Forbes writer Tom Groenfeldt delved into the world of GPU computing in his latest blog on financial technology. In talking with GPU technology consultant Andrew Sheppard, Groenfeldt noted that in the financial space, the use of GPU computing is becoming a competitive differentiator. With large amounts of data to be crunched and a compelling monetary incentive to do so, more and more financial firms are latching onto the GPU as a computational accelerator.
According to Sheppard, one of the big enablers of the technology has been more mainstream development tools for GPU programming. In particular, the collaboration between NVIDIA and Microsoft has made it easier for developers to gain access to the technology. In fact, according to David Rich, director of marketing at Microsoft, over 80 percent of the CUDA download are for Windows.
Although programming in CUDA has its challenges, wrapping it in a comfortable programming environment makes a difference. In this case that means NVIDIA’s Parallel Nsight plugin for Visual Studio, which is just the kind of platform necessary to draw in the non-HPC types. In addition, there are now more than 400 universities teaching CUDA to would be GPU computing developers.
It’s not just about making GPU computing more accessible to the developers, though. From Sheppard’s perspective, a new technology has to be 10 times cheaper or 10 times better to get some traction. While speedups from GPU acceleration do vary across codes, in the financial space, there have been some impressive results. Groenfeldt writes:
One risk management vendor reported that GPU processors had a core application speedup of 100 times. Competitors will be compelled to move to GPUs to keep up, said Rich. Bloomberg used 48 GPUs for a bond pricing application and reduced costs from $4 million to $144,000 and power costs from $1.2 million to $31,000.
Those kind of results are certainly encouraging to the folks at NVIDIA and help justify the effort the GPU maker has put into the technology over the last several years. The company has been evolving its GPU platform aggressively towards a more general-purpose computing platform, adding such capabilities as IEEE floating point support, ECC memory, and other programmer-friendly features.
The investment seems to be paying off. Overall NVIDIA’s GPU computing business has gone from zero three years ago to over $100 million in revenue this year.