The two Saul architecture CPUs that participated in the evaluation this time have a core frequency of 75MHz. They use the most advanced 0.5 micron process technology and integrate the enhanced multimedia instruction set MMX developed by Yuanxin. Thanks to the 0.5 micron process technology, they can integrate 64K full-speed secondary cache-their operating frequency is the same as the core frequency.

At the beginning, the two teams had a heated discussion on whether to integrate the second-level cache, how big the integration is, whether full speed or half speed, and in the end, Su Yuanshan finally made the decision, allowing Xinghai to adopt a more gentle and market-oriented solution.

In Su Yuanshan's plan, Xinghai's CPU department will be the main force in joining AMD to attack the INTEL throne in the next few years, so the market's acceptance ability must be considered. In contrast, the CPU department on Yuanxin will depend on the development of the situation - the best vision is to obtain all the authorization of Intel for the X86 instruction set next year, and hello, I am fine.

I almost had to take the curve to save the country. After Xinghai went public next year, I rely on the raised funds to establish a new joint venture in mainland China with Yuanxin and Texas Instruments, the kind dominated by Xinghai. Then the new company launched the CPU designed by Yuanxin.

The worst thing is that Yuanxin CPU department directly switched to a scientific research unit, stayed dormant for another five years, and then went straight to the 64-bit processor.

Therefore, it is not difficult to understand that Su Yuanshan even attaches more importance to cyrix than the CPU team in the Shanghai Science and Technology Park.

...

The camera faithfully recorded the progress of various tests.

When the test results are counted in real time, all testers, including the design team, show an incredible look.

"The effect of level 2 cache is so great?" Claude's deputy, Holps, who had been connecting with Li Mingliu's team, had been suppressing his excitement. When the test was halfway through, he finally couldn't help but sigh.

"It should be an increase in the command hit rate, which greatly improves the command execution efficiency." Claude's face was very calm, but he could not suppress his surprise: "In addition, the overall structure is much more advanced than Pentium."

Although in the design and simulation stages, the ability of this CPU can be judged by calculation alone, "it is always shallow to learn from paper, and I know that this matter is going to be broken." This is a recognized truth in the field of chip design.

Everything is virtual before completing the flow.

"Now it seems that the biggest problem with this CPU is the cost." Claude took out his phone, glanced at the time, and prepared to call Su Yuanshan.

The test has been conducted for more than half of the time. After the ongoing quake, the new CPU has even completed an epic transcendence after being paired with a brand new motherboard - the FPS is nearly twice as high as the current Intel's strongest Pentium 100MHz.

——Yes, in order to take care of the experience of Socket     7 users, the Sole CPU did not use a new interface, but continued to choose Socket     7 compatible, so that it can run on motherboards that support Pentium model CPUs.

But if it really needs to run perfectly, it still needs to be matched with a chipset designed by Meijie and Yuanxin.

"Claude, something is wrong."

Just as Claude pressed the dial key, several computers on the side also completed a series of multiplication and division calculation tests.

This is a test program sent by Yuanxin, which mainly uses a large amount of scientific calculations to simulate actual engineering, mathematics and related industry operations scenarios.

All test results will eventually yield an accurate result.

"Why is the result of this computer wrong?"

An engineer in charge of the test pointed to the 66MHz first-generation Pentium and looked at the test results in surprise - now, the screens of other monitors have given the test time and final result - that is a Pi that is accurate to the fifth digit after the decimal point. 3.14159

But the first computer came up with a completely different number.-2021.211

"Try again?" Claude immediately put away his phone, and he walked to the computer curiously.

After repeated tests, the answer remains the same.

Several people looked at each other, and Claude even forgot to call Su Yuanshan to report the good news.

After pondering for a few seconds, he whispered: "The test software is the same, and the other five platforms have the same results, so there is only one reason..."

"This test is a floating point division test, and the CPU is responsible. Obviously... there is a problem with this Pentium 66's floating point division."

Claude immediately gave the answer with the keenness of an engineer: "You can find a few more CPUs and find the first generation of Pentium series."

...

The test engineer quickly found several Pentium CPUs that were launched last year from the test laboratory, ranging from 33 to 66. In addition, he also found the second generation Pentium with another main frequency.

To be on the safe side, everyone ran a test on the entire platform and quickly concluded that the answer to Yuanxin, a floating point division test tool, is unique and unchanging.

——The answer to the Pentium 66 host has not changed, it is still -2021.211, which seems so unrestrained.

As the platform was changed, everyone became nervous.

All the engineers who play CPUs are present. They all know that if there is a calculation error in the CPU, it is unlikely that a certain piece will have problems - because if there is a problem with a single piece of CPU, the most likely one is that it will cause a crash during the test or the results will be constantly changed.

The result of this bug can be repeated, which means it may be a design defect or a design error.

Time passed by second by second. When the replaced platform gave the result again, everyone looked at each other... A feeling of not knowing what to say appeared on their faces.

All the first generation of Pentium were wrong!

"Fuck...will Intel make such a mistake?" Hopster stared, hiding the surprise in his eyes.

As a veteran who has been following Intel's pace since the Cyrix era, Hopster knows very well how powerful design and manufacturing capabilities this giant they "temporarily defeated" today have in the CPU field.

"Maybe." Claude calmed down quickly. He thought of many possibilities at this moment, but in the end a ray of light fell into his mind - perhaps, this was a problem with the division table.

In current FPU units, in order to improve the speed of floating-point division operations, all chip designers, including cyrix, use the solution to burn the division table in the core, which is obvious.

But it will also bring trouble because it is "artificial" and not automatically generated through EDA. Then it cannot be simulated in various simulation programs.

So, it is possible to go wrong.

Could it be that Intel made a mistake in the division table during the design process?

Thinking of this, Claude quickly grabbed the phone and began to dial Su Yuanshan's cell phone.

At this moment, the good news is already secondary. It is important to figure out which calculation in Yuanxin’s scientific computing testing software triggered this loophole!
Chapter completed!