DSP technology, in the eyes of some people, has already faced the feeling of a late hero, as we know it now. Freesacle, ADI, and NXP have long since stopped the development of new technologies, and currently there is only one big TI left with the banner of the Digital Signal Processor.
In the eyes of many people, the development of ARM and Intel has far exceeded the DSP in the past decade. Especially in ARM, its A9 dual-core and A15 quad-core have occupied the dominant position in the embedded market, but we can't deny that in the industrial field, the application of DSP still occupies a very important position. At Intel, after the launch of Silvermont, the low-power X86 is also known as a strong force. Some people will think that ARM and Intel can replace DSP in the future.
On this issue, electronic engineer Chen Junzhi gave a view like this:
He believes that for DSP, there are two explanations. The first Digital Signal Processing is digital signal processing. It is a discipline technology. Simply put, the analog semaphore in the real world is converted into digital semaphore and then analyzed and discrete. Digital operations in transformation, modulation and demodulation, etc. It is different from simple arithmetic operations such as addition and subtraction multipliers;
The second explanation is the Digital Signal Processor, a digital signal processor that simply embeds a special microprocessor that can quickly process digital signal analysis, discrete transform, modulation and demodulation.
As can be seen from the explanation, the DSP chip is a microprocessor chip that incorporates digital signal processing functions.
Then ARM, and Intel's chips are also microprocessor chips. In fact, everyone is working on computing. So what is the difference between a microprocessor without digital signal processing and a processor with this function?
The DSP processor has an indicator, that is, "multiplication and accumulation operation", which is a 32-bit multiplication multiplied by another 32-bit product plus a 32-bit number. This operation is completed in one or two instruction cycles. It is impossible to replace ARM and Intel microprocessors in one or two instruction cycles. They have to run several instruction cycles for multiplication. The DSP chip comes with a MAC module, which can be calculated N times faster than ARM and INTEL under the same frequency. Similar to the Fast Fourier Transform function, which often takes a long time to produce results in a chip without DSP functionality.
Then you will say, then my ARM and Intel chip industry will embed this module. Yes, it can be embedded. However, once embedded in this chip, the ARM chip can also be called a DSP chip.
Therefore, DSP is just a technology, a chip embedded with this technology, anyone can produce. In fact, some ARM chips are embedded with DSP functions. For example, many of Qualcomm's series of ARM master chips for mobile phones have their own DSP functions.
Intel's own development is the main control in the field of computing. In the embedded field of industrial aerospace medical and other fields, they use much digital signal processing. They don't involve much, so their stuff is basically not embedded in the DSP module.
In order to avoid direct competition with TI's C2000 series DSP, Microchip called its dsPIC series DSP chip DSC, which is actually a DSP chip, and they have been launching new products. Also, Freescale, ADI, NXP, they also have DSP production, but they can not compete with TI's dedicated DSP. The main point is that it is not enough to compete for FPGA chips.
Here again led to the FPGA technology, when it comes to this, perhaps it is really the real enemy of DSP. At present, the major manufacturers of FPGA chips are Altera, Xilinx and so on. They don't belong to Intel, nor do they produce ARM chips. They only focus on FPGA chips.
We know that a microprocessor chip is actually a large-scale ultra-high-density integrated digital circuit in a chip. The smallest unit is actually a transistor, and the transistor is integrated into a gate circuit. The gate circuit forms a chip internal CPU and a memory unit through a series of arrangement and combination. This is all cured before leaving the factory.
Then can I modify a certain gate of the CPU module to make it use it for other purposes? It has not been modified since it was factory-made, but now with FPGA, you can use the program to compile a CPU core of your own invention and embed it into the FPGA chip. Then you want to add the MAC module of the DSP, etc. You think that a CPU core is not enough, you can add another kernel (provided that the FPGA chip capacity and resources are large enough, of course, this chip is expensive).
Therefore, now Freescale, ADI, etc. see FPGA chips so strong, industrial-grade DSP costs are high, the price of selling is similar to FPGA, the price is lower than FPGA but can not compete with TI's DSP market. So no way, I had to quit slowly. Now TI is also dead by old customers. In the unlikely event that high-end FPGAs fall to the same price as they, they will have to quit.
Of course, the shortcomings of FPGAs, in addition to their own high cost, if you want to embed ARM, DSP core, these cores also cost more. At the same time, the HDL development difficulty of FPGA is much larger than that of DSP C, especially in the typical image processing algorithm, which also limits the development of FPGA to some extent. Although Xilinx's Vivado now supports high level synthesis, converting C to HDL, it is also a very difficult problem. This is not a wise choice.
In the eyes of some engineers, as a processor with a dedicated computing unit. DSP is a processor with many mathematical functions to meet the needs of the dedicated market. For example, the Harvard architecture, such as the computing unit with SIMD and VLIW technology, once gave DSP a lot of lead in the dedicated market. The advantages of TI DSP are low power consumption, high computing performance, mature and stable, coupled with obvious price advantages.
