STAC Report: STAC-A3 (backtesting) on NVIDIA DGX-2 using Python

6,250 times as many simulations on a basket of 50 instruments in 1 hour as the previous record

13 May 2019

STAC recently ran STAC-A3 benchmarks using Python on a multi-GPU system from NVIDIA. The STAC Report is available here.

STAC-A3 simulates workloads common in the refinement and backtesting of trading strategies. As a rate-limiting step in a firm's response to changing market conditions, the performance of backtesting infrastructure has a top-line impact. Several trading firms drove the requirements for STAC-A3 in order to facilitate architectural comparisons. Like other STAC Benchmarks, STAC-A3 is agnostic to architecture. One test specification has been implemented in STAC-A3 so far: a per-instrument parameter sweep of a simple mean-reversion algorithm. Other specifications are in development.

The stack under test (SUT) consisted of the NVIDIA-authored “STAC-A3 Pack for Python with RAPIDS (Rev A)” integrated with Dask, numba, and cuDF, running on an NVIDIA DGX-2 server containing 16 x V100 (Volta) GPUs, 2 x Intel Xeon Platinum 8168 processors and 30TB of NVME SSDs for application data.

Highlights from the STAC-A3 SWEEP benchmarks on this SUT included:

• Performed 6,250 times as many simulations on a basket of 50 instruments in 60 minutes as the previous record (STAC-A3.β1.SWEEP.MAX60)
• Running 10,000 simulations on a basket of 48 instruments took less than 6 minutes.
• The time to run 10,000 simulations on a basket of 48 instruments was effectively the same as the time to run 1,000 simulations on the same basket size, as measured in these tests.

At the upcoming STAC Summits, Michel Debiche, Director of Analytics Research at STAC, will discuss these results in the context of previous STAC-A3 projects and the near-term roadmap for the STAC-A3 benchmark suite.

The configuration details, implementation code, and test-harness software are available to firms with premium subscriptions. For information on premium subscriptions, please contact us.