading precisely when trading activity—and the need for speed—peaked.

Processing Power Barriers

CPU-only workloads couldn’t keep up with high-frequency tasks.While scaling horizontally with additional servers seemed logical, it proved both costly and ineffective. Critical Monte Carlo simulations and stress tests that needed to complete in minutes stretched into hours, making real-time risk assessment unattainable.

Disjointed Technology Architecture

Over time, different teams built isolated tools using a mix of programming languages and frameworks. Without a unified platform to orchestrate both CPU and GPU resources, collaboration suffered and system maintenance became increasingly complex and expensive.

The DolphinDB Solution: Shark Platform

After a thorough evaluation, the firm selected DolphinDB Shark to build a unified architecture that seamlessly integrates data ingestion, storage, memory management, GPU acceleration, and compute services.

By pairing DolphinDB’s distributed file system with CUDA-powered GPU processing, they reengineered critical risk workflows. Tasks that once required hours of CPU time were now completed in seconds or even milliseconds — making real-time risk monitoring a reality.

• Multi-modal storage engine Includes built-in TSDB, OLAP, and PKEY engines to efficiently manage diverse financial datasets — from high-frequency market data, reference data, and risk factors.

• Seamless CPU-GPU compute pipeline DolphinDB maximizes CPU performance with native vectorized computing. Shark extends this by enabling GPU acceleration through simple @gpu annotations, allowing compute-intensive tasks to run in parallel on GPUs.

• Multi-paradigm programming Supports SQL, imperative, functional, and vectorized programming styles. Developers can design hybrid workflows that leverage both CPU and GPU resources for batch processing, real-time streaming, and complex simulations.

• Unified stream-batch processing Equipped with over 10 built-in streaming engines, it supports real-time and batch workloads using a single codebase — improving consistency and accelerating development.

The Gains: From Hours to Seconds

The results speak for themselves. After moving to DolphinDB Shark, the firm saw massive improvements across several core workloads:

Use Case	Legacy System	DolphinDB Shark	Speedup
Snowball option pricing	830–2860ms	≤ 40ms	20x–100x
Multi-counterparty, multi-contract	> 1h	≤ 40s	> 100x
Multidimensional Monte Carlo simulation	4.2s	20ms	> 100x

These breakthroughs unlocked true real-time intraday risk analysis, empowering the firm to:

• Compute complex risk metrics across all counterparties in real time

• Replace fragmented, siloed systems with a unified compute platform

• Significantly shorten the development cycle for new risk models and indicators

The Shark platform is currently compatible with NVIDIA GPUs.

Discover what DolphinDB Shark can do for real-time risk management. Reach out at info@dolphindb.com.