From the perspective of architecture, TI's solid Haval structure has not changed much in the past ten years, and the speed of arm and intel is obvious to all. Look at two small examples, intel's branch forecast is a lot of leading TI home; another example, TI's second-level cache, on-chip off-chip also requires users to manage, cache miss penalty is also quite high, and Intel's superb caching technology makes you basically feel Not the difference between the film and the film.
Look at the arithmetic unit, now arm has neon, intel has sse, avx, out of order execution, and gpp for special image calculation; plus FPGA. These product computing units are already very powerful, so the mathematical performance of DSP is no longer a leader. For example, two 4-byte integers in the DSP do multiply-and-accumulate operations. TI's C674x requires 4 clock cycles, and Intel's sse3 handles 16-byte multiply-and-accumulate operations with only 4 cycles.
In the field of complex algorithms, such as image processing, pattern recognition, DSP is really not enough. Running a boosting is hard to die, and the compressed sensing algorithm is pre-research, and most of them still run on the intel and amd platforms. .
Simple algorithm, there are too many optional processors, so the most important thing is the price. I am optimistic about arm, and fpga has no price advantage. Of course, c2000 still has a lot of fans, but it has already eaten a lot of market by other processors.
The above is based on the calculation performance as the core, and does not involve power consumption. In the same amount of calculation, the time consumption is similar, DSP power consumption control is the best, ARM, atom is not as low as DSP power consumption (asic may be good, but not a general-purpose processor). This is also why in the embedded complex algorithm market, or DSP is the mainstream.
As far as Chen Gong seems, the computing advantage of DSP is not in the ordinary four arithmetic. With the Fast Fourier Transform FFT, the FFT of any other microprocessor without a DSP core is inferior to the DSP at the same frequency. We do DSP or other embedded algorithms, which are based on "Automatic Control Principles" or "Signal Systems" rather than "Data Structures".
He believes that DSP is not a wise choice for boosting. DSP can design a real-time sound filter for the frequency domain computing power (all the sounds in the noisy environment are transmitted to the AD circuit through the microphone, and then through the DSP operation, some of them The personal voice is taken out and the voice of someone is released in real time through the speaker. Such a digital filter can be made smaller than a mobile phone). In the industrial control and aerospace field where TI DSP is the largest supplier, most DSP applications have nothing to do with computer algorithm technology.
When it comes to the low power consumption of DSP, of course, when DSP is doing filtering and audio processing, it does not use the DSP function (just use them as a single-chip microcomputer), so the power consumption is very low. But if you use DSP as SPWM or digital filter, the power consumption will be big. I have forgotten the specific comparison values. I used to compare the power consumption controlled by ARM when I used the mine motor control. (But it seems that Microchip's DSP chip consumes less power when filtering.)
In fact, 16bit/32bit DSP chips (or other 32-bit MCU/ARM embedded DSP-capable chips) have been coexisting with various microcontrollers, ARM, PowerPC, X86, etc. in the long-term development, and not only TI, but many other manufacturers. All of them are specialized in the production or embedding of DSP modules (such as Microchip's dsPIC series DSP chips, which have been continuously updated). This means that DSP has signal processing capabilities that are not available in other architecture processors.
Looking at it comprehensively, FPGA is the big killer in the future. What do you think of it? Of course, the whole discussion is just a word, I hope everyone can look at the whole discussion rationally.
Innosilicon Asic Miner:Innosilicon A9 ZMaster,Innosilicon A9++ ZMaster,Innosilicon A9+ ZMaster
Innosilicon is a worldwide one-stop provider of high-speed mixed signal IPs and ASIC customization with leading market shares in Asian-Pacific market for 10 consecutive years. Its IP has enabled billions of SoC's to enter mass production, covering nodes from 180nm to 5nm across the world`s foundries including: GlobalFoundries, TSMC, Samsung, SMIC, UMC and others. Backed by its 14 years of technical expertise in developing cutting-edge IPs and ASIC products, Innosilicon has assisted our valued partners including AMD, Microchip and Microsoft to name but a few, in realizing their product goals.
Innosilicon team is fully devoted to providing the world's most advanced IP and ASIC technologies, and has achieved stellar results. In 2018, Innosilicon was the first in the world to reach mass production of the performance-leading GDDR6 interface in our cryptographic GPU product. In 2019, Innosilicon announced the availability of the HDMI v2.1 IP supporting 4K/8K displays as well as our 32Gbps SerDes PHY. In 2020, we launched the INNOLINK Chiplet which allows massive amounts of low-latency data to pass seamlessly between smaller chips as if they were all on the same bus. With a wide range of performance leading IP in multiple FinFET processes and 22nm planar processes all entering mass production, Innosilicon's remarkable innovation capabilities have been proven in fields such as: high-performance computing, high-bandwidth memory, encrypted computing, AI cloud computing, and low-power IoT.
innosilicon zec miner,Innosilicon A9 ZMaster,innosilicon a9 50ksol miner,innosilicon a9 miner,a9 50ksol zmaster
Shenzhen YLHM Technology Co., Ltd. , https://www.asicminer-ylhm.com