Aerospace and defense applications demand products that are sensitive to their SWaP (size, weight and power) restrictions. In the past this put some of the highest performing GPU products out of reach because their power requirements, acceptable for a desktop computer or a data center, were too high for a device intended for an embedded application. This led embedded solution providers to use GPUs that were targeted mainly at the mobile and laptop market. While these mobile GPUs had lower power requirements they also were not as high performing as the same generation’s desktop GPUs.
With the introduction of the NVIDIA Pascal product line the performance gap between GPUs for desktop and mobile has been largely erased. In fact, NVIDIA no longer designs chips specifically for the mobile market. Instead with their new power-efficient Pascal chips they are able to promote many of the same chips for both mobile and desktop.
The NVIDIA Quadro GPUs include full NVIDIA Management Library (NVML) support, which includes an API that allows the end user to dynamically limit the operating power. This allows Quadro embedded GPU solutions to operate at significantly less than the maximum GPU operating power, providing another tool to allow system designers to meet SWaP targets. By profiling the GPU’s power versus performance system designers can fine tune the device to provide the required processing at an acceptable power level for their specific application.
One of the high-performance Quadro Pascal chips of interest to the aerospace and defense market is the GP104. This GPU can provide an impressive 76 GFLOPS/Watt. This is far more processing performance per Watt compared to the previous generation’s mobile solutions, which provided at most 30 GFLOPS/Watt, while maintaining a very similar operating power.
For those applications that require a lower maximum operating power the Quadro Pascal GP107 is particularly noteworthy. This GPU can provide 46 GFLOPS/Watt. This is still significantly higher than the previous generation’s mobile solutions, providing a high performing module at 25W, or half the maximum GPU operating power.
Figure 1: Low power GPUs are typically required for XMC and VNX modules, while higher performance and higher power GPUs can be used in VPX modules or in custom SFF designs which are typically installed in a chassis that is able to dissipate more heat.
The Quadro product line is NVIDIA’s professional line, with significant advantages to the aerospace and defense industries over NVIDIA’s consumer-focused GeForce line. As with most professional versus consumer products, the Quadro GPUs are more rigorously tested. NVIDIA provides a longer warranty for the Quadro products: 2 years for Quadro versus 90 days for GeForce products. The Quadro line also has extended product availability with full product lifecycle management.
The drivers that are provided for the Quadro product line are tested for professional use. This results in long life, stable drivers, suitable for mission critical use. Quadro products can support a larger total memory compared to the GeForce product line. Quadro products also include support for NVIDIA GPUDirect™ which provides other devices direct access to CUDA™ host and device memory, resulting in significant performance improvements in data transfer times.
With the Quadro Pascal’s larger data processing capability application engineers will feel free to develop more complex algorithms, which will put increased demands on memory utilization. The Pascal architecture provides a larger virtual memory address space, enabling GPUs to access the entire system memory plus the memory of all GPUs in the system. The on-demand page migration engine allows the system to migrate pages from anywhere in the system to the GPU’s memory for processing. This improved memory handling can result in significantly improved algorithm efficiency.
Of course, to complete the requirements for aerospace and defense devices the GPU must be integrated into a product that operates reliability in harsh environments. NVIDIA relies on its partners to integrate their Pascal GPU chips into product designs that fully comply with the manufacturing and quality standards required by aerospace and defense, such as MIL-STD-810, RTCA DO-160, ISO 9001:2015, AS9100 and IPC 6012 CLASS-3 (High Reliability Electronic Products).
With the latest generation of NVIDIA Quadro Pascal GPU chip-down design modules system developers will be able to get the most processing power ever available in a rugged video graphics board.