In recent years clock rates of commodity processors have flattened and performance-per-processor core has stagnated. Blame this condition on the laws of physics: as processor clock speeds increase while die size remains roughly the same, the power/density ratio increases until no practical way exists to dissipate the heat. How do we solve this problem? We circumvent the laws of physics and break the power/performance wall.
Heterogeneous Computing with Specialized Hardware
One way to accomplish this is heterogeneous computing—using specialized hardware that accelerates specific portions of an application. Examples include using attached array processors and I/O cards containing graphics processing units (GPUs) as application accelerators. Another approach is to use reconfigurable hardware, such as field programmable gate arrays (FPGAs), to execute performance-critical portions of an application.
Such specialized hardware architectures can provide extremely high performance with much better power efficiency. However, heterogeneous systems are notoriously difficult to program, mostly due to the complexity involved in distributing and managing execution across multiple architectural models.
Heterogeneous Computing with Hybrid-Core
The heterogeneous architecture of Convey Computer’s hybrid-core systems is different. Convey systems combine the economies and programmability of industry standard Intel® processors with the performance and efficiency of a hardware-based, application-specific design.
These hybrid-core servers employ a reconfigurable coprocessor that augments the capabilities of commodity processors with processing elements optimized for performance-critical operations. Instructions executed by the coprocessor appear as extensions to the x86 instruction set and can be customized to directly address the application’s performance needs (Figure 1).
Figure 1. The Convey Hybrid-core architecture features commodity x86 processors tightly coupled with an FPGA-based coprocessor.
The Convey coprocessor is based on standard FPGAs coupled with standard multi-core Intel® Xeon® processors. Convey’s newest family of servers, the Convey HC-2 and HC-2ex, accelerate computing by providing higher absolute performance, increased functionality, and improved efficiency.
Personalities for Optimized Application Performance
The Convey hybrid-core systems adapt to different workloads through personalities—reloadable instruction sets specifically designed to achieve orders of magnitude acceleration in a variety of applications, including life sciences, data intensive computing, scientific research, military/defense, automatic speech recognition, and more.
Personalities implement specialized hardware architectures designed to execute specific algorithms. They are hardware designs, but they execute within the same virtual address space as the program that calls them. The coprocessor hardware and the x86 cores on the host system have the same view of memory and can use the same pointers to operate on data, subject to the same protection mechanisms. Moreover, personalities are loaded transparently as needed by the operating system, allowing multiple hardware architectures tailored to different algorithms to be used on a single system.
Better Performance: HC-1 versus HC-2
The Convey HC-2 systems increase application performance 2-3 times over previous generations of Convey servers (Figure 2). Earlier versions of the Convey systems improved performance-critical portions of the application on the coprocessor. Convey’s latest hybrid-core systems attain even greater acceleration by increasing performance on the entire application—even those portions not assigned to the coprocessor.
Convey’s HC-2 and HC-2ex provide more flexibility than previous generations of Convey hybrid-core systems. Now available in numerous configurations, customers can better match Convey hybrid-core technology to fit the performance profile of their particular problem. Customers can customize their implementation with a combination of models, Intel processors, memory configurations and I/O devices.
The Convey HC-2 family is completely binary compatible with our earlier generation systems. That means that applications run untouched on the new hardware, only at higher performance levels. Investments in software and personalities are maintained as you migrate to the new systems.
Better Energy Efficiency
Using hybrid-core computing technology not only achieves substantial performance increases, but you also save money in power and maintenance costs. When used as nodes in a HPC cluster, Convey hybrid-core systems deliver higher per-node performance, providing substantially better performance per watt than clusters based on commodity processors. Just one rack of Convey systems can easily replace numerous racks of commodity servers. You can increase the performance of your data center while cutting power consumption and saving money.
The next-generation Convey HC-2 systems are binary compatible with the earlier HC-1, provide 2-3 times the performance of those systems, and an order of magnitude more performance than commodity servers. The Convey HC-2 systems are available immediately for a range of applications and in a wide range of configuration options.
For more information visit Convey Hybrid-Core Technology.