Skip to main content

I tested Intel’s new overclocking tool, and it does AI all wrong

One of the most interesting features of Intel’s recent Core i9-14900K is its AI-assisted overclocking. Available through the Extreme Tuning Utility (XTU), AI Assist is billed as the natural next step of automatic overclocking. It uses AI to push chips further rather than relying on a predetermined list of checks that Intel already offers through XTU.

That’s the pitch, at least. But according to my own testing, AI Assist doesn’t do much of anything.

Recommended Videos

Not only does it provide minimal benefit by being restricted to the Core i9-14900K, it also shows the dangers of haphazardly adding AI in places where it doesn’t belong. That’s a shame, too, because we’ve seen some clear examples of how AI can enhance performance on PCs. AI Assist doesn’t fit in that mold.

Get your weekly teardown of the tech behind PC gaming
Check your inbox!

Too little to matter

Intel's 14900K CPU socketed in a motherboard.
Jacob Roach / Digital Trends

Don’t get your hopes up for a big performance boost out of AI Assist. It takes about 45 seconds to run the feature on your PC, and Intel says XTU tests applications during that window to determine the optimal overclocking settings. AI Assist is supposed to go beyond what the automatic overclocking in XTU offers, but it actually produced identical results.

The result: a whopping 100MHz boost across all cores. I ran the overclock through Cinebench R23, which produced a 1.5% improvement in multi-core performance and a 1.6% improvement in single-core performance. When testing a real application, there’s no difference at all. In our Handbrake transcoding benchmark, the AI Assist overclock produced an identical result to a stock Core i9-14900K.

Performance for the AI Assist feature with the Intel Core i9-14900K.
Jacob Roach / Digital Trends

I’ve never expected huge performance gains out of an automatic overclocking feature, but what’s frustrating is that the automatic overclocking in XTU produced the exact same overclock settings, along with identical performance. It’s not clear what AI is doing here, but from my testing at least, it’s no better than the otherwise limited automatic overclocking already present in XTU.

The application is the problem here, and that’s true on two fronts. First is the Core i9-14900K. This is already a chip pushed to its absolute limits, as it’s essentially a juiced-up version of last-gen’s Core i9-13900K. It doesn’t provide a huge performance lift, as you can read in our Core i9-14900K and Core i9-13900K comparison, so expecting overclocking to do much — AI or otherwise — isn’t a good idea.

It’s also the application of AI to overclocking. You could imagine how AI could be useful for automatic overclocking — train a model on thousands of different PC configurations and their components and use those patterns to figure out optimal settings — but it doesn’t look like Intel is doing much with AI Assist. I don’t know what’s going on behind the scenes, but whatever it is, it doesn’t appear too sophisticated. And based on the final result, that still holds true.

Wrong application

Intel announcing the Meteor Lake release date on Intel Innovation.
Intel

We’ve seen plenty of examples of AI being applied where it shouldn’t be this year, but the disappointing thing about AI Assist is that it could work. It just doesn’t do much right now with the Core i9-14900K and Intel’s seemingly basic AI implementation.

AI can dramatically improve the power of a PC, and we have a great example of AI doing just that: Nvidia’s Deep Learning Super Sampling (DLSS). It started off as just an AI-assisted upscaling tool, but it has grown to apply AI to frame generation and ray tracing denoisers, with all three working to improve performance and the gaming experience on PC.

DLSS and AI Assist aren’t the same thing, but DLSS is a prime example of applying AI to a problem and seeing tangible benefits from it. AI Assist is not. It’s applying AI in a situation where the performance gains are already limited, and doing so in a way that doesn’t do anything to improve the performance or user experience.

Intel has been pushing AI hard, as has the rest of the computing world. Its next-gen Meteor Lake chips come with a dedicated AI processor, despite the fact that the only clear application for one is Windows Studio Effects. We now have AI-assisted overclocking, despite the fact that the days of minor overclocks producing big performance jumps are far behind us. And next month, Intel is launching its next-gen CPUs for laptops with an event titled “AI Everywhere.” I’m not against AI being everywhere, but it needs to make sense.

AI Assist in XTU unfortunately doesn’t. It could be a great feature, particularly for lower-end chips that have some headroom to spare. And it could bring AI into the fold in a place where it matters, given that the model behind the scenes is truly delivering something you can’t get with a one-click overclocking tool. It’s possible we could see that in the future from AI Assist, but it’s not here now.

Jacob Roach
Lead Reporter, PC Hardware
Jacob Roach is the lead reporter for PC hardware at Digital Trends. In addition to covering the latest PC components, from…
Nvidia’s CEO — yes, one person — is now worth more than all of Intel
Jensen Huang at GTX 2020.

Nvidia is one of the richest companies in the world, so it's no surprise that the company's CEO, Jensen Huang, is quite wealthy. The most recent net worth numbers from Forbes puts into context just how wealthy the executive really is, though. Huang has an estimated net worth of $109.2 billion, which is around $13 billion more than the market cap of Intel across the entire company.

Although Nvidia makes some of the best graphics cards, the obscene amount of money the company has racked up over the past two years stems from its AI accelerators. In 2020, Forbes estimated that Huang was worth $4.7 billion, and even in 2023, after ChatGPT had already exploded onto the scene, the executive was worth $21.1 billion. Now, Huang is the 11th richest person in the world, outpacing Bill Gates, Michael Dell, and Michael Bloomberg.

Read more
More than seven months later, Intel CPU instability issue might be over
Intel's 14900K CPU socketed in a motherboard.

We first reported on the Intel CPU instability issue in February 2024, and since then, Intel has offered various fixes that helped, but still failed to fix the problem once and for all. Now, it finally seems like the owners of Intel's best CPUs might soon be able to rest easy. Intel has shared a new update that pinpoints the four causes of Raptor Lake problems and provides a fix.

Intel's July update on the matter disclosed that the company was aware of issues within the microcode and that the problem was related to incorrect voltages. Today's update breaks this down into four operating scenarios that can cause problems. Intel now refers to these long-lasting issues as the "Vmin Shift Instability."

Read more
Intel may have been right about killing Hyper-Threading after all
A Core i9-12900KS processor sits on its box.

Intel is getting rid of one of the features that has defined most of the best processors for more than a decade -- Hyper-Threading. It's the branded name Intel uses for simultaneous multi-threading, or SMT, and the company has already confirmed it won't use SMT on its upcoming Lunar Lake mobile CPUs. Rumor has it the company is also ditching SMT for its Arrow Lake desktop CPUs. Surprisingly, according to new leaks, killing SMT might have been the right call after all.

A handful of benchmarks have leaked for Arrow Lake CPUs. Starting off, the Core Ultra 7 265K and Core Ultra 9 285K both popped up in the Geekbench 6 database. The flagship Core Ultra 9 is a 24-core part, and it achieved a score of 21,075 in Geekbench 6's multi-core test. That's slightly above what you'll see with the Ryzen 9 9950X and on-par with the Core i9-14900K, both of which come with 32 threads due to SMT.

Read more