Quantifi is a financial software company.
We develop trading and risk management solutions
for the financial markets.
One of the biggest challenges that banks
face in the marketplace is the risk
that a counterparty may default. For a mid-sized bank,
calculating the counterparty exposure
can require over 25 trillion trade repricings.
Each trade repricing can in itself be quite complex.
We tested our aggregation model.
The test data we use is a real world data
based on a single counterparty with several thousand trades.
The aggregation model calculates the total exposure
for that counterparty using three different ways.
Running this test, we found that the Intel Xeon Platinum 8180
processor is 1.47 times faster than
the equivalent previous generation
processor with the same memory.
To take advantage of multicore processors,
we use Intel threading building blocks.
That allows us to speed up and optimize
the core parts of our system, and allows
them to run faster on the higher core count of newer processors.
So one of the great things about Intel TPB
is how well it's integrated with Intel's other tools,
including Intel Parallel Studio.
That means you can analyze the performance as you tune it.
We can improve the performance of each trade using
vectorization.
Vectorization allows one instruction
to operate on multiple data at the same time.
Running our code with vectorization enabled
and disabled, we saw that we got an increase of up to 35%
in performance, depending on the core count.
We use [INAUDIBLE] amplifier to help measure
the performance in terms of utilization of cores,
and in terms of vectoization within each thread.
That allowed us to find individual bottlenecks
and steadily improve performance over time.
Our clients have been ecstatic with the results we provided.
Working with Intel, we've been able to provide them
with software that provides more accurate results,
and also reduces their hardware costs.
This has a very big impact on their business.
Không có nhận xét nào:
Đăng nhận xét