Numerix, Intel Publish New Threading Performance Benchmarks for CrossAsset

The study was conducted around CrossAsset's use of Intel's Threading Building Blocks (Intel TBB) Library, which users can deploy to easily write parallel programs taking full advantage of multicore performance, that are portable and composable, and that have future-proof scalability.

In leveraging the Intel's Xeon processor platform, Intel and Numerix worked together to optimize the performance of CrossAsset Analytics on an Intel Xeon processor based platform by vectorizing the core computational library.

As a result, this improved performance of CrossAsset Analytics by 1.78x by using Intel TBB memory allocator.

“The computational requirements imposed by the market and regulators are daunting. Even Smartphones don’t come in single-core CPU form any more, and every efficient computation engine needs to be able to take advantage of multiple cores,” says Ken Roller, global director of strategic relations at Intel. “The question has now become ‘how’, because in general there is no free lunch.”

Adds Steve Karmesin, SVP for Numerix CrossAsset infrastructure development: “Usually you can scale up in separate ‘processes’ on a single machine to keep the computations separate – the advantage being that they don’t interfere with each other, thus they scale up better. Or, you can scale up with multiple threads and get efficiencies by sharing data, but the computations can interfere with each other in multiple subtle ways, slowing down the computation. In leveraging the Intel TBB thread optimized memory allocation system we saw thread scaling much closer to the ‘process’ scaling, but with threads so we can also capture shared memory benefits.”