by Shanee Ben-Zur

A major theme emerging from this week’s GTC is the staggering – and increasing – breadth of applications that CUDA is enabling.

Tuesday, we highlighted sessions that expanded on CUDA’s use in medical applications; Wednesday, the efficiences developers are gaining through CUDA libraries. Today, we take a look at some of the sessions covering CUDA in high performance and large scale computing:

  • High-Performance C to CUDA Mapping – Benoit Meister, of Reservoir Labs, reviewed an automatic C-to-CUDA mapper prototype, which optimizes execution and data movement for a broad class of loop codes. With his powerful mapper, he discussed how using C as an input language can offer higher performance and performance portability.

  • Accelerating Explicit FEM Shock & Blast Simulations – Explicit finite element codes are widely used to simulate the response of structures and mechanical equipment subjected to shock, blast and wave propagation phenomena.  In his session, Nachiket Gokhale, of Weidlinger Associates, revealed how CUDA enabled an order-of-magnitude increase in the speed of these simulations using his firm’s commercial finite element code, NLFLEX. 
  • Enabling Large-Scale CCTV Face Recognition – Presenters Abbas Bigdeli and Ben Lever, of NICTA, instructed attendees on how CUDA and GPGPU can be used to perform large-scale facial search for both forensics and for CCTV face recognition.