DolphinDB's 10–20x Boost
DolphinDB showcased its Shark CPU–GPU platform at NVIDIA GTC, claiming 10–20x speedups for finance workloads like factor mining and intraday risk monitoring — a big tick for firms merging GPU compute into trading analytics pipelines. The demo also highlighted a brokerage that unified siloed systems for intraday risk, signalling practical deployment paths beyond benchmarking. (x.com)
Shark is built from three named modules—GPLearn (GPU genetic-algorithm factor discovery), DeviceEngine (GPU-accelerated factor computation), and SharkGraph (GPU execution graph for DolphinDB scripts). (docs.dolphindb.com) DolphinDB introduced Shark as part of its 3.0 CPU–GPU heterogeneous compute layer, explicitly positioning it to accelerate Monte Carlo pricing and other compute-heavy finance analytics on top of DolphinDB’s time‑series storage. (docs.dolphindb.com) DolphinDB and NVIDIA’s teams integrated RAPIDS primitives—libcudf for column/table primitives and RMM for pooled device memory—to avoid frequent CUDA alloc/free overheads and reuse GPU memory during massive temporary-result workloads. (blogs.nvidia.cn) DolphinDB’s product materials state GPU factor-mining and factor-calculation acceleration “up to 10x” in selected scenarios, while independent coverage of Shark GPLearn reports task-specific speedups in the broad range of roughly 30–175x against Python GPLearn on some datasets. (dolphindb.com) (cloud.tencent.com) Documentation and tutorials show Shark supports single-node multi‑GPU parallelism and factor discovery at daily, minute and tick frequencies—capabilities targeted at intraday factor workflows and sub‑minute analytics. (docs.dolphindb.com) DolphinDB’s Orca stream examples demonstrate a dual‑stream join pattern for real‑time account‑level P&L and position monitoring, the same streaming and compute constructs cited in case studies where brokerages collapsed siloed feeds into a single intraday risk pipeline. (docs.dolphindb.com) (dolphindb.com) Commercial deployment options include on‑prem distributed DolphinDB clusters with Kubernetes/ACK guides and a cloud trial path via DolphinDB’s listing on the Alibaba Cloud Marketplace (announcement dated June 12, 2025), enabling both colocated low‑latency installs and cloud‑based evaluation. (docs.dolphindb.com) (dolphindb.com)
