Advanced Computing in the Age of AI | Thursday, March 28, 2024

AMD GPU Helps Map Universe 

Canadian researchers scanning the universe to develop the most detailed map of our three dimensions ever observed are using AMD's GPU accelerator to crunch the celestial data.

Based on internal benchmark testing, the chipmaker (NASDAQ:AMD) bills its FirePro S9300 x2 Server GPU as the world’s fastest single-precision GPU accelerator. It claims 13.9 teraflops of "peak single-precision" compute performance, according to internal testing earlier in March. AMD said its calculation was based on a formula using a combination of clock speeds and shader units.

(AMD cited recent performance testing by Nvidia (NASDAQ: NVDA) that pegs the GPU rival's fastest Tesla board, the Tesla K80, as delivering up to 8.73 teraflops of single-precision compute performance.)

AMD, Sunnyvale, Calif., is targeting its server GPU based on its third generation graphics core architecture primarily at datacenters, adding that it comes with 1 Tb/sec of memory bandwidth. Along with Nvidia's, AMD also claimed its server GPU outpaced by a wide margin Intel Corp.'s (NASDAQ: INTC) high-end Xeon processor, the E5-2699 v.3, in terms of Gb/sec of memory bandwidth.

Canadian researchers will show whether those performance claims hold up as they use the AMD server GPU to create a detailed 3D map of the largest volume of the universe so far observed. The Canadian Hydrogen Intensity Mapping Experiment, or CHIME, uses four 100-meter-long cylindrical reflectors rather than traditional dish-shaped telescopes to gather signals. The reflectors cover an area larger than the area of five hockey rinks.

CHIME's reflector array.

The signals gathered by the mammoth reflectors are then subjected to computational analysis using an AMD GPU cluster to map the universe in three dimensions.

The CHIME array collects radio waves thought to have been travelling through time and space for up to 11 billion years. The radio wave data are fed into a supercomputer running a cluster of FirePro GPUs for intensive number crunching designed to map expansion of the universe.

The CHIME project was launched to investigate the discovery that expansion of the universe is accelerating rather than slowing down. The result will be an extremely detailed map showing hydrogen radiation intensity across billions of galaxies. The map would aid researchers seeking to understand the mechanism behind accelerated expansion of the universe.

Advances in GPU performance along with chip scaling that have reduced their overall cost are making the chip technology available to researchers who require massive amounts of computing power to visualize the universe. "We now have the computational power, bandwidth and efficiency to make this data analysis possible to study the impact of dark energy on our universe," Keith Vanderlinde, of the Dunlap Institute for Astronomy & Astrophysics at the University of Toronto, noted in a statement provided by AMD.

(NASA calculates that about 68 percent of the universe is made up of dark energy, which affects the expansion of the universe.)

In support of CHIME and other HPC initiatives, AMD said it developed a computing platform based on open source parallel processing. The Linux-focused platform was optimized for HPC and ultra-scale computing applications. The compute kernel provides direct access to graphics hardware.

AMD is aiming its FirePro graphics GPU at more academic research projects like CHIME.

About the author: George Leopold

George Leopold has written about science and technology for more than 30 years, focusing on electronics and aerospace technology. He previously served as executive editor of Electronic Engineering Times. Leopold is the author of "Calculated Risk: The Supersonic Life and Times of Gus Grissom" (Purdue University Press, 2016).

EnterpriseAI