What is GPU Workbench?
GPU Workbench™ is a complete platform for developing and deploying real-time applications that use NVIDIA CUDA technology. Based on the latest available GPU and CPU products, GPU Workbench systems are powered by Concurrent’s RedHawk Linux operating system specially optimized for real-time CUDA performance. GPU Workbench systems can be configured with up to 96 Teraflops of double-precision compute power.
GPU technology is widely used by scientists and engineers in applications that require massive parallel processing. Compute intensive processes in molecular biology, cosmology, particle physics, radar and sonar data analysis, economics, medical imaging and many other disciplines can now execute in a small fraction of the time needed to run them on a CPU. In addition to providing a CUDA platform, GPU Workbench offers key features for time-critical CUDA applications that require guaranteed hard real-time response to external events.
Jump to Section
RedHawk Linux Performance and Determinism
At the heart of each GPU Workbench solution is Concurrent’s RedHawk Linux real-time operating system. Compatible with the popular Red Hat® Linux, CentOS and Ubuntu distributions, RedHawk provides high I/O throughput, fast response to external events, optimized interprocess communication and NUMA memory management. RedHawk is the ideal Linux environment for complex real-time CUDA applications. Proprietary GPU drivers supplied by NVIDIA frequently make demands upon kernel resources that can dramatically impact realtime performance. RedHawk addresses these special challenges and guarantees the performance of real-time processes when CUDA processes are concurrently running on a system.
RedHawk Linux, which includes the latest version of the NVIDIA CUDA SDK, reduces the process dispatch latency of real-time processes in CUDA applications from hundreds of microseconds to under 10 microseconds. RedHawk minimizes jitter and provides better overall performance.
NightStar Debugging and Analysis Tools
GPU Workbench supports Concurrent’s powerful NightStar development tool kit. Users can debug, analyze, monitor, schedule and tune their real-time applications on GPU Workbench systems or remotely. Each tool runs on the target system nonintrusively, thus preserving the execution behavior of the real-time application. NightStar’s NightTrace tool includes an application illumination feature allows programmers to automatically trace CUDA API function calls and examine the values of parameters passed and returned as well as get detailed timing analysis of CUDA kernel executions. NightTrace also permits a user to graphically view the interaction between the Linux kernel and multiple application threads in realtime. In addition, NightTrace allows users to add trace points into the CUDA kernels that are executed by the GPU.
Customized GPU Platforms
Concurrent offers fully customized real-time GPU Workbench platforms that can contain up to eight of the latest NVIDIA Tesla and graphics cards and up to four CPU sockets. Systems come in standard tower, rackmount or desktop enclosures with up to 11 integral PCIe slots and optional expansion chassis. Configurations can contain up to 24 disk drives with optional RAID. Multiple GPU Workbench systems can be interconnected via 10 Gbit Ethernet or high-speed fabrics.
Real-Time Clock & Interrupt Module
The Real-Time Clock & Interrupt Module (RCIM) is a multifunction PCIe card designed for time-critical applications that require rapid response to external events. The RCIM includes a synchronized clock readable by multiple GPU Workbench systems, eight programmable timers, and twelve input and twelve output external interrupt lines.
An optional, on-board GPS module is available to align the RCIM’s synchronized clock to GPS standard time. One GPS-equipped RCIM can synchronize multiple systems in a cabled RCIM chain, or multiple systems equipped with the GPS module can operate from a common time base without any cable connections between the systems. POSIX timers based on absolute GPS time can be used to simultaneously start the execution of programs on systems which are not physically connected.
Custom Engineering from Concurrent
Concurrent is available to design and deliver customized GPU Workbench configurations for customers who require complete competitive solutions for demanding real-time applications. Concurrent engineers can provide special packaging including peripherals and enclosures, integrate third-party I/O cards, develop and integrate RedHawk Linux drivers, and perform application rehosting. Hardware and software is designed and developed to exact customer specifcations.