AMD A new artificial intelligence chip was launched Thursday that targets Nvidia’s data center graphics processing units, or GPUs.
AMD said at a new product launch event on Thursday that the chip, called Instinct MI325X, will begin production before the end of 2024. If AMD’s AI chips are viewed as close to alternatives by developers and cloud giants NVIDIA products, which could put pricing pressure on Nvidia, which has a gross margin of about 75% and whose GPUs have been in high demand over the past year.
Advanced generative AI, such as OpenAI’s ChatGPT, requires large data centers filled with GPUs to perform the necessary processing, creating demand for more companies to provide AI chips.
Nvidia has dominated much of the data center GPU market over the past few years, but AMD has historically been in second place. Now, AMD aims to take share from its Silicon Valley rivals, or at least a chunk of the market, which it says will be worth $500 billion by 2028.
“Artificial intelligence demand actually continues to grow and actually exceeds expectations. It’s clear that investment rates are continuing to grow everywhere,” AMD Lisa Su said at the event.
AMD did not reveal new major cloud or network customers for its Instinct GPUs at the event, but the company has previously revealed that both Meta and Microsoft have purchased its AI GPUs, and that OpenAI uses them For some applications. The company also didn’t reveal pricing for the Instinct MI325X, which is typically sold as part of a complete server.
With the launch of the MI325X, AMD is accelerating its product plans to release new chips every year to better compete with Nvidia and take advantage of the boom in AI chips. The new artificial intelligence chip is the successor to the MI300X, which began shipping late last year. AMD said the 2025 chips will be called MI350 and the 2026 chips will be called MI400.
The launch of the MI325X will put it into competition with Nvidia’s upcoming Blackwell chips, which Nvidia says will begin shipping in high volumes early next year.
The successful rollout of AMD’s latest data center GPUs is likely to pique the interest of investors looking for more companies poised to benefit from the artificial intelligence boom. AMD’s shares are up just 20% so far in 2024, while Nvidia’s shares are up more than 175%. Most industry estimates say Nvidia has more than 90% of the market share for data center AI chips.
AMD shares fell 3% during trading on Thursday.
AMD’s biggest obstacle to gaining market share is that its rival’s chips use its own programming language, CUDA, which has become the standard for artificial intelligence developers. This essentially locks developers into Nvidia’s ecosystem.
In response, AMD said this week that it has been improving its competitive software called ROCm so that AI developers can more easily switch more AI models to AMD’s chips (AMD calls it an accelerator).
AMD positions its AI accelerator to be more competitive in use cases where AI models create content or make predictions, rather than in use cases where AI models process terabytes of data to make improvements. This is said to be partly due to the advanced memory AMD uses on its chips, which allows it to serve Meta’s Llama AI models faster than some Nvidia chips.
“What you see is that the inference performance of the MI325 platform on Llama 3.1 is 40% higher than the H200, and a lot of people are training as well,”
Also competing with Intel
While artificial intelligence accelerators and GPUs have become the most talked about parts of the semiconductor industry, AMD’s core business has always been the central processing unit (CPU), which is at the heart of nearly every server in the world.
AMD said in July that its data center sales in the June quarter more than doubled from last year to $2.8 billion, with AI chips accounting for only about $1 billion.
AMD’s spending on data center CPUs accounts for about 34% of total spending, the company said. This is still less than IntelWith its Xeon series of chips, the company remains the market leader. AMD aims to change that with a new series of CPUs it unveiled Thursday called EPYC fifth-generation.
The chips come in a variety of configurations, ranging from a low-cost, low-power eight-core chip costing $527 to a 192-core, 500-watt processor for supercomputers that costs $14,813 each.
AMD says the new CPUs are particularly well suited for feeding data into artificial intelligence workloads. Almost all GPUs require a CPU on the same system to boot the computer.
“Today’s artificial intelligence is really about CPU power, and you can see that in data analysis and a lot of these applications,” Su said.
watch: AMD CEO Lisa Su said technology trends will take years to emerge and we are still learning about artificial intelligence
About The Author
