laitimes

Break through the traditional "supercomputing" to accelerate the development of new drugs

In recent years, traditional structure-based design methods have gradually reached their limits, resulting in exponential growth in capital expenditures for drug research and discovery. Thus, motion-based drug design solutions, known as molecular dynamics simulations, are among the most exciting and innovative advances in the industry for the field of drug design. The realization of molecular dynamics simulations has extremely high requirements on computing power.

In the past, only large supercomputers had the ability to provide sufficient computing power for molecular dynamics calculations. But Snow Lake Technology, which is committed to providing "dedicated supercomputing" solutions, hopes to break this bottleneck and develop a solution that is higher in performance, lower cost and more energy efficient than traditional supercomputers.

Dedicated engine for molecular dynamics based on FPGAs

As mentioned above, in order to promote the development of medicine, drug development has advanced to the molecular level. At this level, scientists are able to gain a deeper understanding of the role of molecules and biological processes, unraveling the mysteries of their movements.

Dr. Sheng Nan, Chief Scientist of Life Science Computing at Snow Lake Technology, said: "Just as it is far easier to capture the details of the crime than 'watching the surveillance video' when solving a case than 'watching the scene photos', molecular dynamics simulation allows drug designers to more directly 'see' the entire effective process of drug molecules, thereby greatly improving the success rate of new drug research and development."

Founded in 2017, Shanghai Xuehu Technology Co., Ltd. is an innovative enterprise dedicated to the research and development and solutions of high-performance computing products. The company provides innovative, dedicated, high-performance computing solutions and products for life sciences, oil and gas exploration, autonomous driving, distributed storage, and other related fields.

The "Yaddle MD" designed and developed by Snow Lake Technology is a dedicated engine for molecular dynamics based on FPGAs. It is a proprietary solution for molecular dynamics computing implemented on the Xilinx Alveo Accelerator Card that provides compatibility support for commonly used molecular dynamics software and other analysis plug-ins through a molecular dynamics-specific API with the help of the Xilinx Runtime Library (XRT).

Break through the traditional "supercomputing" to accelerate the development of new drugs

Molecular dynamics calculations used in drug research and discovery processes

In terms of performance, Snow Lake Technology's Yaddle MD FPGA molecular dynamics dedicated engine runs on the Alveo accelerator card, which is nearly 50 times higher than the CPU-only molecular dynamics software solution and twice as high as the GPU-accelerated molecular dynamics software solution, as shown in the figure. At the same time, the power consumption of FPGAs is only 2/3 of that of the CPU and 2/9 of the GPU, so that the Yaddle MD can provide nearly 100 times the analog per kilowatt-hour of power than the CPU and nearly 10 times that of the GPU.

Break through the traditional "supercomputing" to accelerate the development of new drugs

Snow Lake Technologies' Yaddle MD FPGA molecular dynamics engine delivers higher performance and lower energy consumption than competing solutions

The value of adaptive Alveo accelerator cards

Because complex algorithms have evolved far beyond the design cycle of chips, fixed-function GPUs and ASIC devices cannot keep pace. Built on Xilinx's 16nm UltraScale architecture, the Xilinx Alvio Accelerator Card is designed to meet the needs of the rapidly changing modern data center and delivers up to 90x performance than CPUs for common workloads such as machine learning inference, video transcoding, database search, and analytics.

Break through the traditional "supercomputing" to accelerate the development of new drugs

Alveo U200 accelerator card

The Alveo Accelerator Card provides a unique reconfigurable acceleration capability for building Yaddle MDs, enabling continuous optimization of its flexible adaptation algorithms while supporting any type of workload while reducing total cost of ownership.

Yaddle MD uses the Alveo U200 accelerator card to implement ultra-high-performance FPGA molecular dynamics algorithms capable of performing very complex calculations. Based on a single Alveo card, yaddle MD can achieve higher performance than using CPU and GPU processors, providing powerful computing power that was previously only available to supercomputers, which can greatly accelerate the kinetic simulation of biomolecules.

With Yaddle MD, users can:

Complete molecular dynamics calculations are implemented on monolithic FPGAs, freeing up large amounts of CPU resources

Realize distributed pipeline computation of inter-atomic interactions, which significantly improves computing performance

Encodes bonding information between atoms and decodes it during real-time computation

Provides compatibility support for commonly used molecular dynamics software and other analytical tools, including format conversion, through toolsets

Dr. Sheng Nan said: "High performance is the key to ensuring the efficiency and accuracy of molecular dynamics calculations, and Snow Lake Technology's high-performance molecular dynamics special engine based on Alveo can not only save customers a lot of time and cost, but also make kinetics-based drug design from impossible to possible."

At present, Snow Lake Technology's Yaddle MD engine has begun to provide MD acceleration services through Alibaba Cloud, and its Alveo card-based acceleration solution is also being developed and tested on a small scale. At the same time, the pace of innovation of Snow Lake Technology continues, and its plan will continue to expand simulation capabilities in the form of a combination of multiple Alveo devices in the future, further accelerating molecular dynamics computing applications and expanding throughput.

(7856913)

Read on