Rebirth of the Tech Madman

...

At this time, Intel CEO Andy Grove was listening carefully to Tang Huan's outlook on the next generation of personal computers and the role that the intel80486 could play in it. After all, no matter how the engineers on both sides cooperate or argue on technical issues, the final decision will still be made by the big boss.

For Intel, which puts all its treasures on the microprocessor business, there are currently two major problems-the latter has a slight upper hand in the battle between cisc and risc, and the main product x86 series cpu is always better than Motorola mc68000 series in terms of performance. Competitors in the same cpu category are weaker. Even the latest generation of microprocessors of the two sides - intel80486 and mc68040, also continued this trend of strong and weak comparison.

His own products are so unsatisfactory that Andy Grove, who is outspoken, wants to use Tang Huan owns 20% of Intel's shares as an excuse to lobby the other party to abandon Motorola's mc68000 series cpu, but he is embarrassed to say so.

In this case, Zheru's deep skills in compilers, drivers, operating systems and other low-level system software are very important-in terms of code optimization, as long as Tang Huan asks his subordinates to pay a little attention to any party, other can be promoted.

It has to be said that the status of software is becoming more and more important now. In addition to being able to stick to users through the application software operation habits that have been developed on the terminal, it also largely determines whether the new hardware can play the designer's hope. performance.

And Zheru has now achieved such a situation of supporting two companies with one hand, and Intel has a tendency to rely on others invisibly. Of course, this kind of embarrassment is different from being subject to IBM at the beginning, and it is entirely due to the natural ecological checks and balances.

Fortunately, Tang Huan affirmed that the attitude of intel80486 seems to surpass that of mc68040, which made Andy Grove feel relieved.

In terms of performance, the mc68040 is indeed slightly better than the intel80486. However, the heat generation of the mc68040 is quite difficult, and it may be difficult to make up for it through the process, which has brought a considerable impact on the design of the computer host. Tang Huan raised his hand and pointed outside, You have also seen the direction of today's press conference. Multimedia and video games can squeeze out the performance contributed by hardware specifications at any time. Intel80486 and mc68040 still need the cooperation of Zheru in this regard. Processor to share the pressure.

Andy Grove naturally had no way to refute, so he had no choice but to aim at the future and said: The intel80486 has integrated the cache and the floating-point unit into the chip. Let’s learn from the design concept of risc, such as superscalar. This requires further communication between Intel, Zheru and Fangyuan on patent licensing.”

At this stage, the reason why the risc processor is faster than the cisc processor is that the industry and academia are quite pessimistic about the prospect of the latter. One of the main reasons is the parallel computing capability of the former.

Compared with serial computing, parallel computing can be divided into parallel computing in terms of time and parallel computing in terms of space. Time parallelism is mainly instruction pipeline technology, while space parallelism uses multiple processing units to perform concurrent computing tasks, such as superscalar.

Instruction pipeline technology splits the computer instruction processing process into multiple steps, and then hand it over to multiple hardware processing units for parallel execution to speed up the instruction execution speed - very similar to the assembly line in the factory, the computer instruction is equivalent to the product on the assembly line conveyor belt, Each hardware processing unit is equivalent to a worker next to the assembly line.

Intel has implemented a certain degree of instruction pipeline technology inside the intel80486-under the most ideal conditions, the intel80486 can process one instruction in one time period, and then achieve twice the performance of the intel80386 with the same clock frequency.

The superscalar architecture can be regarded as mimd—a design scheme of multiple instructions and multiple data, which often implements instruction pipelining technology, and the mechanism is naturally more complicated.

For microprocessors with this architecture, there will be multiple internal execution units, such as arithmetic logic units, shift units, multipliers, etc., so that the entire CPU can be assigned multiple CPUs at the same time within one clock cycle. Instructions are executed in different execution units, finally achieving instruction-level parallelism.

As early as the 1960s, Control Data Corporation's supercomputer cdc6600 was the first to carry out superscalar design, but the realization of this mechanism on a microprocessor was the first of its kind in the 1980s with Zheru's risc microprocessor holder.

The holder of a lot of cutting-edge technology concepts such as the superscalar architecture has been successfully realized, and its performance is outstanding. Top-level models help Fangyuan supercomputer become the fastest computer in the world; mid-to-high-end models help Fangyuan network servers and Fangyuan personal workstations compete for the market share of traditional mainframes and minicomputers; lightweight models help telecommunications switches, Internet routers and other equipment with data throughput capabilities Achieving a leap forward; the most interesting application of a specific model is the cpu that became the third-generation Xiaobawang game console. Although it has not yet been officially released, the evaluation media is overwhelmingly full of praise.

Another successful application of this parallel technology is video processors, audio processors, and other multimedia hard-coded decoding chips. For example, the current CPU does not have the ability to softly decompress and play VCD, so it must rely on a special video accelerator card.

For such technologies, even if the chip is reverse engineered in the laboratory, it cannot be fully analyzed. Whoever is ahead will naturally occupy the advantage in patents.

At this stage, Intel is facing the siege and interception of Motorola, amd, National Semiconductor and other companies within cisc, and there are uncertain factors in product prospects brought about by risc outside, it is not an exaggeration to describe life as walking on eggshells. In terms of patents, that is, the little brothers who can bully amd and the like, even bend down in front of Zheru.

