A few days ago, Longxin released a 3A3000 and a series of products at the press conference. And the dragon core developer plan, Longqi University Program, and Longxin Industry Fund, also announced at the press conference. However, the author introduces the next generation of CPU - 3A4000, which is R & D, is being developed.
What is the difficulty of designing the Dragon Core 3A4000, what is difficult to achieve this indicator? What is the possibility of Longxin 3A4000 performance?
About the progress of 3A4000
Regarding the dragon core 3A4000, the dragon core has given a high hope, especially after absorbing the lack of 3A2000 and 3A3000, has made modifications, and the change is relatively large, the ideal goal of the dragon core is to implement the SPEC06 fixed point in the GCC compiler, but this It is the ideal state. After all, the CPU jumps or not meets the plan, even in Intel, AMD has this black history.
First introduce the progress of 3A4000. The research and development of 3A4000 is slow, mainly two targets, one is the GCC compiler, and the SPEC2006 test can reach 10 /g, and the other is the main frequency to 2G.
What is the concept of 10 / g? AMD's Zen probably published 3.4G, 31.5 points, and the following is 9.26 / g below. However, because some optimization options are not open, and the compiler is GCC4.6, change the GCC compiler version, open some optimization, 10 + / g results ZEN still can be achieved.
Intel's Haswell, like a class, because there is no maintenance from the AVX instruction, and the cache is "castrated" and other factors, it is 9.3 / g, and I5 4460 because it is "castrated", it can be 11 / G, netizens test Skylake, approximately 11 + / g (after all, INTEL is in the year of the toothpaste).
Therefore, if the dragon core 3A4000 can be 10 / g, the gap between the microstructure and intel, AMD is relatively small. However, considering the go up, the larger the microstructure optimization is more difficult. Like 3A4000 is basically a little bit of 3A3000, it is difficult to achieve this goal.
How big is the difficulty? As a reference, the force of the AMD is used to use the force of nine bull two, and the microstructure GS464E of the dragon core 3A3000 is twice the bandwidth is twice the band, and the GS464E and K10 are equivalent to the overall GS464E and K10. It is equivalent to a dragon core to complete AMD from K10 to ZEN's span.
AMD has been used for so long, and the dragon core is to achieve this, and it is difficult to know.
The second is the main frequency, the dragon core 3A3000 is not going, although there is a relatively limited factor in its own back-end design capacity, but also related to human, financial, and competent efforts to work with the foundry. The industry's foundry is only 60% of Tongtu, and the ST of Longxin has only 80% of the Thai power over the same process, while TSMC and Intel have a certain gap.
In addition, if the main frequency should be done, the IC design company and the foundry must establish a very close relationship. Open some technical information, like Intel and the AMD, IBM is IDM vendor, so it can open technical data design and technology Manufacturing continuously engines. Although IBM and AMD sold the wafer factory, it still maintains past contacts in the past, and many technical reserves and materials have retained, so they can continue to do high-frequency CPUs. But the dragon core and ST cannot establish this relationship.
Some of the substitutes in the territory remain to improve - the consistency of domestic processes has a certain gap than overseas processes, including different batch consistency, simulation model and actual transistor parameters consistency and foreign crafts There must be a gap, there must be more margins when designing, and Monte Carlo simulates the range of parameter fluctuations.
Longxin 3A2000 was developed because of the 3A3000 in the ST Formula for 1 year, while the 3A3000 in the ST Film was very smooth, and it took about 1 year from the project to the flow. In other words, the Dragon Core has established a close relationship with the domestic foundry factory. The domestic foundry is only able to say: Chen Chen is not arrogant!
In addition, the more complex the microstructure, the larger the back-end design pressure, such as the hard work difficulty of doing the A53 clocking 2G and putting the zen's main frequency to 2G is not a grade. Like 3A4000's microstructure, design goals close to Haswell's things, for any domestic team, the backend design is very large, and the dragon core 3A3000 and dragon core 3A4000, still use the same manufacturing process, so it is impossible Through the improvement of the process, improve the frequency, completely relying on its own design capabilities to increase the clock 30%, and it is difficult to know.
