Two days later, the server that Duke commissioned to buy from Cusai arrived as scheduled, and Duke received the arrival notice from Cusai and drove to the company.

"Boss, look, this is the computer you want, see if it's the one you want to buy." Cusai waited in the company, saw Duke coming, and quickly brought Duke to the server.

Duke turned on the computer, entered the BIOS and checked the configuration, and it was indeed two AMD 20-core Terramar processors.

"That's it, thank you, Cusay, go about your business" Duke saw that it was indeed what he wanted, so he sent Cusai away.

After entering the system, Duke took out a USB wireless network card that he had with him and connected it to the server, and now Kerry can only access the network wirelessly, and Duke carries a wireless USB network card with him.

After quickly installing the driver of the wireless network card, Duke started Kerry to plug it in, and soon, Kerry combined with the AMD Terramar processor instruction set obtained in advance, after an unknown number of computing tests, finally, after about 2 hours, "the new CPU system hardware virtual completion." Duke's mind suddenly shook, and Kerry gave the message of virtual success.

The Terramar processor is fully equipped with the original performance and full instruction set of the Terramar processor, but the internal implementation is different. Kerry then used this as the core to virtualize a new independent system with dual CPUs, which could directly run the Windows 2012 version and the common Red Hat Linux system on this virtual hardware.

Seeing that Kerry completed the cloning of a set of dual-socket CPU server hardware system, Duke downloaded the latest version of the CPU test application, and started the test on Kerry's internal virtual system and the actual server around him at the same time, and the results showed that Kerry's simulated Terramar system showed exactly the same test results as the actual hardware Terramar system, proving that Kerry's cloning was very successful.

Duke then instructed Kerry to complete the simulation of the 4-way, 8-way, 32-way, 64-way, and 128-way servers, and after this actual simulation test, Kerry fully mastered the hardware simulation of the multi-channel server, and on this basis, Duke began to instruct Kerry to complete the simulation of the Titan II supercomputer according to the CrayXT5 design architecture.

In terms of supercomputer design principles, there is no mystery in it, and Titan II is nothing more than a number of computing nodes equipped with four 20-core Terramar processors and 32GB of memory, which functionally divides the entire system into three parts: computing array, acceleration array, and service array.

Since all of this is just a series of computer nodes, Kerry soon completed the simulation of the supercomputer system, but Duke found that the Titan II simulated by Kerry seemed to be severely inferior to the real Titan II.

According to Hewitt's introduction to the performance of Titan II, its stable computing speed reached 15 petaFlops (that is, 150 million times per second), and now Kerry's software Titan II is simulated according to a simple multiplex server, but the performance is only about one-thirtieth of its performance.

What's going on?

Duke looked back at the computing unit used by Titan II, only to know that he had thought a little worse, because in the composition of Titan II, the computing array and service array are composed of computing node machines and service nodes using general-purpose processing units (CPUs) respectively, and the acceleration array is composed of a large number of acceleration nodes based on graphics acceleration processors (GPUs), which is a kind of "CPU+GPU" heterogeneous collaborative computing.

Because purely in terms of floating-point computing power, the floating-point computing power of a GPU is equivalent to dozens or even hundreds of times that of a CPU, and the memory bandwidth of a GPU can reach more than ten times that of a CPU, and the latency is lower, and the external data throughput capacity is also stronger than that of a CPU.

Therefore, when designing, for the simple floating-point arithmetic part, if a supercomputer is composed of GPUs of the same scale, then the floating-point arithmetic power of this node will be increased by as much as a hundred times. The advantage of CPU is in the calculation of more logical and complex data structures.

Considering that both nuclear reaction simulation calculations and climate estimation simulations involve a large number of floating-point operations, this "CPU+GPU" heterogeneous co-architecture is also used in the Titan II system, and a large number of TeslaK20GPUs produced by NVIDIA are used as the main floating-point computing components.

Fortunately, it is not only supercomputers that adopt this heterogeneity, but also many enterprise-class servers to choose from, and Duke told Cusay to buy one in Boston the same day.

