AMD discuss AI Innovations, 5G/6G Integration, and India Strategy

At India Mobile Congress 2024, Dataquest had an insightful conversation with Gilles Garcia and Neeraj Varma, senior executives from AMD, about the company’s pioneering advancements in AI and 5G technologies. During the interview, they discussed how AMD's diverse AI portfolio including GPUs, CPUs, and AI-embedded solutions is positioning the company as a leader in the competitive AI market. They also shed light on AMD’s collaborations with Indian partners, its role in driving AI integration in 5G and 6G networks, and how their technology is shaping the future of quantum-safe encryption.

Excerpts:

DQ: How is AMD's AI portfolio different from others in the competitive AI market? 

Gilles Garcia: Two days ago, we hosted a major advanced AI event led by our CEO. One key message during the event was that "one size does not fit all." This means that in our AI portfolio, GPUs, CPUs, AI-embedded PCs, and other components each have their own specific roles. Every customer has unique applications, and some might require GPUs, others might need CPUs, and some could use AI-embedded PCs at the edge. Our portfolio is broad and spans the entire range from data centers to edge applications—ensuring efficiency for both inference and training models across networks. One key difference for AMD is that we don’t force customers to use GPUs at the edge where they might not be necessary or where power consumption is a concern.

Neeraj Varma: Another significant point is that we’ve made substantial investments over the past two years to unify our AI infrastructure from a software standpoint. We want to ensure that whether AI is being implemented on a GPU, CPU, FPGA, or adaptive SoC, the software ecosystem remains consistent. We are building this unified platform so that data scientists and algorithm developers can work in a common environment. Now, whether you are a large data center provider, running cloud services, or working on IoT at the edge, we offer products that meet different AI workloads.

DQ: Can you explain how AMD is integrating 5G, 6G, and AI? 

Gilles Garcia: In my group, we handle both wired and wireless networks, and we have a 6G initiative. AI is going to be closely tied to 6G, and we are already working with many customers who are testing or trialing their algorithms for 6G. These AI-driven algorithms will make networks more efficient—whether it’s for power management, user mobility, or introducing new applications. At our booth, we are showcasing two AI applications: one that demonstrates AI integration in 5G and 6G, and another that focuses on applications that can run on top of these networks.

Right now, we have the technology that fits into the far edge and can be deployed into radios or base stations. This technology uses AI engines to help customers transition to 6G by enabling new workloads and AI-based radio components. For example, baseband and radio components will be enhanced through AI to improve spectrum efficiency and interference management.

DQ: When discussing AI, people often associate it with generative AI and large language models (LLMs). How does AMD's AI portfolio go beyond that? 

Neeraj Varma: While generative AI and LLMs are popular and widely discussed, AI encompasses much more than that. AI is playing an increasingly important role in areas like IoT at the edge and network optimization. For example, AI helps make networks smarter and more efficient. AI enables network optimization and improves performance across a wide range of industries. AI has broad use cases that go far beyond generative AI and LLMs.

Gilles Garcia: AI will also be a game-changer in security. Security is all about learning and predicting, and AI allows for the creation of predictive algorithms. We have already demonstrated machine learning-based DDoS detection, where AI is used to detect and prevent large-scale server attacks. AI will continue to expand into security solutions in the future.

DQ: With the high cost and shortage of GPUs, how do you envision the future of edge AI? 

Gilles Garcia: We see three primary models for edge AI applications. The first type will require CPUs, where models and workloads can run directly on the CPU. The second type might need AI accelerators, which could be PCI Express cards or AI engines within switches and routers. The third type involves specialized use cases, such as laptops running AI applications for gaming or AI-enhanced features like Microsoft Copilot.

While GPUs are often seen as the go-to for AI workloads, their power consumption makes them less ideal for many edge use cases. In edge environments, we see limited demand for GPUs. Some edge data centers might use GPUs, but for most edge applications, CPUs, embedded AI technologies, FPGAs, SoCs, and AI-embedded PCs are more suitable and efficient.

DQ: Can you tell us about AMD's “Doctor as a Service” and its role in healthcare? 

Neeraj Varma: One of the demos we are showcasing at our booth is an AI-driven virtual ophthalmology doctor. This is part of the government's E-Sanjeevani telemedicine platform. The model runs on our latest Turin Gen 5 EPYC CPUs, deployed on a "Make in India" server built by our partner, Velankani, in Bangalore. The virtual doctor supports multiple Indian languages, and the demonstration showcases how our CPUs and software are capable of running healthcare models. The whole idea is to enable virtual consultations that can help address the shortage of doctors, especially in rural areas.

Gilles Garcia: In many countries, including India and even parts of Europe, there is a shortage of doctors. The goal of this solution is to provide people with access to an e-doctor or virtual doctor who can offer consultations, diagnose conditions, and prescribe medications. This AI-driven solution helps bring healthcare services to rural or underserved areas where access to doctors is limited. We are working with doctors to train these AI models to ensure they provide accurate diagnoses and prescriptions based on real-world medical practices.

DQ: Beyond healthcare, are there other significant collaborations AMD has in India? 

Neeraj Varma: Yes, we have several significant collaborations in India. At IMC 2024, we are showcasing several "Make in India" products through our customer partners. For instance, we have radio units from companies like Reliance Jio, Tejas Networks, and Lekha Wireless. We are also showcasing servers from companies like Netweb India and Velankani, which are built using our CPUs. One key differentiator is that some of these servers feature motherboards designed and manufactured entirely in India, not just assembled from imported components. This end-to-end manufacturing process is a major milestone for India’s tech industry.

DQ: How is AMD contributing to AI integration in 5G and 6G for telecom networks? 

Gilles Garcia: We already have AI-enabled radios in the field. For example, we’ve partnered with companies like Samsung and NEC, which are using AMD technology in their radios with integrated AI engines. These AI engines run beamforming algorithms that enhance network performance in 5G radios. Our customers are deploying 5G radios with our AI technology, and they are even exploring adding additional AI functionalities to evolve their networks toward 6G.

Our technology, known as Versal, is based on 7-nanometer technology and is already being deployed in 5G networks. Our 4th and 5th generation CPUs are also available for base stations and can handle AI workloads. So, while we are constantly working on new technologies, the technology we have today is already enabling customers to integrate AI into their 5G networks.

DQ: Are you launching any new products at IMC 2024? 

Neeraj Varma: We didn’t plan specific product launches for IMC 2024, but just two days ago, we hosted our “Advancing AI” event. During that event, we announced our 5th Generation CPUs, the MI325X GPU, and gave a preview of upcoming GPUs. We also introduced new networking products designed for GPU connectivity.

Gilles Garcia: We also launched the Ryzen AI family and the next generation of AI-embedded PCs. While these announcements were made just two days ago, we are showcasing them here at IMC 2024, particularly for the telecom industry.