Due to more and more US sanctions, China cannot receive the latest AI accelerators and much more, which is also related to artificial intelligence. However, now the US government has concerns that China will still gain access to such technologies, and not just like that, right away to the most productive supercomputer for AI training.

We are talking about the Condor Galaxy system, which in itself is unique and extremely unusual, but we have already written about this. This system is being built by G42 from the UAE. The company is reportedly controlled by Sheikh Tahnoun bin Zayed and has ties to Chinese firms that the US considers a security threat. Including with Huawei. According to the New York Times, the CIA fears that G42’s cooperation with Chinese companies could lead to G42 becoming a conduit for the transfer of classified American technology and genetic data. In addition, China will have access to an incredibly powerful supercomputer.

The bottom line is that Condor Galaxy will consist of several identical installations (CG1, CG-2, and so on), located in different countries. There will be nine of them in total, and only a few will be located in the United States. The CIA expresses concern about four, but does not specify where the four systems will be located.

Let us remind you that Condor Galaxy is a unique solution. Each of the CG clusters consists of 64 modules, with each module actually containing only one chip. We are talking about the Cerebras Wafer Scale Engine 2 (WSE-2) chip, which is the size of an iPad and contains 850,000 cores. As a result, only one CG system has a performance of 4 ExaFLOPS (FP16), and the total performance of the entire supercomputer will be an incredible 36 ExaFLOPS. This is very, very much, however, just today Nvidia and Amazon announced the construction of a Ceiba supercomputer with a performance of 65 ExaFLOPS in the same mode.

As you know, from the moment it became known that supplies of GeForce RTX 4090 video cards to China would be prohibited, adapters have risen sharply in price in China itself, but prices have also increased in the USA and Europe. Now those responsible have been found. At least partly.

There are reportedly unscrupulous sellers who do not sell their RTX 4090 in their countries, but stockpile them for resale to China. Of course, because of the double or more markups, you can make very good money on this. Of course, such actions lead to a shortage of adapters in a particular market, increased demand and, as a result, higher prices. At least this is what sellers in Vietnam do, but probably not only there.

Prices in the USA and Europe have recently risen to 2000 dollars/euro for the most affordable options, and are not going to go down yet.

Let us remind you that Chinese craftsmen began to convert the GeForce RTX 4090 into accelerators for AI, and now they have even created the RTX 3080 20GB and Radeon RX 580 16GB.

Microsoft’s Copilot assistant, integrated into Windows 11, will soon move to the GPT-4 Turbo language model.

General Director of Advertising and Web Services at Microsoft, Mikhail Parakhin, said that the company still needs to correct some shortcomings. However, no time frame for the transition has been announced yet.

As a reminder, GPT-4 Turbo features a large number of improvements relative to GPT-4, including a context window for 128,000 tokens. True, with Copilot it may well be limited.

In addition, this language model is trained on data up to April 2023, which gives it an advantage over older versions.