In order to ensure that the simulated Titan II is consistent with the real Titan II in the laboratory, this time Duke went to the laboratory to get the detailed CPU and GPU composition structure information of Titan II, and this parameter is not a secret, so when Kerry completed the software simulation of TeslaK20GPU, he re-implemented the software simulation according to the Titan II architecture.

16.5petaFlops！ When the software simulation of Titan II test out of the result, Duke was excited, Kerry finally realized the reproduction of Titan II, according to the Titan II system source code version of the recompiled system, also successfully run on the Titan II system simulated by Kerry, it can be said that Kerry has now fully realized the operating environment of the nuclear reaction simulation program.

Everything is left to Dongfeng, and all that remains is how to download this app into Kree Space.

"Duke, you're here." Suddenly seeing Duke, who he hadn't seen for a few days, Hewitt was not surprised, and greeted him normally, he knew that his junior brother was a veritable genius, and now he was studying materials, computer science, and nuclear physics at the same time, and he didn't know how that head grew.

"Hello senior brother, I'll come over today to see if the permission application has been approved." Duke politely returned.

"Oh, there should be no problem, Professor Andrew has been here all this time." Hewitt said. "How's it going, thinking about starting research? Have you read the guide to using this system now? ”

"I've taken a cursory look at it, and now I have a little experimental design, and if there are no problems with permissions, I'd like to see how it looks on the machine." Duke said.

"Really? Duke, have you really designed an experiment? Hewitt was a little surprised, Duke's learning speed was terrifying, because it took him a semester to make the first experimental design that could be used on a computer. The tome guide is so thick that it takes a lot of time to read it all over.

"I don't know if it's right, I just followed the guidelines and wanted to study the operation mechanism of natural nuclear reactors, so I designed such an experiment." Duke said.

"Natural nuclear reactor research?" Hewitt was stunned for a moment, looking at Duke thoughtfully.

Of course, he knew that there was a place on the earth where a natural nuclear reactor had been formed naturally, probably in a uranium mine in Gafeng, Africa, where such a natural nuclear reactor had occurred, but how it was formed, a very small number of scientists have been studying it over the years, but due to limited conditions, there are not many results.

Unexpectedly, Duke chose this relatively unpopular topic as soon as he came. Of course, this research is very meaningful, but there are too many problems, and only the approximate mechanism of occurrence has been deduced from the results of the current research.

Since these natural nuclear reactors have completed their historical mission, there is no longer a single truly operational natural nuclear reactor on Earth, so scientists can only study with a very small number of nuclear reactor derivatives. There are too few clues that can be found. It's really not easy to achieve success on this subject.

"Duke, the direction you've chosen is really challenging." Hewitt came back to his senses and continued. After speaking, he shook his head, this little junior brother is really different. Hewitt secretly said in his heart that for the first time, Duke's academic prospects in nuclear physics were given a negative evaluation.

Since the discovery of the existence of natural nuclear reactors more than 30 years ago, so few achievements have been released over the years, and Hewitt does not believe that Duke can be a special exception.

Duke didn't understand Hewitt's mentality, among the few nuclear fission knowledge he received from Billem, how to produce nuclear fission in natural conditions is a small amount of popular science knowledge left, and there are various parameters of this occurrence environment, so for Duke, this is really just a very ordinary thing.

It's just that how to make this discovery process as natural as possible is a real challenge. There are some things that Duke must study thoroughly in theory before taking out experiments to verify them to be normal.

If the entire process of the formation of a natural nuclear reactor is displayed in its entirety, it will scare many experts and scholars in the field of nuclear physics.

With Duke's emotional intelligence now, he will never do such a stupid thing again.

Bidding farewell to Hewitt and entering his office, Duke turned on the machine in a cheerful mood, and sure enough, he found that the application process had been completed, and now he could enter the core application of the nuclear laboratory, and had the permission to use the laboratory's nuclear reaction simulation program.