China plans to use low-cost NVIDIA chips in its hypersonic weapons. The US will not ignore it

  • The Jetson TX2i module is a low-cost platform designed for industrial applications
  • The US Department of Commerce allows the sale in China of chips that do not exceed 70 TFLOPS FP32

The SoC of this module incorporates two 64-bit Denver cores, two ARM Cortex-A57 cores and a Pascal GPU with 256 CUDA cores. A priori it is a relatively modest configuration. In fact, according to NVIDIA itself, this module delivers a maximum performance of 1.26 TFLOPS FP32, so it is not subject to the sanctions approved by the US. At the beginning of this month, the US Administration reviewed the requirements that semiconductors produced by US companies must meet so that they can be sold in China. And NVIDIA can sell its Jetson TX2i modules.

It can do so because the US Department of Commerce approves the sale in China of those semiconductors whose performance does not exceed 70 TFLOPS FP32. As we have just seen, the NVIDIA module to which we are dedicating this article has a much lower performance, so presumably its marketing in China is allowed. However, the fact that it is a relatively modest platform compared to NVIDIA’s most powerful GPUs does not mean that it cannot be useful in some usage scenarios.

The US is going to react, but it no longer has room for maneuver in this area

Chinese researchers have discovered that the Jetson TX2i module, which, interestingly, can be easily purchased on the Internet, can very efficiently process the fluid dynamics models that allow hypersonic missiles to be managed in real time. In fact, according to SCMP this hardware has cut the latency of other platforms designed to deal with this usage scenario from a few seconds to just 25 milliseconds. To make matters worse, these scientists have published a guide that indicates step by step how this NVIDIA module can be integrated into a hypersonic weapon.

Those responsible for this project are technicians from the Beijing Research Institute and the Dalian University of Technology. According to them, this NVIDIA hardware is perfect for real-time optimization of the fuel delivery, diagnostics and fault-tolerant control systems of supersonic combustion engines. The result of their research was peer-reviewed and published on March 13 in the journal ‘Propulsion Technology’ of the Chinese Academy of Sciences.

Given the current situation of tension between China and the US, we can be sure that the Government led by Joe Biden is not going to sit idly by. It is reasonable to assume that it will once again modify the requirements that semiconductors produced by American companies must meet to prevent this NVIDIA module or other similar ones from being marketed in China. The problem is that the effectiveness of this measure is debatable. And it is because in all likelihood Chinese chip designers and manufacturers, such as Huawei, SMIC or Hua Hong Semiconductor, will be able to replicate and produce the Jetson TX2i module without too much effort given its relative simplicity.