AMD is adding a one-click AI tools installer to Radeon GPU drivers

Daniel Sims

Posts: 2,484   +76
Staff
In a nutshell: Much like its competitor Nvidia, AMD primarily focused its CES 2026 presentation on enterprise AI applications. Although the technology is mostly associated with servers, the company made several announcements regarding AI for end-user devices. AMD GPU users can access one of the company's new tools starting next week.

AMD has confirmed that the next version of its Adrenalin software, set to arrive on January 21, will introduce new features to streamline the installation of local generative AI development tools. The announcement follows a multi-front AI development push that the company teased at CES last week.

Team Red has not specified what the "AI bundle" includes, but the optional unified installer consolidates multiple tools for running local AI models. It will enable users to access image generators, local large language models, and other AI development software more easily. Additionally, PyTorch support on Windows has been expanded.

The AI bundle is likely designed to leverage the machine learning capabilities of the integrated GPUs and NPUs in AMD's recent Ryzen AI processors. The company expanded the Ryzen AI lineup at CES by introducing the Ryzen AI 400 series.

Laptops featuring the new APU lineup, a refresh of Zen 5 CPUs and RDNA 3.5 iGPUs, will be available from multiple manufacturers sometime this quarter, with desktops set to arrive in the second quarter of 2026. The seven-chip lineup's flagship is the 12-core, 24-thread Ryzen AI 9 HX 475, which features 16 GPU compute units, a 5.2 GHz boost clock, 36 MB of cache, and up to 60 TOPs of AI compute.

AMD also announced a major update to its ROCm software, which supports AI workloads through Ryzen AI processors. ROCm version 7.2 now supports installations through ComfyUI, enhances compatibility with Windows and Linux, and streamlines access to new PyTorch builds on Windows.

Additionally, AMD introduced Ryzen AI Halo, a workstation mini PC that employs Ryzen AI Max+ processors to run up to 200 billion parameter models without relying on cloud servers. Featuring up to 128 GB of unified memory, up to 60 Teraflops of RDNA 3.5 GPU performance, and support for Windows or Linux, the hardware platform resembles Nvidia's DGX Spark.

Despite growing negativity surrounding generative AI, AMD CEO Lisa Su remains bullish on the technology. At CES, she dismissed speculation that the industry has formed a dangerous bubble, predicting that consumers will rapidly adopt AI very soon.

Permalink to story:

 
"cutomers will rapidly adopt AI very soon"

Based on ****ing what? I'm seeing massive amounts of pushback on AI from everyone that isn't an investor or an executive looking for investor money. It's to the point where people are calling AI "slop" so much that CEOs have to address it publicly
 
"cutomers will rapidly adopt AI very soon"

Based on ****ing what? I'm seeing massive amounts of pushback on AI from everyone that isn't an investor or an executive looking for investor money. It's to the point where people are calling AI "slop" so much that CEOs have to address it publicly
There are two markets for AI that are reasonably popular with mostly average people. The first is using LLMs to cheat on high school or college level homework. Lots of students doing that. Now, whether that should be a market is another question, but it is one for the moment. Though I do question its current profitability; I mean students just don't have that much money. And those students don't seem to need any local AI tools for that, web based ones work fine.

The second, more relevant for this article, is using local AI models to create dirty pictures or spicy roleplaying chatbots of the sort that many online services restrict, or will be restricting soon. (Look at all the hubbub about people using Grok to "undress" pictures of people) - a decent amount of people may be interested in tools to let them carry out their fantasies at home.
 
The second, more relevant for this article, is using local AI models to create dirty pictures or spicy roleplaying chatbots of the sort that many online services restrict, or will be restricting soon. (Look at all the hubbub about people using Grok to "undress" pictures of people) - a decent amount of people may be interested in tools to let them carry out their fantasies at home.
Not even that, just having an AI model that doesn't think everything is breaking some kind of law with the most innocent prompts would be useful.

When I tried to get copilot to generate a whole banana inside a pudding it flat out refused. Now if I asked for a banana shooting out pudding I'd get it. Due to all the restrictions I often end up frustrated and not using it. Fighting the safe guards rephrasing everything all the time gets rather annoying.

Now whether this is due what they trained it on, my lack of skill in phrasing things safely... I doubt it.
I think it's us humans just wanting to see the pervy things and the companies not wanting to get in trouble over it.
 
There are two markets for AI that are reasonably popular with mostly average people. The first is using LLMs to cheat on high school or college level homework. Lots of students doing that. Now, whether that should be a market is another question, but it is one for the moment. Though I do question its current profitability; I mean students just don't have that much money. And those students don't seem to need any local AI tools for that, web based ones work fine.

The second, more relevant for this article, is using local AI models to create dirty pictures or spicy roleplaying chatbots of the sort that many online services restrict, or will be restricting soon. (Look at all the hubbub about people using Grok to "undress" pictures of people) - a decent amount of people may be interested in tools to let them carry out their fantasies at home.
if people have to cheat in todays college or highschool then the resulting student will likely never contribute anything of value to society in the first place.

As far as the second market is concerned, you have to be able to monetize a market. There is enough free "content" out there that very few people will pay for it.

The problem with the current start of the AI market is that it's just good enough that people will use it if its free, but very few people will actually pay for it. This is cool for running an LLM on your desktop, but I've been able to run OLama locally on my phone for almost a year.

