GPU Computing Creates New Paradigm
The origin of high-performance computing (HPC) was based on clustering multiple CPUs, and more recently, clustering scores of multi-core CPUs to achieve higher levels of computational power. The GPU (Graphics Processing Unit), and the recent advent of GPU computing, have created a quantum shift in computing architecture by introducing a hybrid model whereby GPU I/O cards now work in conjunction with CPUs.
Utilizing the latest Intel® multi-core CPUs and PCI Express 2.0/3.0 technology, Trenton servers are ideal for building GPU computing solutions based on NVIDIA® Tesla™ GPU 20-series processors and the CUDA™ parallel computing architecture. The TCS4501 4U rackmount computer supports up to four Tesla cards in a rugged enclosure.
Servers Designed for GPU Computing
While CPUs are excellent sequential processors adept at serial operations, GPUs were designed from the start to excel at processing similar data that can be split into many pieces and processed in parallel. This is due to the fact that GPUs contain hundreds of parallel cores which are capable of running thousands of parallel threads.
Maximizing GPU performance, however, begins with the underlying hardware architecture. At Trenton that means single board computers & PCI Express backplanes which are designed from the start with data throughput in mind.
Strategic NVIDIA Partnership
Programmers take advantage of this architecture by directing the most performance-critical sections of their programs to run on multiple GPU cores. By offloading the CPU, 10x or greater performance increases are common, and this factor is sure to increase as future generations of GPU computing technology are developed.
As a NVIDIA Preferred Solutions Provider Trenton creates robust computing solutions for customers wishing to integrate NVIDIA’s broad range of computing products; including the latest Telsa K80, M60 and K40 GPU Accelerators, into their system operations. More information on performance benefits can be found on the Trenton NVIDIA K80 Benchmark Infographic. End users can also request access to a NVIDIA Remote GPU Compute Cluster to test their code on GPU Accelerators.
GPU Computing Applications
- Machine vision
- Computational Finance
- Asymmetric Cryptography
- Seismic image analysis
- Signals intelligence
- Thermal analysis
- Modeling & Simulation
- Genomics Research
- Video surveillance
- Radar image processing
Trenton’s knowledge of high-speed bus technology, BIOS configuration, system power requirements, shock & vibration, and thermal characteristics is incorporated into every computer system, single board computer, backplane and embedded motherboard.