The NPU: Neural processing unit or needless pricey upsell?
- Reference: 1731331930
- News link: https://www.theregister.co.uk/2024/11/11/npu_debate/
- Source link:
In the past 12 months, these AI-focused processors have been touted as the next essential upgrade – one that everyone apparently needs to make the most of artificial intelligence. But is this all just marketing hype, or do NPUs genuinely offer the transformative value they promise?
What are NPUs?
NPUs are specialized processors within system-on-chips (SoCs) designed to handle AI-specific tasks, like for example, background noise suppression, real-time video enhancement, and basic generative AI functions. Companies including Intel with its VPU in Meteor Lake, AMD with Ryzen AI, and Qualcomm with the Hexagon AI processor have all embedded NPUs into their silicon, claiming they will revolutionize the computing experience by making devices smarter and more efficient. The idea is to offload AI workloads from the CPU and GPU to save power, theoretically improving battery life and providing faster on-chip AI processing.
But are these AI-enabled processors genuinely game-changing features or are they occupying precious die space that could be better utilized to meet the real needs of users?
The reality behind NPU benefits
While NPUs do add efficiency, particularly in mobile devices where every watt saved is valuable, their impact on laptops – where battery life is already robust – is harder to justify. The tasks NPUs handle are largely niche and have a limited impact on the average user's experience. If you're someone who frequently uses voice commands or relies heavily on video call enhancements, an NPU might save some battery life. But for most users, at present it's a nice-to-have feature rather than an essential one. CPUs and GPUs have managed these functions adequately for years, and while NPUs might lower power consumption slightly, the innovation is more about incremental efficiency than offering meaningful new capabilities.
Take Intel's Meteor Lake VPU, for instance. It's marketed as a solution for on-device AI tasks like video call background blurring and noise cancellation – tasks that CPUs and GPUs have been handling effectively. The primary benefit is a marginal boost in power efficiency, which, while useful, is unremarkable when you consider the overall computing experience. AMD's Ryzen AI takes a similar approach, offering efficiency gains without groundbreaking functionality. Qualcomm's Hexagon processor, drawing from its mobile pedigree, brings similar capabilities to laptops but doesn't significantly expand the range of applications for most users.
TOPS trumps
When discussing NPUs, vendors often highlight TOPS as a metric of performance. Intel's upcoming Lunar Lake platform boasts a 48 TOPS NPU, AMD's Ryzen AI 300 series is capable of 55 TOPS, and Qualcomm's Snapdragon X Elite features a 45 TOPS NPU. These numbers are thrown around as if they have substantial meaning for real-world users.
However, TOPS is a theoretical measurement of peak performance under ideal lab conditions. It's calculated based on the number of multiply-accumulate (MAC) units and operating frequency, but it doesn't necessarily translate to actual performance gains in everyday use. For the average user, these figures are as meaningful as theoretical horsepower in a car they will never drive at top speed.
The trade-off: Die space utilization
Including an NPU consumes valuable die space, which could potentially be allocated to enhance more universally beneficial features including CPU cores or GPU capabilities. Using AMD's Zen 5-based Ryzen AI 300 mobile SoC as an example, the NPU occupies about 10-15 percent of the die space – a significant portion. If that space were instead used to add more CPU cores, users could experience noticeable improvements in multi-threaded applications, benefiting developers, content creators, and power users alike.
[1]Apple quietly admits 8GB isn't enough in 2024, M4 iMac to ship with 16GB as standard
[2]As Arm rivals cook up custom silicon, Mediatek sticks to tried-and-true Cortex recipe
[3]Arrow Lake splashdown: Intel pins hopes on replacement for Raptors
[4]AI PCs will dominate shipments by 2026, but not because of demand
Alternatively, expanding the integrated GPU could offer better graphics performance, a feature that would be appreciated by gamers and professionals using graphics-intensive applications. Given that GPUs have traditionally been the go-to hardware for AI workloads, enhancing GPU capabilities could serve a dual purpose.
Are NPUs truly future-proofing?
Manufacturers promote NPUs as essential for future-proofing laptops in an AI-driven world. However, given the rapidly evolving nature of AI, it's challenging to predict which hardware features will remain relevant. While NPUs do offer some advantages for specific AI tasks, most users are unlikely to notice their absence. The majority of everyday computing tasks – like web browsing, document editing, and media consumption – do not require AI-driven optimization.
Even for users who occasionally interact with AI-powered features, CPUs and GPUs are typically sufficient for handling these workloads, albeit with slightly higher power consumption. The promise of NPUs lies more in potential future applications than in current, tangible benefits for the average consumer.
Gimmick or genuine innovation?
While AI has many practical applications – such as speech-to-text conversion and real-time translation tools – the inclusion of NPUs in laptops feels premature. The technology seems to be a solution in search of a problem, driven more by marketing strategies than by actual user demand. Until AI applications become truly mainstream and indispensable in daily computing, NPUs may remain an overhyped feature rather than an essential component.