Finally, although the results of the current microstructure optimization are more gratifying, and there is already the simulation results. But because of 2G frequency, it will eventually do some weigh. Finally, how to take out how the product is only tested.
The author makes a prediction, the dragon core 3A4000 may use the ST 28NM FD SOI process, if time, the conservative estimate is 1.8-2.0, the SPEC06 GCC compiler is fixed at 16-18 points.
If you spend some time to optimize, and the dragon core team acts, the 3A4000 clock is 2G, and the SPEC06 GCC compiler is fixed to 18+ points.
If the dragon core team is super-level, the ST does not drop the chain, perhaps the SPEC06 fixed point 20 points under the GCC compiler.
About Longxin walks in the road and development direction
It must be emphasized that although the network is full of unreliable reports on dragon core, the dragon core has a very clear understanding of its ability, and intel, AMD gap.
In 2020, the direction of the dragon core is the application of industrial control, Netan, and oil, electricity, transportation, finance and other industrial applications. The application of these industries is related to national information security, and on the other hand, there is no integration like desktops and mobile phones. Software ecology problem.
For the PC of the dragon core, 3A3000 is mainly targeted by the party and government army office and dragon and lover, developers. Long Core has no ability, and there is no heart to enter the civil market with Wintel.
The idea of Longqin is to step by step. The first step can support yourself. In the special market and some industrial embedded applications, it can raise the team, and the money earned can support the development of the Longxin No. 3 series chip. This is a lot of achievements, which can be said that Long Core is the least in several official background IC design companies, and it is also extremely rare to realize the government weaning, self-contained, self-contained.
The second step is to a specific market, such as the party and government office PC, and encourage developers, enthusiasts use dragon core computers. At the same time, further development of the industry market, let more industry and embedded equipment use dragon core, expand its own base, increase market share and business income, and gradually establish software ecology of industry alliances and specific markets - such as software to meet the party and government Office use, or satisfied with a high scientific research unit, or use of the top ten military research institutes.
The third step is to enter the civil market. It is still very distant to this step, there is a long way to go. Longxin is currently in the second step.
The reason why Longxin's road will be so difficult, because all things rely on themselves, it is impossible to obtain a national huge funding support, go out to get a kernel outside, and then use the best craft of Taiwan to do a CPU / SOC, or it is impossible Like some companies buy ARM IP to integrate, and there are also ready-made AA systems to support commercialization.
Longxin must expand its own instruction set, develop a variety of IP, rewrite the operating system according to the open source software, build an industrial alliance and software ecology. This step is naturally more slow than cooperating with overseas manufacturers or joint ventures, but wins on the roots, but can also achieve autonomous control, but also master the ability to design high performance CPUs.
And cooperation / joint venture, it means that many things outdoor manufacturers have helped you do it, you will be able to do it, and there will be many restrictions in the joint venture or cooperation, and foreigners are not live Lei Feng. As well. For example, in the joint venture, there is a line of drawings. Chinese engineers cannot be changed. If they have to be changed, they must be reported by foreign companies. This limiting provision eliminates the possibility of Chinese engineers by modifying the original design, gradually educating foreign technology.
In the long run, use joint venture / cooperation model, not only unable to form a third level independent of AA, Wintel, but also its ability to develop. This is also why several major cars and foreign giants have been joint venture for many years, but in independent innovation, it still lacks the reasons why.
This article originally address: https://www.eeboard.com/news/cpu-3a4000-wintel-aa/
Search for "Love Bo.com" to pay attention, daily update development board, intelligent hardware, open source hardware, activities, etc., you can make you master. Recommended attention!
[WeChat scanning picture can be directly paid]
Technology early know:
Intel finally hit a safe vulnerability patch with nearly ten years.
Intel releases new M.2 wireless module Gigabit Wi-Fi / Bluetooth 5.0 two in one
Improve the Valley, the fascination of Meizu: The processor problem is a headache!
Ledith semiconductor launches a new embedded visual development kit for mobile related edge applications
Don't you want research equipment? Diy So Easy
Our other product:
