Cortex-A 73, a CPU that will not overheat - Gary explains

Cortex-A72 - refinement and revision of the Cortex-A57. It looks forward about a year and we found in the heart of the Cortex-A72 SoCs such as Kirin 950 and 955, which are used in mobile phones such as Huawei and Huawei Mate 8 P9. Now ARM has announced another premium 64-bit ARMv8 new processor, the Cortex-A73. We know that the work on the new ARM CPU core, code named Artemis, and now it's official. So, what's Cortex-A73 brings to the table? Is it faster? Sure ... but more importantly it has made great strides in the field of power efficiency during periods of continuous use.

power efficiency and heat dissipation is everything when it comes to mobile CPU and they are also factors that affect the performance of mobile CPUs. On the desktop is not a problem because the PC is connected to power and has a large cooling fan, but the mobile world is very different. To keep things efficient mobile CPU designer has a few tricks that they can use. One is to throttle the CPU when it becomes too warm, which means to run at a lower clock frequency; Another is to use a heterogeneous multi-processing (HMP) setup like big.LITTLE, and uses more power efficient CPU cores for the time being; and the third is to use frameworks such as ARM Intelligent Thermal Power Allocation, which can dynamically manage the thermal budget of System-on-a-Chip - thermal budget reallocation from the CPU to the GPU (and vice versa visa) when needed.

When a smartphone is not very busy the CPU is free to spike highest level of performance for short bursts. Measures such as opening an app, web page rendering, or start the movie all make a momentary spike CPU performance. But after an open application CPU usage drops, and once the web page is displayed CPU just sits idle when you read the text, and so forth.

However, if you start an activity that forces the CPU performance is high, such as playing a complex game, then after a while the heat generated by the CPU (and GPU) will force Android to take action and re-arrange things so that the heat can be dissipated properly. As I mentioned earlier, which may very well include throttling the CPU that runs at a lower frequency (and therefore generate less heat).

Does this mean that the CPU has a peak performance levels that generate more heat than thermal budget allows, which is OK - even good, for short bursts. However, when used over a sustained period CPU usage should be modified so as to keep the nominal power budget, but that comes at the expense of performance ...

But what if the ARM CPU core can produce designs which generate roughly the same amount of heat when the spikes of CPU performance for short bursts, and when used for a sustained period of time? Or in other words, what if ARM can design a CPU that can maintain peak performance in the normal per-core power budget. Well, that is the purpose of the Cortex-A73.

warning

Before we dive deeper into the Cortex-A73 design, I need to clarify a few things. First, there are several different components on the SoC to produce heat, including GPU, image processors, video processors, display processors, and so on. If the overall heat level of the SoC increases due to the activity of the GPU, the CPU can still strangle even though it was not part of producing heat. Second, how any given SoC makers to implement Cortex-A73 in silicon include a process node is used will affect the results of performance / efficiency as a whole.

Cortex-A73

So let's look at some metrics around Cortex-A73. It is a 64-bit ARMv8 CPU core design that can run at speeds up to 2.8GHz and can be used in a big.LITTLE configuration. It can be built in a variety of process nodes, but it is expected that manufacturers SoC will make the Cortex-A73 SoCs based on 10nm or 14nm / 16nm. Overall 10nm Cortex-A73 offers 30% power savings compared to 16nm Cortex-A72, while generating 30% more performance. Some of those profits derived from the use of 10nm instead of 16nm, but the Cortex-A73 offers at least 20% power saving and performance advantages of about 10% to 15% when compared with the Cortex-A73, if they are both built using the same process node.

Micro-architecture

Cortex-A73 has been specifically designed for mobile work load and thus the internal optimization (including the branch prediction, pre-fetching and caching) was created with the mobile phone in mind. There are several important architectural changes in the Cortex-A73 when compared with the Cortex-A72.

double decode pipeline, compared to decode three-wide on the A72

The use of 4-way 64K instruction cache, instruction cache than 48K 3-way.

The new branch predictor with large Branch Target Address Cache (BTAC), along with Micro-BTAC to accelerate branch prediction.

Out-of-order execution engine optimized for high memory throughput with four full units of load / store out-of-order (two loads and two stores), compared with only one load and one unit A72 stores.

The new enhanced L1 and L2 cache took the algorithm that uses a complex pattern detection

Hexa-core than the octa-core

The use of octa-core processor has been very successful for cheaper mid-range phone. SoCs such as the Qualcomm Snapdragon 615/616 or P10 MediaTek has proved that there is a market for devices using eight 64-bit Cortex-A53 cores. Cortex-A53 has been very successful here because the cost / performance ratio, as well as high levels of power efficiency. But what is interesting is that the hexa-core Cortex-A73 SoC, with two A73 cores and four A53 cores, occupying roughly the same size as the silicon octa-core Cortex-A53. Traces of silicon is everything when it comes to the cost of making a SoC and even a fraction of a square millimeter can make the difference between SoC benefit and one that loses money for manufacturers. Cortex-A73 occupy less than 0.65mm2 per core.