Skip the content
Image : AMD
AMD’s hardware teams have attempted to redefine AI inferencing by using powerful chips such as the Ryzen AI Max or Threadripper. In software, AMD has been largely absent when it comes to PCs. AMD executives claim that this is changing.
AMD’s Advancing AI Thursday event focused on enterprise-class graphics cards like its Instinct line. AMD relies on a software platform called ROCm that you may not be familiar with. AMD has released ROCm 7, which, according to the company, can boost AI inferencing three times by using the software alone. It’s finally coming to Windows in order to compete with Nvidia’s CUDA dominance.
Radeon Open Compute is AMD’s open stack of software for AI computing. It includes drivers and tools that run AI workloads. Remember the Nvidia GeForce RTX RTX5060 debacle a few weeks ago? Nvidia’s latest GPU would have been a lifeless piece of silicon without a driver.
AMD was in a similar situation early on. AMD had to make a decision. Without the unlimited coffers of Nvidia and other companies, it chose to prioritize enterprise GPUs for big businesses instead of client PCs. Ramine Roane is the corporate vice president for the AI solutions group. She called this a “sore issue:” “We focused on ROCm cloud GPUs but it wasn’t working on the endpoint – so we’re repairing that.”
Mark Hachman / Foundry
In today’s world, simply shipping the best product isn’t always enough. Capturing customers and partners willing to commit to the product is a necessity. It’s why former Microsoft CEO Steve Ballmer famously chanted “Developers developers developers” on stage; when Sony built a Blu-ray drive into the PlayStation, movie studios gave the new video format a critical mass that the rival HD-DVD format didn’t have.
Now, AMD’s Roane said that the company belatedly realized that AI developers like Windows, too. “It was a decision to basically not use resources to port the software to Windows, but now we realize that, hey, developers actually really care about that,” he said.
ROCm will be supported by PyTorch in preview in the third quarter of 2025, and by ONNX-EP in July, Roane said.
Presence is more valuable than performance
This means that AMD processors are finally going to have a larger presence in AI applications. If you own a laptop or desktop with an AMD Ryzen AI processor inside, or one with a Ryzen AI max chip, your computer will have more chances to run AI applications. PyTorch is a machine learning library that is used by popular AI models such as Hugging Face’s ‘Transformers. This should make it easier for AI models like Hugging Face’s “Transformers” to utilize Ryzen hardware.
The ROCm feature will be added to all “in-box” Linux distributions: Red Hat in the second half 2025, Ubuntu (same) and SuSE.
Roane provided some helpful context on what model sizes each AMD platform is capable of running, from a Ryzen A300 notebook to a Threadripper Platform.
Mark Hachman / Foundry
But performance also improves substantially
ROCm 7’s AI performance improvements are substantial. They include a 3.2X improvement in Llama 3.0 70B, a 3.4X improvement in Qwen2-72B and a 3.8X increase in DeepSeek R1. The “B” represents the number of parameters in billions. The higher the parameters the better the quality of the outputs. Roane stated that these numbers are more important than ever before, as inferencing processors are growing faster than processors for training.
(Training”) generates the AI model used in products such as ChatGPT or Copilot. “Inferencing”, on the other hand, is the actual process of using AI. You could train an AI to be able to answer all questions about baseball. But when you ask if Babe Ruth is better than Willie Mays you are using inference.
Mark Hachman / Foundry
AMD said that the improved ROCm stack also offered the same training performance, or about three times the previous generation. Finally, AMD said that its own MI355X running the new ROCm software would outperfom an Nvidia B200 by 1.3X on the DeepSeek R1 model, with 8-bit floating-point accuracy.
Again, performance matters — in AI, the goal is to push out as many AI tokens as quickly as possible; in games, it’s polygons or pixels instead. Simply offering developers a chance to take advantage of the AMD hardware you already own is a win-win, for you and AMD alike.
The one thing that AMD doesn’t have is a consumer-focused application to encourage users to use AI, whether it be LLMs, AI art, or something else. Intel publishes AI Playground, and Nvidia (though it doesn’t own the technology) worked with a third-party developer for its own application, LM Studio. One of the convenient features of AI Playground is that every model available has been quantized, or tuned, for Intel’s hardware.
Roane said that similarly-tuned models exist for AMD hardware like the Ryzen AI Max. However, consumers have to go to repositories like Hugging Face and download them themselves.
Roane called AI Playground a “good idea.” “No specific plans right now, but it’s definitely a direction we would like to move,” he said, in response to a question from PCWorld.com.
Mark Hachman, Senior Editor, PCWorld.
Mark is a technology writer with over 30 years experience. He has written over 3,500 articles, covering PC microprocessors and peripherals, Microsoft Windows, and other topics, for PCWorld. Mark has written for PC Magazine, Byte and eWEEK as well as Popular Science, Electronic Buyers’ News and Electronic Buyers’ News. He also shared a Jesse H. Neal Award with Popular Science for breaking news. He recently gave away a collection consisting of several dozen Thunderbolt Docks and USB-C Hubs, because his office has no room.
