Customers often ask, “Will this CPU be fast enough?” or tell us, “We need a quad-core, at least 2.5 GHz.” These situations can be equally tricky to handle – we want to make sure our customers get the most appropriate performance match for their application, without unnecessarily increasing the cost of the hardware. So, we need to make sure our customers are comparing apples to apples, not apples to…orangelos.
In the 90’s and early 2000s, it was simple: The higher the clock speed, the higher the performance, with Intel and AMD continually trying to one-up each other in the GHz race. Now, there are a lot of factors that come into play, such as:
- CPU core architecture
- The number of cores/threads
- Amount of cache
- On-demand overclocking or “Turbo” features
- Integrated graphics
For example, when VIA launched the Nano CPU a few years ago, it had the same 1.6 GHz clock speed as Intel’s competing ATOM N270, but lacked the N270’s hyperthreading. Due to the more advanced core architecture though, the Nano was actually significantly faster in most applications and benchmarks. Clock-for-clock, it was effectively almost twice as fast in some of our testing.
Similarly, based on PassMark CPU scores, we can see that the AMD T56N (1.6 GHz, PassMark score of 733) falls right between the last-gen ATOM D525 (1.8 GHz, 715) and the current-generation ATOM D2550 (1.86 GHz, 748). This is thanks again to a more sophisticated core architecture.
When we look at the Celeron P4500 the difference is much more dramatic: at 1.87 GHz it gets a PassMark score of 1256, compared to 748 for the D2550. This, despite a clock frequency difference of just 10 MHz.
The recent shift in focus from CPU to GPU and heterogeneous computing confuses the issue even more. With discrete-class graphics and a low-power CPU, AMD’s G- and R-series Fusion APUs offer a bit of a hybrid approach (hence the orangelo reference), which allows for new possibilities in the industrial and embedded space.
With the such a dramatic increase in graphics performance, the Fusion APUs can decode multiple HD video streams for digital signage applications, perform image processing for machine vision, and provide high-performance vector-based computations, all while consuming less than 25w.
So how do you figure out which processor is best for your application? Check back next week for Part II.