When Verizon set out to build its 5G network five years ago, the company knew it would gain a deeper understanding of what this next-generation technology would mean for the computing, augmented and virtual reality, gaming and telco industries.
What they found is that the combination of 5G and GPU-based edge computing brings a lot more than speed and performance to the table. It enables a whole new computing reality.
“The network is so fast and offers so much bandwidth, we can really offer a high-fidelity experience for media, gaming, IoT and autonomous vehicles,” said Raheel Khalid, chief engineer at Verizon.
Verizon quickly learned that 5G will deliver multiple computing capabilities, including gigabit speeds with latencies under 20 milliseconds. That led Khalid and his team to start experimenting with NVIDIA GPUs with the goal of beefing up Verizon’s high performance computing operations.
Fast-forward to today, and the team is combining powerful GPUs from NVIDIA with 5G to create a distributed data center.
Khalid knew that the closer Verizon could move GPU-based processing to devices, the more advanced services its network could support. These improvements will also enable devices to become thinner, lighter and more battery efficient, opening the door to the kind of memory-intensive parallel processing that enables rendering, deep learning or computer vision.
“Computing power goes up, and all the limiting factors start melting away,” Khalid said.
Scaling GPUs
At the last few GPU Technology Conferences, Verizon has demonstrated a variety of APIs that are only possible when combined with GPUs and high-speed networks. Examples include ray tracing, rendering and real-time transcoding. The company has also been branching into neural-based applications.
But the real focus has been on scaling to squeeze more out of its GPUs, which had previously only been able to support between one and three mobile users. With refinements to its architecture, Verizon is now supporting between 16 and 64 computer vision users on a single GPU, or as many as 1,024 rendering and ray-tracing users.
NVIDIA GPUs are being embedded throughout Verizon’s network — at its mega data centers, at the hundreds of smaller ones that those feed, and at the thousands of smaller cell sites supported by those.
The idea is to spread HPC capabilities in the most effective manner possible, using every site, guaranteeing the best performance and moving from cell site to cell site as needed.
“Users accessing the edge service get the best possible experience without interruption or degradation of quality,” said Khalid. “We can create special lanes on the network to ensure flawless performance of application.”
Find out how Verizon is pairing 5G with an NVIDIA Quadro RTX server on the network edge to create dynamic real-world applications for AR and VR by attending a webinar on Tuesday, Sept. 10. Register here.