Bloomberg is well known company in the financing world and use advanced mathematic equations for calculating the price of everything from company stock to inflation expectations. You need plenty of power to do this and in the past Bloomberg has used a Linux-based processor farm. Last year when it came to the conclusoin that CPU power alone didn’t cut it, it would be too costly, it started looking for a software solution that used the parallel power of GPUs.
The result was very promising and together with NVIDIA’s engineers Bloomberg tailored its software to run on GPUs instead of regular CPus.
Part of the pricing application, data gathering, doesn’t lend itself well to GPU computing, Edwards notes, because it can’t be parallelized. The x86 servers also prepare the problems to be parallelized. But about 90% of the work does run on the GPU platform, he says.
Even if it couldn’t move the entire workload to NVIDIA’s graphics circuits Bloomberg says it increased performance by incredible 800%. This made it possible to replace 1000 servers with 48 servers with two Tesla GPUs each. The faster analysis aside the new farm only consume one third of the power of the old farm.
There is no coincidence that NVIDIA has high expectations for GPGPU and not just focus on games and 3D performance.
Different Tesla solutions from NVIDIA
