EM Photonics, today, announced the general availability of its CULA accelerated computing math library. This release includes NVIDIA GPU-accelerated versions of many commonly used LAPACK linear algebra routines. I had a chance to sit down with EM Photonics CEO Eric Kelmelis here at the NVIDIA GPU Technology Conference.
Applications ranging from video games, to medical imaging, to scientific computing have come to depend on the superior processing capabilities of GPUs. By every measure, this trend is rapidly growing and impacting more and more markets,” said Eric Kelmelis, CEO of EM Photonics. “To bridge the current gap between what GPUs can offer and how they can be used to accelerate applications, we have developed CULA in close association with NVIDIA. A broad range of users took advantage of our beta release over the last few months and achieved 5-10x performance gains over CPU implementations.”
You may not hear the words “EM Photonics” or “CULA” often, but you will soon. They are doing for LAPACK what NVIDIA has already done for BLAS. NVIDIA’s new Fermi hardware is indeed impressive. Billions of transistors are begging to be raked over the proverbial coals. However, without a reasonable programming interface available to port existing codes to the new platform, one can argue the cart before the horse.
EM Photonics has been running silent and deep for the last several years in the various GPU efforts. They’ve landed various DARPA and DoD grants/contracts. Most recently, NASA Ames had them work the GPU magic on several production routines.
The CULA linear algebra library enables developers for a wide range of technical computing applications including computational fluid dynamics, electronic design automation, finite element analysis, and electromagnetic simulations, to take advantage of the performance boost of the GPU”, said Andy Keane, General Manager for the Tesla high-performance computing group at NVIDIA. “With this release, EM Photonics is making a meaningful addition to the NVIDIA CUDA eco-system by providing a mature, complete math library”.
CULA is currently available in three forms: Basic, Premium and Commerical. CULA basic is FREE and it comes with six common LAPACK routines. CULA Premium lists at $395 and contains a significantly larger number of routines. Keep and eye on these folks. ORNL is buying a 20PF GPU-accelerated system that needs solvers and EM Photonics has an early lead.
For more info, check out their website here.