In order to make Intel's research and development more smooth, there is no problem in principle. Tang Huan said happily: You are not clear about my pursuit of microprocessor performance. Haven't I been manufacturing intel80286 and intel80386 by myself these years? In order to get products with higher clock frequency.

Andy Grove asked curiously: During this raging battle between risc and cisc, not only did Don not express his attitude, but he still seems to have confidence in cisc?

In this kind of technology dispute, everyone insists on their own opinions. No one can guarantee that their predictions are 100% accurate. I represent the interests of many teams, so I don't want to talk about it. However, I do still have doubts about the future prospects of cisc. Great hope. Tang Huan replied frankly: After all, in a complex environment, the advantages of cisc cannot be replaced by risc. The real technology development trend should be that cisc constantly absorbs the advantages of risc, just like Intel The next-generation microprocessor intel80586 will introduce a superscalar architecture.

Andy Grove laughed, This should be the most exciting news I've heard in recent months.

However, Intel seems to have little confidence in the future of its main product - the cisc-based x86 series microprocessors. It is also planning to release a new risc product - inteli960 this year? individual shareholders.

At present, the fact that the performance of risc processors exceeds cisc processors cannot be ignored. In terms of the excellence of product design, x86 family microprocessors are still not the best one in Intel's research and development project sequence. We have to make two-handed preparations. ’” explained Andy Grove.

In a way, Intel's inteli960 is a bonus.

When intel_iapx_432, which tried to directly support high-level programming languages ​​in hardware, failed completely, Intel turned to risc to find a solution. And Siemens, which tried to acquire Intel but was preempted by Tang Huan, based on its own business needs, cooperated with Intel on a project called biin to develop a fault-tolerant multi-processor high-performance computer system. High-end markets such as transaction processing, national grid services, military, and industrial controls such as nuclear reactor management have a share.

The general cooperation model between the two parties is that Siemens, which has strong financial resources, will provide money, and Intel, which is specialized in technology, will provide people—most of them come from the failed intel_iapx_432 project team. In addition, in order to avoid restrictions on sensitive regulations, Siemens owns the software and documents, Intel owns the chip design, and mutually authorizes each other in terms of the distribution of work results.

However, this high-end field is also highly competitive. The system fault-tolerant application market includes tandem_computers and stratus_computer, and the parallel computing market has competitors such as sequent_computer_systems, pyramid_technology, and alliant_computer_systems. IBM and Tang Huan also participate in it in different forms. As a result, until now, the biin project has spent nearly 300 million U.S. dollars, but the effect is minimal. Intel doesn't care, but Siemens, which is responsible for paying the money, can't help but think about quitting.

The upper management can only use numbers to calculate the gains and losses, but the lower engineers are not willing to admit that the R\u0026D process has failed. The person in charge of the project, Glenford Myers, tried his best to lobby that the risc microprocessor inteli960 developed for the biin system was excellent enough to be introduced to the market even as a separate product—a general-purpose cpu.

Intel's management, who had originally intended to bet on both cisc and risc for insurance, agreed to this proposal smoothly.

Andy Grove finally said: We have evaluated that inteli960 is doing well, and it should have a good prospect in the embedded application market. If network cards, scsi controllers and other products use it as a CPU, it will not only meet the performance requirements. If necessary, it can also greatly improve stability.

I believe in this point. After all, inteli960 can be called overkill. Tang Huan nodded, But I want to remind you that the reason why Intel gave up the company's traditional business of memory before was because of too much pressure from market competition. , also in order to be able to focus more on the microprocessor business; but now you are hesitating between cisc and risc, and you are suspected of going astray, so you have to think twice. When you are doing marketing, do you say that the new risc Is the product faster than the x86 series processors that are contributing to the company's main profit? Users will definitely have doubts about how to choose.

The x86 series processors are indeed not as good as standing out from the crowd, but don't forget that they have a considerable number of applications, and many of them are trump products with a wide range of applications and a solid user base. Pure technology Excellence is not the only decisive factor in winning the market, you don’t understand this truth, do you?”

Andy Grove pondered and probed: What Tang means is that Intel's product development direction should be to stick to the x86 series processors, and introduce risc technology on the basis of cisc, and take the road of integration?

I really think so. Tang Huan nodded bluntly, At this stage, as a separate product, the video accelerator card will undoubtedly contribute a lot of profits, but as long as you have some long-term vision, you can judge It can be seen that this is only a transitional solution, and future microprocessors can fully include its functions.

So, I suggest that Intel's next-generation microprocessor intel80586, in addition to introducing superscalar, should also enrich the instruction set to directly support multimedia operations. Zheruhui will take a friendly and open attitude on related patents.

Andy Grove touched the hearing aid on his head, thought for a long time, and finally began to see the light in the muddy waters of the cisc and risc dispute.

After the enlightened counterpart left happily, Tang Huan ushered in another wave of visitors—Sony President Akio Morita and President Norio Oga.

These two people's recent intensive inspections in the United States have attracted the attention of the richest man, which makes him guess that Sony may also be in troubled waters in a new field for him.