Chapter 1961 Private Supercomputer (Part 2)

The testing after the relocation of the supercomputer is completely a cover, Guo Tailai intends to test the supercomputing chip he made.

The supercomputer currently deployed by Guo Tailai's data service company is the Sugon-4000A developed by the Computing Institute of the Chinese Academy of Sciences, with an operating speed of 11 trillion operations per second. In the early stage, apart from the occasional rendering of the MT30 assembly demonstration animation in the spare time, most of the other computing power is in the implementation of the force analysis and wind tunnel analysis of the body design plan of the supersonic airliner requested by Guo Tailai, which has basically been exhausted. full of.

Taking advantage of the opportunity of deploying the underwater computer room this time, Dawning-4000A will also carry out some upgrades. Although it is not up to 5000A, it is also a small version higher than 4000A. more than seven times.

Of course, Guo Tailai will definitely bring private goods, and put a batch of servers he made into it at the same time. Once the underwater computer room is submerged, it will only be opened for maintenance after a long time, and Guo Tailai's server will be relatively safe. It is easy to do as long as a separate line is used. Even Guo Tailai can keep some nanorobots in the computer room for remote control through satellite lines, which is very convenient.

The Dawning supercomputer will definitely be delivered after testing. Basically, there is no need to worry, but Guo Tailai’s own supercomputer definitely needs to be tested. It just happened to be used to make a rendering for "Avatar" to test the usage status.

Guo Tailai said so, and it was a supercomputer rendering, how could James disagree? The post-rendering of Avatar used Weta Digital's six supercomputers, each running at about 200 trillion times per second, and the rendering work lasted for a full year. With an extra supercomputer, it can be completed a period of time earlier, and it doesn't do him any harm. And the boss's own super calculation, confidentiality and priority will definitely be taken care of, so there is no need to worry at all.

"Okay!" James agreed without hesitation.

There is no need to worry about negotiating with Weta Digital. Anyway, Guo Tai said that it is only a small part of small-scale rendering, which is used to test the operation of the supercomputing, and not much will be allocated.

Guo Tailai needs a supercomputer. Not only does the design company need to use supercomputers for wind tunnel analysis and body force analysis, etc., he also needs a supercomputer to design the aircraft he wants. It's not suitable for everyone to know about that monster plane in the early stage, so let's figure it out by ourselves!

During this dive, whether it was Guo Tailai, Chu Fei, or James, they all relaxed physically and mentally, which was far more satisfying than an ordinary vacation. The three of them quickly threw themselves into their work with vigor.

James continued to go back to the post-production of Avatar, and Chu Fei also began to prepare for the global promotion of Transformers, but this time the promotion work is mainly concentrated in the United States and China, and Guo Tailai will be able to accompany Chu Fei to attend the premiere.

Guo Tailai's work is naturally the same as before, but there is a new job added, processing the underwater computer room and installing and debugging his own server.

The twelve-inch silicon crystal cylinder purchased before can be fully used now. The chip he wants to make has already been designed, just process it.

The design drawing of this super chip carried in the control system is a mature giant chip in the future, with 1.2 trillion transistors, and only one chip can be made on a 12-inch wafer. The largest chip in history, it is specially designed for AI tasks and supercomputing.

This giant chip is the final mature version of WSE (Wafer Scale Engine) produced by Cerebras, with an area of ​​42,225 square millimeters, 1.2 trillion transistors, 400,000 cores, 18G bytes of on-chip memory, and a memory bandwidth of 19PByte/s , fabric bandwidth 100Pbit/s.

The 400,000 cores are connected together through a fine-grained, full-hardware, on-chip network, providing a total bandwidth of 100Pb (100x1000Tb) per second. Provides supercomputer-class computing power, high-performance memory and bandwidth in a single wafer-level solution. It can be said that a piece of WSE chip is a supercomputer, and it will be a supercomputer in the next fifteen years, and it also has its own AI learning ability.

Guo Tailai produced 10,000 pieces of such chips, and integrated them according to the integrated system. It can be said that if successful, this will be the most powerful supercomputer in the next ten years, bar none. The overall performance is millions of times stronger than the Sugon 4000A currently in use.

The post-rendering work of Avatar took Weta Digital’s six supercomputers for a whole year, and it may take less than an hour to put it in front of this new supercomputer. It's a pity that Guo Tailai's powerful computing power is to ensure the perfection of the monster plane he made, not for the rendering of special effects in film and television. Guo Tailai has no intention of revealing this artifact to outsiders, so let's master it first.

The biggest problem with WSE is that the tape-out success rate is too low. A 12-inch wafer can only make one chip. If there is a slight mistake, this chip will be scrapped.

But here in Guo Tailai, the problem is not big, although there will be occasional mistakes, but the efficiency is much higher than that of photolithography. The same is true for subsequent packaging, which has led to the fact that, except for Guo Tailai, almost no one can do this at present, and Intel, the leader in the chip industry, can't do it either.

Another trouble is that the heat dissipation is relatively high, and liquid cooling must be used. Fortunately, Guo Tailai is currently using the underwater computer room directly. The entire computer room will be submerged in water, and the cooling water of a Dushu Lake will be circulated to ensure that a large amount of operating heat can be taken away in time.

Guo Tailai's processing workshop was kept secret from the outside world, and no one knew what he was doing inside. Especially after the two chiefs had dinner, they each sent people to strengthen the security level there. Except for Guo Tailai and a few people around him, no one could enter the processing workshop at all. In this case, Guo Tailai processed a large number of chips without pressure at all.

It was already mid-May when I came back, and by mid-June, Guo Tailai had finished processing all the supercomputers and underwater computer rooms. The underwater project next to the hotel has also been completed, and it is only waiting for the supercomputer to be relocated and then entered the water.

It took a week for the data company's supercomputer to be installed. After the power-on test was successful, the underwater computer room was filled with water and sealed, and it was successfully placed underwater. The water-cooling cycle starts, and the Sugon supercomputer starts up successfully and starts running.

A small part of James' rendering work was also submitted to Guo Tailai through the satellite network. Guo Tailai activated the WSE supercomputer, installed the rendering software, and began to calculate. Ten seconds later, the rendering of the two-minute special effects shot was completed. I saw the picture after the successful rendering, perfect!

The good news came one by one. After more than two years of construction, the Iron Man Building was completely completed and handed over. Next, there will definitely be work such as investment promotion, but some scenes from the "Iron Man" crew can come here for actual shooting, which is great news.

On June 22, the U.S. stock market retreated from its high level, and the stock market plummeted. The Dow fell 2.37%, the S&P fell 2.29%, and the Nasdaq fell 2.07%. , making a fortune in a muffled voice. But the drop this time is obviously a bit too big.