Today NVIDIA announced the release of their fastest ever accelerator for scientific computing – the Tesla K80. It has more raw compute power than any other GPU card available on the market, featuring two of the new GK210 GPUs. But what performance improvements can we expect for typical financial applications? Xcelerit have been given early access to the card and put it to the test, comparing it to their previous flagship, the Tesla K40.
Last month saw a gathering of who’s who in Fixed income circles at the Hotel Condes de Barcelona for the 10th Fixed Income Conference. Delegates were able to avail of a pre-conference workshop on Wednesday followed by two days of events where experts in every aspect of quantitative finance were represented.
A Guide to xVA Algorithmic Optimisations
I have seen many xVA acceleration initiatives in Banks focusing exclusively on software acceleration using GPUs or many-core hardware. This is omitting an essential part of the problem which can have a massive impact on overall application performance: algorithmic optimisation. In general, this is not only true in computational finance but also across many other fields in science and engineering.
In this post, let’s zoom in on the hot topic of xVA (various risk value adjustments), and let’s see how 10x to 100x performance gains can be achieved through algorithmic optimisations – without any hardware or infrastructure changes.
In our previous blog post, we have seen that Haswell (Xeon E5 v3 series) achieves a significant performance boost compared to the previous generation Ivy Bridge processor (Xeon E5 v2 series). But let’s see how its performance compares to Intel’s flagship accelerator processor, the Xeon Phi, for a popular application in computational finance.
Intel Xeon Phi 7120P
Intel Xeon E5 v3 (aka Haswell)
Intel just released its new Haswell server processor line (Xeon E5 v3 series), promising significant performance gains over previous-generation Ivy-Bridge processors. In this post, we will compare the two generations for a popular financial application – a Monte-Carlo LIBOR Swaption Porfolio pricer.
Haswell Processor Die
Today Xcelerit announced the release of version 2.6 of its SDK, increasing performance significantly and further improving the user experience. A complete list of the updates and new features can be found in the Release Notes shipped with the package.
We’ve just published a new white paper which looks at how to boost Reverse Time Migration algorithms while maintaining flexibility and portability. This white paper features an implementation that achieves up to 9,000 MCells/s on a single K40 GPU (8th order in space, 2nd order in time, 3D isotropic).
This white paper shows how to develop high performance Reverse Time Migration (RTM) implementations that are both flexible and easy to maintain, by using the Xcelerit platform. As energy exploration is pushed towards more complex geologies, RTM has become the de-facto standard algorithm to construct images of the Earth’s subsurface from measured seismic wave data. High performance hardware such as GPUs allow to cope with the tremendous computational complexities involved – at the cost of increased software development and maintenance effort.
Xcelerit were represented at last month’s Global Derivatives conference in Amsterdam’s Okura Hotel. This is the largest annual gathering in Quantitative Finance
We found the latest advances for reducing the complexity of xVA calculations particularly exciting. For example, Jesper Andreasen (Dankse Bank) presented a number of smart algorithmic optimisations for xVA, Adil Reghai (Natixis) showed how to use algorithmic differentiation cleverly to reduce CVA complexity, and Andrew Green (Lloyds Banking Group) presented a highly efficient method to compute FVA of VAR-based inital margin. Additionally, for the first time this year, the conference featured a stream about “Innovations in Computational & Numerical Efficiency,” with a number of implementation-centric talks on how to cope with the computational complexities of financial analytics.
The audience of more than 500 quants were revived at break time by Xcelerit’s crack team of baristas, who sent them back to the following sessions freshly caffeinated and ready to absorb more.
On May 28th, Xcelerit in association with the Wilmott Forum delivered a seminar on “xVA and Risk-Aware OTC Pricing: Getting Ready for the New Normal” to an audience of quantitative analysts and other financial industry specialists in London.
The risk trainer Justin Clarke started by explaining the forces driving the drastic changes facing banks’ risk departments and trading desks. He illustrated the complexities involved in computing the multitude of adjustments to OTC prices required for compensating the associated risks (xVA). His talk was followed by a session with Xcelerit’s CEO, Hicham Lahlou, who demonstrated how the calculations needed can be achieved within the turnaround times required, using Xcelerit’s tools. Xcelerit’s Jorg Lotze, concluded the seminar with a live demonstration, showing the audience the achievable performance on GPUs without requiring major modifications to the existing codebase.
The event was over-subscribed and even though we maintained a waiting list, many people were still left disappointed. We are happy to announce that a video of the event is now being made available on our website. If you missed it, click here to gain access to the material.
We’ve just published a new white paper which covers how to cope with the computational complexity involved in calculating various valuation adjustments, such as Credit Valuation Adjustment (CVA), Debit Valuation Adjustment (DVA), and Funding Valuation Adjustment (FVA).
These adjustments are commonly referred to as xVA. This white paper gives an overview of the different xVA adjustments, shows
how they are typically computed, and outlines where the computational complexities lie. We give recommendations on how to achieve high performance, portability, and scalability for centralised in-house xVA implementations. We show how, by careful software design, we can easily harness, not only the power of multi-core CPUs, but also accelerator co-processors such as graphic processing units (GPUs) and the Intel Xeon Phi.
You can download the paper here.