In the meantime, consumers might benefit more from enhancements in processing power, graphics capabilities, and overall system performance – improvements that offer immediate and noticeable advantages. As it stands, NPUs are an interesting development, but perhaps not the game-changing innovation they're slated to be - at least for users.
PC makers are keen to promote hardware containing an NPU, possibly because AI PCs promise to lift average sales prcies across the sector by [5]five to ten percent . ®
[6]
JavaScript Disabled Please Enable JavaScript to use this feature.
Get our [7]Tech Resources
[1] https://www.theregister.com/2024/10/28/apple_imac_m4/
[2] https://www.theregister.com/2024/10/22/arm_custom_silicon_interview/
[3] https://www.theregister.com/2024/10/10/intel_arrow_lake_deep_dive/
[4] https://www.theregister.com/2024/09/25/analysts_ai_pcs_shipments_gartner/
[5] https://www.theregister.com/2024/09/13/win_11_refreshes_delayed_pc_makers/
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2ZzI4N0x1tDYrMVKhYc6K0QAAAQk&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[7] https://whitepapers.theregister.com/
Almost correct...
From the opening paragraph--
"If you haven't heard of neural processing units (NPUs) by now, you must have missed a year's worth of AI marketing from Intel, AMD, and Qualcomm."
Should have read...
"If you haven't heard of neural processing units (NPUs) by now, you must have missed a year's worth of AI marketing hype from Intel, AMD, and Qualcomm."
Re: Almost correct...
TBH, whenever marketing is involved, hype is implied
What does an NPU actually do?
I mean the actual functionality, not the hype.
GPUs run a "map" function on a huge array of data at selectable floating point resolution. The "reduce" performance tends to be pretty awful and better done by CPU. So I can use them for tasks requiring a 'map' as long as I can readily get the data in and the results back out.
What does an NPU do?
Can I write an NPU 'shader'? If so, what is the hardware optimised for?
Re: What does an NPU actually do?
Was wondering the same thing. Some kind of shonky hardware almost-a-neural net or something?
All I see...
Is another possible attack-vector for malware, spyware and ransomware...
What memory does this NPU have access to ?
Are there any protections built-in to prevent access to certain memory regions ?
I think it's just a matter of time before someone finds a way to get these things to do something they shouldn't.
I think a lot of AI implementations...
...Will wind up being built into Cloud Computing. If you let the users run models on their own hardware you give them too much control and can't charge them as much for it.
Multi tasking?
How do NPUs handle multi tasking? In fact can / do they at all? Does an executable "claim" it and prevent others from using it? Does the NPU need to run a complete iteration of the neural net, then hand over to other processes? I really want to know how this stuff works.
Maybe I should ask ChatGPT or Gemini, but they'll tell me something that they're 100% sure of, but could be a complete hallucination.
It's a scam to sell more hardware.
Because W11's fake 'minimum requirements' didn't con enough of us.
Same reason why they block older versions, remove legacy compatibility and demand subs rather than standalone.
The TOPS rating is like those vehicle mpg, emissions and EV range ratings that aren't real world. It's not vendor and customer, it's predator and prey or farmer and sheep. They are no better than politicians, which is the worst thing you can say about anyone.
NPU = FAU. **** All Use.
Poorly written set of poll options
Shirley the first and third choices in your poll are just saying the same thing in different ways?
Re: Poorly written set of poll options
They are but don't ... already got my coat ...
Under a rock
Apparently I've been living under a rock for a year because I have never heard of these things. And I thought I was reasonably current with tech.
Or perhaps I did hear about them, promptly filed them under Next Overhyped Thing, and completely ignored them since?
Do we actually need AI on our computers? What is the real world benefit? At what cost (not just the extra cost for the CPU - what about power draw, heat, memory usage, performance, reliability etc?) Will the AI be like the LLM's (aka Kit-Bash solutions? If you're wondering about that - take a look at some of the AI generated images and texts... they're... unreal) Will they deliver what's promised? Can we opt out?
The latter, to me, is the most important point: Can I take the non-AI version, please? After all, if it's a choice of take it or leave it, then we can't do a real world comparison as to if these NPU's are delivering anything to us other than bigger bills. Got to do side-by-side comparisons so we can make informed decisions as to if NPU is useful or just that Needlessly Pricy Upsell.
You forgot the evil empire...
called Apple.
Haven't they had NPU's in some shape or form in their CPU for some years now?
All silently working in the background to steal your data and send it to the mothership (or so the detractors try to tell us)
They must be even worse than Intel or Amd because there is no choice now to have a device without an NPU...
El Reg is slipping or has the change in ownership made it a more apple friendly place. We deserve to know.