Choosing the Best Processor for the Job
September 18, 2020 | BY: Mike SouthworthDownload PDF
When it comes to embedded computing architectures for defense and aerospace applications, there are a number of questions that should be asked when selecting the processor (or processors), such as:
- Can a standard central processing unit (CPU) handle all required tasks?
- Is a graphics processing unit (GPU) or a field-programmable gate array (FPGA) needed?
- What architecture considerations should factor into your decision for minimizing size, weight, and power (SWaP)?
- Since the end goal of most systems is to implement an algorithm to perform some pre-determined task or function, which architecture is going to achieve the goals?
The CPU is the brain of what people typically think of as a computer. It runs an operating system (OS), as well as user applications. It can receive inputs, either from the user or from other devices, executes a series of instructions, and then finally outputs a result of those instructions to a screen, a printer, another program, etc.
CPUs contain at least one processor core, which is the individual processing unit that receives and executes instructions for a computing task. Modern CPUs are traditionally multi-core processors, containing anywhere from two to 64 cores. Most modern computers contain multiple CPUs, each with integrated cores that often support double the number of virtual cores (known as threads). For example, an application server with two 16-core CPUs may have a total of 32 cores and 64 threads. Multi-threading enables more simultaneous work to be accomplished at a time, which improves the overall computing throughput. However, multi-threading may reduce throughput in certain cases, like I/O bound operations.
When talking about the speed of a processor, we typically look at two principal factors: the clock rate and the instructions per clock (IPC). Your CPU processes many instructions (low-level calculations like arithmetic) from different programs every second. The clock rate measures the number of cycles your CPU executes per second, measured in gigahertz (GHz).
During each cycle, billions of transistors within the processor open and close to perform computing work. Put the clock rate and the IPC together and you get the instructions per second (IPS). With today’s processors, millions of instructions per second (MIPS) is the more common unit of CPU performance. Other factors, such as the performance of the memory hierarchy, also affect processor performance even though they’re typically left out of IPS calculation.
A common type of CPU board architecture in military and aerospace applications is a single-board computer (SBC), which integrates a microprocessor(s), memory, input/output (I/O) and other features required of a functional computer. A performance-optimized SBC for highly-parallel signal processing applications is the digital signal processor (DSP), a class of CPU with a higher number of parallel processing CPU cores – typically 8-16 multi-threading cores per processor – offering 16-32 parallel processing threads, often with two or more CPUs on the same processing module.
Almost every modern computer today has a CPU, regardless of its size, function, or performance.
This white paper reviews popular processor architectures to help you make an informed decision when defining your electronics payload.