Chapter 72: Supercomputer
With the necessary technology accumulated and validated, Li Qingsong immediately began construction on a new chip factory.
This factory was much larger than its predecessors. The entire industrial park, including all supporting facilities, covered a massive 10 square kilometers.
This wasn’t just because the factory was designed for a higher production capacity—an annual output of approximately 20 million 45-nanometer chips of various specifications and models. It was also because the 45-nanometer process was far more complex than the previous 800-nanometer one, involving more steps and demanding stricter environmental controls.
From the initial purification of silicon dioxide to the final packaging and shipping of the chips, the entire process involved well over 1,000 steps!
In total, there were several hundred other factories dedicated to supplying it with materials, power, and other auxiliary services.
In the past, building such a massive production base would have been a major undertaking even for Li Qingsong, requiring several years of effort.
But now, with his maximum number of consciousness connections soaring to over five million, the entire chip-making complex was fully constructed in a mere six months, with all equipment installed and calibrated.
All the auxiliary factories were completed as well, and the complex immediately began production.
Meanwhile, the older chip factories with their less advanced processes didn’t shut down; they continued to operate at full capacity.
The reason was simple: while the 45-nanometer chips were the most advanced and certainly excellent, they weren’t needed for every application.
For instance, the chip controlling an excavator’s hydraulics could be satisfied with a millimeter-scale process chip. Why use a 45-nanometer one?
It would be an absurd waste.
In fact, across Li Qingsong’s entire cluster of bases, it was the chips with less advanced, lower-end processes that were used in the greatest quantities.
With factories producing chips of various process nodes simultaneously, the launch of this newest plant pushed Li Qingsong’s total chip production capacity past the 500 million mark, reaching an annual output of 520 million units.
Even before the first batch of 45-nanometer chips came off the production line, Li Qingsong had already started on another key project.
A supercomputer!
Supercomputers utilize a vast number of chips for parallel processing, enabling performance that far exceeds ordinary computers. They are indispensable infrastructure for nearly every field, including industrial development and scientific research.
The AI technology Li Qingsong planned to develop would be impossible without one.
Li Qingsong already had a supercomputer. It used about 20,000 800-nanometer chips and had a floating-point performance of around one hundred billion operations per second.
That level of computing power had been sufficient before, but it was no longer enough.
This time, Li Qingsong planned to build a supercomputer using approximately 130,000 chips, capable of 400 trillion floating-point operations per second!
It wasn’t that Li Qingsong didn’t want to build an even larger one. The reality was that increasing a supercomputer’s performance wasn’t as simple as just crudely piling on more chips.
Far more important than the number of chips was the machine’s parallel processing capability—that is, designing an algorithm and its corresponding hardware to efficiently distribute computational tasks among all the chips.
Getting them to work in concert was the most crucial part.
Beyond that, factors like memory bandwidth and capacity, storage system read/write speeds, and data transfer latency all acted as constraints on the supercomputer’s potential power.
Without breakthroughs in these other areas, simply adding more chips could sometimes paradoxically decrease the supercomputer’s overall performance.
Li Qingsong assigned several thousand Clones to run massive optimizations before they finally engineered this Supercomputing System.
The supercomputer center was now complete, waiting only for the chips to arrive.
After just a few days, the specially designed chips for the supercomputer were finally ready.
Once the chips were transported to the supercomputer center, Li Qingsong had his Clones rapidly begin the installation.
Li Qingsong divided the roughly 130,000 chips into 65,536 nodes. Each node was equipped with two chips, interconnected with specially developed fiber optics, and the system’s total memory reached 64 terabytes.
A dedicated power station with an installed capacity of 15,000 kilowatts, specially built by Li Qingsong, supplied its electricity.
The machine’s power consumption was terrifying. It devoured 15,000 kilowatt-hours of electricity in a single hour, amounting to 130 million kilowatt-hours a year.
With such terrifying power consumption came equally terrifying heat generation.
To handle this, Li Qingsong designed a custom water-cooling system. A vast network of pipes, totaling over a thousand kilometers in length, was laid out around the supercomputer center, covering an area of more than ten square kilometers.
Once installation was complete, at a command from Li Qingsong, the machine finally roared to life.
Instantly, countless calculations began firing within every single chip. Heat was generated with every state change in the most fundamental circuit components.
The heat from a single chip was minuscule. But the combined heat from over one hundred thousand of them was terrifying.
In the cooling pipes, the frigid water began to heat up instantly and was circulated outward.
The surface of Ceres stays at a perennial low of over 250 degrees below zero Celsius. Now, the heat from the supercomputer was being dispersed into the planetoid’s crust through the network of buried pipes.
Once cooled, the water flowed back to the supercomputer center to absorb more heat and begin the cycle anew.
This constant cycle allowed the supercomputer to operate within its optimal temperature range.
As the supercomputer began to output heat, sensors Li Qingsong had placed nearby detected a slow rise in the temperature of this region of Ceres. Some frozen gases with low melting points even began to thaw and diffuse into interstellar space.
Observed through an infrared telescope, this patch of ground would even appear noticeably darker than the surrounding regions.
If Ceres had an ecosystem, Li Qingsong could say that this single supercomputer was capable of altering the local ecology all on its own.
With testing complete and all systems performing as expected, Li Qingsong immediately uploaded the learning algorithms—written by a thousand of his programmer Clones—to the supercomputer.
Then, data began to pour in. It flowed from the chips inside every factory, power plant, research base, cultivation center, and even livestock farm. It streamed from every truck, train, tractor, excavator, spaceship, and satellite. Data from virtually every chip in his entire operation began to converge on the supercomputer.
This colossal torrent of data was the nourishment for the learning algorithm.
Fed on this high-quality data, the learning algorithm continuously iterated and optimized itself, rapidly maturing.