High-Performance Computing
Our team designs, deploys and optimizes High-Performance Computing (HPC) for back-testing and trading. All else equal, the trader with more computing power wins — they can explore the vast space of parameterized strategies more thoroughly. Three cases where HPC made the difference:
Parameter optimization
Traders need a stable, optimal parameter plateau for a robust, non-sensitive strategy. In high-dimensional parameter spaces the curse of dimensionality makes exhaustive search impossible; HPC architecture and algorithms dramatically speed up the search and widen the explored area. Our low-cost HPC infrastructure is available to quantitative traders.
- ›Confidential: upload only an (encrypted) executable, never source code
- ›No computation, no payment
- ›Lease rather than buy — no hardware or maintenance cost
- ›CPU and GPU supported
- ›Suites of optimization algorithms to suit your needs
Implied volatility pre-calculation
For high-frequency options trading, we pre-compute every possible implied volatility (IV) for the next trading day before the open. In live trading, the IV is read from memory instead of computed on the fly. Black-Scholes (European), Barone-Adesi-Whaley and binomial-tree (American) models are supported out of the box; proprietary pricing models can be accelerated while staying private.
- ›cu1906C44000: (6083 − 1313) / 1 = 4,770 option prices × (50080 − 45310) / 10 = 477 future prices ≈ 2.28 million IVs for one pair
- ›All cu options in [C44000, C54000] and [P44000, P54000]: 35,257,455 IVs
- ›At 5 s per binomial-tree IV, ~108 hours — far beyond a 6-hour trading session
- ›Our HPC infrastructure and algorithms speed this up 1,000×; pre-cached IVs translate into higher expected profit
Chinese stock Level 2 data processing
Level 2 data — ticks, transactions, orders, sizes and order queues — amounts to about 10 GB a day. Back-testing and factor generation over, say, five years, often in repeated passes, is extremely demanding. Our HPC L2 solution makes it practical.
- ›Efficient data-reading and computing APIs
- ›Massive compute for otherwise impossible jobs
- ›High performance at low cost
- ›High-density data compression
- ›Robustness through multiple instances
- ›Automatic fault checking and reporting
- ›APIs for Python, Java, C++ and more
