Russian and Chinese Scientists Do Incredible Things with Nvidia GPUs

Restrictions on supplying advanced Nvidia chips to Russia and China have prompted scientists there to look for ways to optimize computing performance on Nvidia's non-sanctioned products, especially on popular GPUs.

The more it is banned, the more extraordinary things Russian and Chinese scientists create.

Thanks to reverse engineering and in-depth research into Nvidia GPU architecture, a research team from MSU-BIT University in Shenzhen, China, has developed an algorithm that can increase computing speed up to 800 times after optimization.

Founded in 2014, the research team is a collaboration between MV Lomonosov Moscow State University and Beijing Polytechnic Institute. They reverse-engineered the Nvidia architecture supporting the CUDA environment, and the results of the research were published in detail in a paper in the Chinese Journal of Computational Mechanics.

'This computational efficiency allows researchers to reduce computation times from days to hours or even minutes with just a typical home GPU. This is a significant step forward in PD (peridynamics) research,' the paper's authors say.

Research increases the computing power of conventional GPUs from Nvidia by up to 800 times.

Based on the improved algorithm, the team developed the PD-General framework for calculations in the field of exodynamics, focusing on evaluating materials with non-uniform deformation, especially cracks. This opens up the opportunity to accelerate calculations by up to 800 times in the field of materials science, an important factor for the aerospace industry.

Previously, Chinese scientists pointed out the superior capabilities of Nvidia graphics chips, but their potential was not fully exploited due to US restrictions. Scientists believe that this could limit the advances that humanity could have exploited in the future.

Marvin Fry

Update 10 February 2025