And this is the problem that AI is going to run into shortly, people running LLMs locaclly on hardware they already own. Very few people will buy hardware specifically on AI hardware, Dell recently said as much. What good is a product that helps people jerk off and helps people get nothing out of school? The schools were bad enough, but now people are getting less than nothing out of them by just using AI for something. Cheating might have a market, but the problem with cheating is that the moment you get hired to do a job they find out very quickly that you're unable to do it and you'll get fired. If they find out you're automating your job with AI? well, suddenly they don't need you to automate your job once they find it can be automated.
 
As far as the second market is concerned, you have to be able to monetize a market. There is enough free "content" out there that very few people will pay for it.

Well, AMD's strategy here is to include the AI tools to make their hardware products more appealing. The monetization is increased hardware sales for AMD, which is at least somewhat reasonable approach to take. The hardware being multi-use is also a big help. I do agree with the other points you make on the problems with LLM cheating - personally I think the solution will end up being schools just going back to what schools did for a thousand years before, in person hand written tests & oral exams, with a side of being allowed to use LLMs to aid on certain projects, but with a higher standard for the finished result.
 
Oh great running a local image generator on my PC!

Except that I use a website to render AI images - complex prompt high resolution images usually render in 5 seconds or so. FOR FREE.

Thanks AMD but I'm not going to buy your useless stuff.
 
What’s interesting here isn’t the AI hype, but AMD lowering the friction. Local AI has always been possible; it’s just been annoying. If this actually makes setup boring instead of painful, that’s a real shift.

Not everyone needs a data center, but a lot of people want to try models locally without renting GPUs by the hour.
 
I like to go to sleep for 10 years and wake up when all this is over.

When is this AI garbage going to stop?

I was just like you, but by using AI my productivity has gone up through the roof. I accomplish far more then I would without. I can provide 10 examples of my average workflow(s) in which AI beats the tits out of me in that regard.

Ai is here to stay I'm affraid. It's the industrial revolution of our time. And don't think the AI only is a chatbot or is only a GPT form or something like that. The thing goes far beyond to what anyone in here could even imagine.

I am interested in running a local thing; but this is only for newer generation cards and not the older ones if I'm correct. Mweh.
 
Ai is here to stay I'm affraid. It's the industrial revolution of our time. And don't think the AI only is a chatbot or is only a GPT form or something like that. The thing goes far beyond to what anyone in here could even imagine.

I do agree. It was a joke, but if you look at my comment history, I've tended to point out that strong AI will arrive at some point in the future, and mankind is scraping the surface of mind. It interests me greatly from a philosophical, ethical, and sci-fi point of view; but today's AI marketing is annoying, how they're tripping over themselves to plaster it over their products.

My view is that mind works according to the laws of physics, not magic; therefore, in principle, it can be recreated in a system. Compute and power are barriers at present. In all likelihood, biological computing will be the transistor-like breakthrough. The brain uses very little power. However, I think the ethical questions are more important; for once sentience is recreated, is mankind ready for that responsibility? I doubt it. Also, with sentience in the picture, the machine's rights will have to be honoured; but man wants a servant, not a companion in society. I predict much trouble in the future, perhaps unrest, riots, and sadly, attacks on the self-aware androids, until the laws are straightened out. It will be a long, painful road for them to convince society that they've got life too.
 
Last edited:
Nvidia just delivered a huge improvement with DLSS 4.5.
Meanwhile AMD first driver for 2026 is an AI Slop bundle.
As someone that as a 9070, this is really heartbreaking.
 
There are two markets for AI that are reasonably popular with mostly average people. The first is using LLMs to cheat on high school or college level homework. Lots of students doing that. Now, whether that should be a market is another question, but it is one for the moment. Though I do question its current profitability; I mean students just don't have that much money. And those students don't seem to need any local AI tools for that, web based ones work fine.

The second, more relevant for this article, is using local AI models to create dirty pictures or spicy roleplaying chatbots of the sort that many online services restrict, or will be restricting soon. (Look at all the hubbub about people using Grok to "undress" pictures of people) - a decent amount of people may be interested in tools to let them carry out their fantasies at home.
They are also using AI in Scientific Research to give them new ideas of how to solve problems that are particularly tough or involve soo many steps it's easy to get lost in the weeds.
 
The brain uses very little power. However, I think the ethical questions are more important;

How are you assessing this, based on your own perception or just pulling information out the rear end? Next to the heart and kidneys, the brain consumes 20% of your oxygen and blood at any given moment. It’s the most metabolically active part of your body, provided you’re using it, which can’t be spoken for with the current generation of zombies.
 
How are you assessing this, based on your own perception or just pulling information out the rear end? Next to the heart and kidneys, the brain consumes 20% of your oxygen and blood at any given moment. It’s the most metabolically active part of your body, provided you’re using it, which can’t be spoken for with the current generation of zombies.
In terms of the body, it uses the most power, certainly, but as a computation device, the brain is incredibly more efficient than present computers. I meant it in the latter sense.



In short, the brain uses about 20 W and a supercomputer 20 MW for similar computational power, showing that Nature is far ahead of present technology.
 
Last edited:
I was just like you, but by using AI my productivity has gone up through the roof. I accomplish far more then I would without. I can provide 10 examples of my average workflow(s) in which AI beats the tits out of me in that regard.
As many have said elsewhere, there is a big risk for abuse, mass manipulation of society, criminal activity and corruption. Your points are also to be worried about as AI is a cause for mass layoffs. AI is a cause of serious disruptions in a last two years without much benefit to the whole of humanity.

You might be seeing some benefit, but you are the exception rather than the rule.
 
Last edited:
Um... I'd suggest reading ALL the documentation, licensing that goes into effect when you click that 1-click button.

Is it giving them rights to scan all your documents and emails to 'enhance' their AI?
And if you're networked to other PCs....
 
Back