AMD aims to knock AI king Nvidia off its throne

amd advancing ai
Photo by PaulSakuma.com Photography)

AMD is moving decisively for dominance in the data center and AI. To do so, it must deal with old (Intel) and newer opponents (Nvidia). In San Francisco, it puts its assets on the table.

AMD chooses the slogan Advancing in AI for its annual conference in San Francisco, leaving little to the imagination. Not very original, since these days every tech conference revolves around AI. Yet that slogan is very relevant to AMD’s current direction: the data center is the destination, AI points the way. PCs are barely mentioned during the event, but even then a reference to AI is never long in coming.

“We believe that high performance computing is the fundamental building block of innovation. AI is the most exciting application of HPC. AMD wants to be the end-to-end leader in AI,” CEO Lisa Su explained the company’s focus in her opening remarks. The self-confidence Su exudes reflects a great belief within AMD that it can deliver on that ambition.

Rapid advance in server market

Upon hearing the name Advanced Micro Devices, as AMD is fully known, your first reflex may be to see if a sticker from the company is on your PC. For many, AMD is synonymous with laptop and desktop processors, and for nearly fifty years that was the company’s main source of revenue.

That changed when Su, who may celebrate her 10th anniversary as CEO at the conference of all times, took over. Su felt the company needed to expand its horizons beyond the PC. That resulted in 2017 in the first generation of Epyc processors, named after Naples, with which AMD dipped its toes into the deep waters of the server market.

amd advancing ai
Photo by PaulSakuma.com Photography)

Seeing Naples and then dying is out of the question at AMD. Su proudly holds the now fifth generation of Epyc in the palm of her hand during her keynote. The square is barely visible from the press stand, but everyone is trying to get as close a view of it as possible from every angle.

This year’s tour of Italy stops in Turin. With up to 192 cores and built on the latest Zen 5(c) architecture, Epyc Turin may not lack muscle to carry heavy (AI) workloads in data centers.

“You have to know that five years ago we represented maybe two percent of the server market. By the first half of 2024, our market share will have risen to 34 percent. Data centers now comprise 50 percent of AMD’s business. Thanks to these new innovations, we believe we can continue our rapid advance. Epyc Turin is a beast,” said Dan McNamara, head of the server branch within AMD, sounding as confident as his CEO.

Epyc Turin is a beast.

Dan McNamara, VP Server Business Unit AMD

Old and new acquaintances

With the beast Epyc Turin let loose in the AI arena, AMD wants to gobble up its competitors. One competitor knows it inside out: the eternal feud with Intel is also playing out in the server industry. Now that Intel is struggling, AMD sees an opportunity to deliver a knockout blow. While waiting for the market figures to finally turn around, it is already declaring technological victory over Intel.

No matter how many superlatives AMD pastes on Epyc Turin, the company itself realizes that good CPUs alone are not enough to rule the data center. A certain company called Nvidia has managed to win many souls with its Hopper GPUs. With its Instinct accelerators, AMD is now entering Nvidia’s territory and has no intention of budging.

“Until now, AI applications were largely borne by CPU capacity. New developments such as agentic AI are triggering a shift in market demand, creating more need for GPU capacity,” Su said. With the MI300X, AMD already had a direct opponent for Nvidia Hopper H100, now with the MI325X it has also found a challenger to the H200. The answer to Blackwell is fully in the pipeline and is expected in spring 2025.

read also

AI agents: iPhone moment or old wine in new bags?

That AMD and Nvidia would meet in the data center was written in the stars. For years, both companies fought a battle in the shadows over who could make the best graphics cards for PCs, though only gamers and hobbyists lay awake from that. The AI craze offered an opportunity to get to the forefront, and you could say that (for now) Nvidia has succeeded best, although of course AMD sees it differently.

Now Nvidia and AMD are looking each other in the eye again on the highest stage. At the same time, they must face each other there as well, as AMD works with Nvidia to pair Epyc processors with Hopper GPUs. Keep your friends close and your enemies closer: the saying applies nowhere more than in the tech industry.

For a long time, enthusiasts will not have to wait for AMD’s latest chips. In a barrage of announcements, successively Dell, HPE, SuperMicro and Lenovo took to the stage to show off their latest servers running AMD Epyc Turin and MI325X. “Maybe next time we should suddenly hand out order forms,” jokes Forrest Norrod, head of the Datacenter Solutions Business department.

Feeding the beast

With CPUs and GPUs, AMD’s arsenal is not yet complete. In data centers, networking plays a crucial role in linking hardware together. Since acquiring Pensando, AMD also has DPUs(data processing units) on offer. The chipmaker is happy to put that in the spotlight again during the event.

“The combination of CPUs and GPUs makes the parts stronger. Networking used to be relatively simple, sorry to all the networking specialists in the room. Because of AI, it has become much more complex,” Norrod said.

Soni Jiandani, who joined AMD’s networking division after the Pensando acquisition, agrees with that statement. “AI encompasses a whole ecosystem. We need to innovate on all fronts. AI systems connect to the front-end, GPUs communicate with the back-end of the network. The beast must be fed.”

AMD is keen to present itself as the party that is everywhere in the data center. In the presence of delegates from Cisco, Microsoft Azure and Oracle, among others, Jiandani got to show the press the latest Pensando Salina DPU. That operates in the front-end of the network, but AMD is also thinking about the back-end with Pollara, an Ethernet-compatible network card that injects extra performance, scalability and efficiency into a data center network.

From PCs to data centers, and back again?

The renewed focus on data centers does not mean that AMD has completely forgotten its roots in the PC industry. Still, it takes until the last ten minutes of the keynote before we get to see a first PC. Vlad Rozanovic, guest speaker for Lenovo, announces a ThinkPad PC with Ryzen AI Pro. Otherwise, the launch of the Ryzen AI Pro label for Copilot+ PCs seems to serve mainly as an invitation for Microsoft to come and promote Copilot once again.

read also

AMD aims to knock AI king Nvidia off its throne

Vamsi Boppana, AI specialist at AMD and McNamara admit that there is even some contradiction in Copilot+’s presence at the event. The explosion of AI is creating endless investments in data centers, which AMD obviously loves to see happen, while the idea behind Copilot+ is just about getting AI workloads back out of data centers and running locally on the device.

“Different workloads require different technologies and methods. You don’t always have time to go through the cloud,” Boppana said. To that, the gentlemen are asked whether AMD has a pocket-to-cloud-strategy like Lenovo aspires to. “If people are ever walking around with Epyc processors in their pockets, we will be very happy,” McNamara replies with a quip.

amd advancing ai
Photo by PaulSakuma.com Photography)

AMD got big with PCs and now wants to rule in the data center. The moment it felt it was hitting a ceiling within the PC market, it made the leap into servers and data centers, and today it is reaping the benefits. AMD is ready to compete with Nvidia for the AI crown.

Or will AI just ensure that AMD will soon have to take the opposite path toward PCs (and Intel) again? Whatever path AMD takes, it always seems to run into old acquaintances.