Tired of it so-called AI PCs already? Buckle up, because we, or perhaps artificial intelligence, haven’t even begun to peak yet. More energy efficient ARM-based CPUs don’t get nearly as much of a spotlight as the x86 type. Still, chip maker Qualcomm is trying to draw your attention to its latest Snapdragon chips by promising the best AI performance available on a consumer-end chip.
Sure, we’ve known the Snapdragon for months. Maybe, just maybe, some app developers and Windows itself will finally make good on the promise of on-device AI.
The Snapdragon The Snapdragon X Plus is heading in the same direction, but will be slightly slower than its bigger brother. Specifically, Qualcomm’s Geekbench 6 benchmarks show that the X Plus is 37% faster and 54% more efficient than an Intel Core Ultra 7 155H, which is widely used in most countries mobile PCs coming to market this year. Cinebench 2024 tests also showed that it was significantly faster than Intel’s basic Meteor Lake chip and the AMD Ryzen 9 7940HS.
Gizmodo had the opportunity to test some of these benchmarks on test machines manufactured by Qualcomm. While these benchmarks seemed accurate, we would much rather have consumer devices on hand to test and compare. It can also be compared to Apple’s M3 chip, perhaps with a little less enthusiasm. The M3 chip might still have the edge, especially in multi-core performance, but again, it’s something we’d much rather test internally. Otherwise, the X Plus promises to support one internal laptop screen at 4K and 120 Hz or up to three external monitors at 4K and 60 Hz. It will also be WiFi 7 compatible and allow recording with a 4K HDR webcam.
AI performance is much more difficult to quantify. Qualcomm relies on the quality of its NPU, which is identical to that of the X Elite and the X Plus. The chipmaker says its NPU reaches a maximum of 45 TOPS, which is equivalent to “trillion operations per second.” Sure, that could mean that Stable Diffusion running on the device can spit out a few deep fakes faster than the competition, but we’re still waiting to see what dedicated user applications will make use of the hardware.
In 2024 there will be a total of four SKUs for the Snapdragon brand. Without boring you, the other two elite versions reach their top speed at slightly slower speeds and do not have the full GPU performance of the top tier. We’ve seen some Qualcomm machines running emulations of games like… Baldur’s Gate III with an X Elite chip at frame rates of over 30, but at medium to low graphics settings. It’s debatable whether Qualcomm claims that the X Elite’s Adreno GPU should be gaming capable when many games still don’t support the ARM architecture. Apple has been dealing with this problem for ages, and even after a major gaming push in its M3 chip, the lack of playable titles still exists. I can’t imagine trying high-profile games on just the X Elite without a discrete GPU.
Meanwhile, the X Plus only has one SKU and has ten cores with a speed of 3.4 GHz and no dual-core boost. It has a GPU performance of 3.8 TFLOPs, but also has an equivalent NPU to the X Elite. The more expensive chip will likely still deliver better AI performance compared to the case is.
What does this mean for the AI running on the PC?
The X Plus and the We can probably expect re-releases of existing versions Work laptopsalthough we don’t know what the price difference will be between an X Plus and a Core Ultra 155H or an X Elite and the Intel Core Ultra 9 185H.
Companies like Microsoft and Intel have bombarded us with incredible AI promises over the last six months, and lo and behold, it hasn’t quite come true as they said it would. We tested quite a few laptops (the Meteor Lake chips don’t have a desktop version yet), and while they performed well and benchmarked well, perhaps even ahead of other mobile processors in their price range, This is not really provided by AI on the device. The NPU, which is supposed to handle routine smart tasks, never seems to rev up, whether you’re using background blur in Zoom or trying to use the dedicated Windows Copilot in Windows 11.
In conversations with Intel, we were repeatedly told that the current generation of chips can support AI on the device. Executives have told Gizmodo that the problem lies not so much with chipmakers but with app developers. For once, it’s not the hardware people trying to keep up with software requirements, but the other way around.
But that precludes whether app developers find much use in hardware that supports AI and the massive cloud data centers of giants like Nvidia, OpenAI and Microsoft. Microsoft’s Build conference is scheduled for May 21st, and that’s where we might finally get a look at the next generation of Copilot, which will hopefully break some of these obvious, growing problems we continue to have with Windows 11.
Most importantly, Copilot could transform into a true AI assistant, able to turn settings on or off, adapt your computer to the context in which it is running, and generally be more than a glorified chatbot you already see in the Bing -Search find. This is based on early demos we saw at some OEMs’ PC showcases last year. So we’ll have to wait and see what Microsoft has in store.