Unleashing the Power of GPUs: Insights from the 2018 GPU Technology Summit
Graphical Processing Units (GPUs) have transcended their original purpose, becoming integral to a myriad of high-performance applications such as Artificial Intelligence (AI), real-time media rendering, and big data analysis. This evolution has created escalating demands for both digital storage solutions and memory hardware, necessitating advanced management software to facilitate these burgeoning applications. The 2018 GPU Technology Summit showcased innovative hardware and software that emphasize how specialized storage and memory can synergize with high-speed processing to drive these applications forward.
The Impact of Moore’s Law
As Moore’s Law—predicting the doubling of transistors on microchips approximately every two years—begins to plateau, the GPU emerges as a means of continuing the quest for increased processing speeds. In contrast to traditional CPUs, which are reaching their limits in terms of density and efficiency, GPUs possess the architecture and parallel processing abilities necessary for tackling increasingly complex computational tasks.
Real-Time Video Rendering Breakthroughs
One of the standout applications of GPUs highlighted during the Summit was the advancement in real-time video rendering through ray tracing technologies. Lasering in on gaming and entertainment, NVIDIA’s CEO Jensen Huang demonstrated real-time, 4K resolution content flowing at 60 frames per second (fps). This was made possible through innovative techniques like prebaked lighting for static objects, which minimizes the processing required for dynamic scenes.
The ray tracing technology also supported intricate optical effects known as caustics, allowing for realistic light behavior with complex geometries. Notably, a real-time Star Wars rendering utilized deep learning for predictive ray tracing, showcasing the transformative potential of combining AI with GPU capabilities. By harnessing the power of the Quadro GV100, the industry could witness a monumental reduction in video rendering farm sizes—from 168 kW to just 24 kW—along with a striking seven-fold decrease in equipment footprint.
The Behemoth: NVIDIA DGX-2
The Summit spotlighted NVIDIA’s DGX-2, a monumental GPU computing platform delivering an impressive 2 petaFLOPS in operational speed. Sporting 512 GB of HBM2 memory and equipped with 30 TB of NVMe SSDs, this 250-pound powerhouse symbolizes the apex of GPU computing capabilities. A captivating application showcased by Disney Imagineering demonstrated real-time video for an immersive 4K theme park ride, setting the stage for unprecedented experiences in media and entertainment.
Real-time rendering is rapidly becoming a staple not only in games and rides but also holds the potential to revolutionize video editing and content distribution, making high-quality graphics more accessible and versatile.
Innovations in Digital Storage and Memory
At the heart of the GPU revolution lies significant advancements in digital storage and memory, showcased by a diverse array of companies at the Summit, including Pure Storage, Liqid, Nyriad, and WekaIO. These organizations epitomize the fusion of memory technology with GPU processing to maximize efficiency and performance.
Pure Storage unveiled their AIRI system, an AI-ready platform that integrates their FlashBlade technology with four NVIDIA DGX-1 supercomputers, capable of delivering up to 4 petaFLOPS of performance. This seamless integration is poised to cater to organizations aiming to conduct AI development on-premise, mitigating dependency on cloud services.
Liqid introduced a novel disaggregated network, compute (GPU included), and storage solution that operates at a bare-metal level. With a PCIe-switch fabric enabling dynamic resource allocation, Liqid empowers businesses to efficiently configure their systems. Their innovative approach eliminates unnecessary virtualization layers, simplifying resource management and reducing associated costs.
Storage Innovations
Nyriad showcased its NSULATE technology, developed initially for the Square Kilometer Array. This advanced storage solution integrates parallel computing with distributed storage, amplifying computing density. Demonstrating its prowess, Nyriad showcased its software running on a combination of NVIDIA Tesla and Netlist NVDIMMS, emphasizing the collaborative potential of GPUs and cutting-edge storage.
WekaIO presented their WekaIO Matrix software designed for performance-intensive tasks, reaching notable throughputs of up to 5GB/s with NVIDIA TensorRT inference optimization over Mellanox’s InfiniBand network. Their collaboration shows promise for businesses requiring high-performance infrastructure, particularly in demanding AI applications.
Conclusion
The advancements showcased at the 2018 GPU Technology Summit illustrate a vibrant intersection of GPU technology, advanced memory solutions, and storage capabilities. As industries pivot towards harnessing these innovations, the potential for enhanced processing in AI, rendering, and big data analysis is undeniably significant, paving the way for impressive developments in technology and media.