Battle of embedded CPU architecture

Myself: I have been in touch with XScale in Marvell. Now I am in touch with TI's Davinci. This article on the is still well written.

 

The original article is as follows:

Many years ago, I carefully studied the architecture of the embedded SoC system. Later I found that after several architectures in the industry were finalized
What is innovation, so it is boring and no longer concerned.
I did not expect that after so many years, the wars of these architectures are still the same and there is no ending.
What architecture?

One is pure speed-based Processor SoC. If you cannot find a typical representative, you can use Xscale to represent it. Although
However, it also has the so-called MMX, but the effect is not very good. Network and NAS acceleration rely on it, but multimedia acceleration relies on
It is also too low. In the end, it still relies on its clock speed. It is said that it has more than 1 GHz. This is Intel's best choice for Marvell.
Expensive heritage-speed;

What are the benefits of a pure speed player? Two words: simple, simple, beautiful, software porting is the most cost-effective and labor-saving.
Upgrading is fast. To mention performance, you only need to increase the clock speed. What are the disadvantages? More and more processor manufacturing processes
High, 90nm is out of date, and 32nm is running out. The more advanced the process, the lower the cost, isn't it? NO, there is a balance here.
I don't know the complicated way of cost investment and recovery, but it is definitely expensive than other architectures I will talk about below.

The second type is the CPU clock speed center SoC, but there is a SIMD instruction set, it is very violent, for specific media Acceleration
Or other network processing operations are very effective. What are the benefits of this speed center type? Relatively simple process requirements, such as Beijing
The multimedia performance of a company's MIPS processor SoC after SIMD is higher than the Xscale of MHz (when
However, it also has very lightweight hardware acceleration), but the SoC processor clock speed is only over 300 MB! This uses. 1
8 technology can be manufactured. How cheap is it and the cost advantage is. This is innovation, and the performance is better designed at a cheaper cost.
Chip, which deserves pride in Engineering.
What are the benefits? Another typical Blackfin Processor of ADI also belongs to this type. It does not have SIMD,
It is the design of DSP. The programming model is simple, that is, the benefits of this type of processor.
It is terrible. It can run the operating system or complex software, and it can also be used to process specific algorithms.
Processing. Undoubtedly, it is better than the first architecture. I once thought it was the most perfect architecture in the industry.
Those powerpcs with SIMD (altivect.

But when I found out that this was not the Terminator, and another architecture that had been struggling to survive, I suddenly found a solution to the problem.
This is why I suddenly want to write this post.

Third architecture: processor + coprocessor, or processor + hardware accelerator. The typical architecture is the SoC of arm + dsp.
Chip. The most typical is TI's SoC. It's all about this arm + dsp. How many years have it been fighting with ADI?
I think he will win soon. This is what TI did before Da Vinci's dsc27, and after Da Vinci's omap
This is the case. It's just a matter of replacing the arm core with its own better dsp core with a faster update.

I once thought that this architecture is very retarded. It is not only complicated in development and debugging, but also complicated in arm, dsp, and debugging integration.
Working hard, the development threshold is too high, and the two processors in this architecture do not have a high clock speed. Otherwise, the power consumption is unbearable.
Two processors in SoC! Too much electricity consumption. Complicated manufacturing costs are not low. I have despised it for many years.
Arm7core, the previous dsc27 and so on, can only run a simple rtos such as ucos. Now those da Vinci
Arm9core cannot be used anywhere. It is more than 200 MB clock speed. It is impossible to run a modern OS well.

In fact, this architecture is the final stage of a large-scale product, that is, ASIC solution.
Method, low cost, dedicated! Speaking of this, if you are a bit confused, it is because you are from an academic school and have no industrial experience.
The architecture is forced to bind two slow CPUs together. Compared with the second architecture,
It is very bad, because 2 can run the OS at high speed, and can do what dsp does whenever necessary.
Scenario 2 and 3 both have good performance.

However, at last, the era of 3 has come. This era is to run modern OS, complex software, and develop DSP at the same time.
Things. 2. The capability is to do only one thing at the same time. 3. You only need to upgrade the two cores at the same time to solve this problem.
Slow, but it is much easier to double your clock speed than 2!

So is the battle of this architecture so far? NO, I think it's still coming. There are opportunities for all three architectures.

Their main battlefield is a 10 thousand-level product, or people want it to be the most omnipotent product.
Complex things are what you call a mobile phone. I don't know if it is a mobile phone.
Do you want your mobile phone to be omnipotent? Call, listen to mp3, play games, watch movies, take photos, take videos, and access the Internet...
When the last two architectures fail, they will not disappear. They will dive into those that only need to do one thing at the same time.
There are also a lot of such products and applications. This is the charm of embedded systems...

Of course, there are other architectures, such as multi-core and multi-core, that started with the Embedded SoC and then spread to Intel's X86
I have never been very interested in X86, but those multi-core MIPS are still busy on high-end network devices ..
The emperor often takes turns to do this. This year, the Battle of architecture continues...

It is not the theoretical battle of books, but the battle of industry. The debate on book theory will never replace the industry.
People, join the industry, that is heaven, that is also hell :)))

.


Myself: in fact, Xscale's multimedia and WIFI functions are very powerful. PXA168 I have also participated in development. Haha