AMD Aims for Nvidia’s AI Leadership (NASDAQ:NVDA)

NVIDIA Corporation (NASDAQ:NVDA) has long been unbeatable in the GPU space, and its lead, especially in AI applications, has often seemed insurmountable. While that may not be in danger of changing tomorrow, a new AI benchmark from vendor-led organization MLCommons has revealed that It shows that competition is definitely increasing by showing processor performance in a fair and controlled way. release A few weeks ago, Advanced Micro Devices, Inc. (Am) is closing the gap with Nvidia’s performance in AI inference workloads. Is AMD on the verge of catching up? Let’s take a closer look.

AMD puts Nvidia in hot water

AMD has bounced back from the brink of bankruptcy to surpass Intel Corporation (INTC) in the datacenter CPU business, while NVIDIA has been able to maintain its lead in GPUs despite AMD’s forays into that market. The accelerator market is growing for a number of reasons. In my opinion, the main reason is the overwhelming performance advantage of Nvidia’s processors. While Intel struggles to scale out its 10nm (Intel 7) node, Taiwan Semiconductor Manufacturing Company Limited (Taiwan Semiconductor ManufacturingTSMC) suddenly appeared, allowing AMD to design processors at better nodes and take market share from Intel. Because Nvidia was essentially fabless, it was able to avoid these pitfalls and focus on what its engineers do best: designing cutting-edge processors.

And they are really good. Whether it’s gaming or the datacenter, AMD has always been several generations behind Nvidia in terms of performance and power consumption. But AMD has had success in the datacenter server market with its EPYC series of CPUs, and that has allowed them to reinvest that success back into GPU research and development, and it’s starting to pay off.

While Nvidia remains the overwhelming leader in AI training performance, the real golden egg in AI is inference workloads. To quickly summarize the difference, training is the process of “teaching” an AI model with a dataset, while inference is the process of the taught model making predictions based on data it has never seen before. Intel CEO Pat Gelsinger gave the example of creating and using weather models. While only a few organizations forecast the weather, hundreds of millions of people check the weather forecast every day. It’s easy to see why inference benchmarks are catching people’s attention.

MLPerf Inference v4.1 includes the latest performance results from numerous chipmakers, but I will focus on AMD and Nvidia here. In my opinion, the notable takeaway is that Nvidia’s lead is shrinking; for the first time, AMD has demonstrated parity with Nvidia’s current generation processors on inference workloads.

Note: “Genoa” and “Turin” refer to the generation of AMD’s EPYC that the servers are running and these tests were performed on an 8xGPU configuration.

As you can see, AMD’s MI300X is virtually on par with Nvidia’s H100 80GB GPU in terms of tokens/sec in both server and offline inference workloads (server mode is more consistent with real-world interactions). We don’t know exactly what the retail price of these GPUs is, but we do know that AMD is aggressive in pricing their high-speed server products, and Nvidia, with its 50+% net profit margins, is not. This is the same strategy AMD used to undercut Intel in the CPU market, and with the price and performance parity mentioned above, this strategy may drive some customers away from Nvidia.

Additionally, it’s important to note that the MI300X actually comes with 192GB of HBM3 memory, significantly more than the 80GB on the H100 and the 141GB on the H200. The model used in these evaluations, the Llama 2 70B, is fairly lightweight, so these benchmarks may be underestimating the performance of AMD’s processor.

The benchmark also includes some perspective on Nvidia’s matchups.

From these two images, I think it’s clear that AMD wants to highlight parity with the H100, while Nvidia wants to highlight the lack of parity with the H200. Both are valid points. AMD wants to draw attention to the fact that they have made significant progress in providing a viable alternative to Nvidia’s dominance, while Nvidia wants to show that the performance gap is still large.

I’m sure many Nvidia bulls are reading this article and wondering why they’re worried when the company is still a generation and a half ahead. My answer to that is to look at what AMD did to Intel. Nvidia’s management is much more competent, but mistakes and failures happen. For example, the release of Nvidia’s next-generation GPU Blackwell series was delayed due to a small design flaw that affected yields, despite what was ultimately a fairly short delay. As processors become more complex, design flaws become more likely and can erase what seemed like a huge technological advantage.

As for Blackwell, Nvidia submitted benchmarks for just the B200 (the more powerful Blackwell chip), showing impressive performance of 10,755 tokens/second in server mode on the Llama2-70B model (it’s unclear whether this is due to hardware improvements or FP4 support), which represents a nearly 4x improvement over the H100 and MI300X, and a 2.5x improvement over the H200.

All of this sounds impressive, but the B200 will sell for more than double the retail price of the MI300X, not to mention significantly higher power requirements. The B100 will be somewhere in between. Still, the value proposition will undoubtedly be tilted in Nvidia’s favor, even if the price is significantly higher. And since the cycle never ends, AMD is also targeting a new release in Q4 with the MI325X.

AMD is expected to announce more details about the chip’s release at its annual Advancing AI event, outlining significant performance, efficiency, and memory boosts. Specifically, the MI325X will feature 288GB of large HBM3E memory (more dense than HBM3). This is significantly higher than the 192GB HBM3E in the B100 and B200, which could give AMD a value advantage in inference workloads for larger AI models.

That said, it’s important to remember that hardware is only one side of the equation. Nvidia’s true strength lies in its CUDA software layer, which is the gold standard for developers writing real applications using these models. AMD has built ROCm to be a viable alternative, but adoption is still minimal compared to CUDA, and until that changes, Nvidia’s current customer base will likely be sticky and unwilling to switch ecosystems.

On a pure compute basis, AMD appears to be closing the gap on Nvidia in AI inference workloads, and while Nvidia still offers a lot of advantages that AMD has to chip away at, AMD bulls should be excited about the progress.

Investor View

Okay, that was a lot of technical talk, but what about the impact on stocks?

I believe the overall AI market will continue to hold AMD and NVDA in high valuations for years to come. The GPU accelerator market will likely remain strong as cloud providers and other large technology companies scale for the nascent AI revolution. Nvidia has the expertise, margins and market position to continue to succeed despite AMD’s advances.

This wave of Blackwell processors could boost already impressive margins further with new high-margin processors added to the product mix, and Nvidia is a buy on the back of its continued lead in AI applications and a seemingly unbreakable software moat. Bulls would be well advised to keep a close eye on these benchmarks going forward to see if the company maintains, loses or extends its lead over AMD and other competitors.

However, I believe AMD offers a much more attractive risk-reward profile than NVDA. The GPU accelerator market is poised for gains over the long term, but AMD’s market share is still small, so a potential upside seems inevitable as the company continues to improve its products and close the gap with Nvidia. However, this could take several years, or even longer, to materialize. So, those looking to buy AMD stock to benefit from the company taking share from Nvidia with the MI325X or MI350X may want to wait a bit. But in the long term, AMD is a company with the right moves in all areas and a competitive edge moving in the right direction. These factors make AMD a strong buy.

Thank you for reading!

Editor’s Note: This article discusses one or more securities that do not trade on a major U.S. exchange. Please be aware of the risks associated with these stocks.