Jacob Roach / Digital Trends

I ’m sorry to say it , but I ’d be lying if I articulate otherwise . NPUs are useless . Maybe that fare off as abrasive , but it does n’t hail from a place of cynicism . After testing them myself , I ’m genuinely convert that there ’s not a compelling role for NPUs decently now .

When you go to grease one’s palms your next laptop , there ’s a expert opportunity it will have aNeural Processing Unit ( NPU)inside of it , and AMD and Intel have been making a big fuss abouthow these NPUswill drive the hereafter of the AI PC . I do n’t doubt that we ’ll get there eventually , but in terms of a modern use case , an NPU really is n’t as effectual as you might think .

When talking about Meteor Lake atCES 2024 , Intel explain that it use the various processor on Meteor Lake microchip for AI work . It does n’t all fall on the NPU . In fact , the most demanding AI workloads are offloaded to the integrated Arc GPU . The NPU is chiefly there as a low - office option for steady , low - volume AI workloadslike background blur .

Your GPU steps in for anything more acute . For example , Intel register off a demo in Audacity where AI could separate the audio tracks from a single stereo file , and even transliterate the lyrics . It also showed AI at work , with Unreal Engine ’s Metahuman , transform a video transcription into an animation of a naturalistic secret plan character . Both break away on the GPU .

These are some great use cases for AI . They just do n’t need the NPU . The dedicated AI processor is n’t near as powerful as the GPU for these bursty AI workloads .

The main purpose case , accord to Intel , is that the NPU is efficient , so it should help your barrage life when black market those steady , low - intensity level AI workloads like background blur on it rather than the GPU . I tested that on an MSI Studio 16 go one of Intel ’s new Meteor Lake CPUs , and it ’s not as big of a great power saving as you might expect .

Over 30 minutes , the GPU had an mean major power draw of 18.9W while the NPU average 17.6W. What was interesting is that the GPU spiked ab initio , going up to somewhere around 38W of total top executive for the organization on a chip , but it lento fawn back down . The NPU , on the other hand , started low and slowly went up as prison term go bad on , eventually settling between 16W and 17W.

I need to designate out that this is total package power — in other word , the power of the full chip , not just the GPU or NPU . The NPU is more effective , but I do n’t know how much that efficiency matters in these low - intensity AI work load . It ’s plausibly not salvage you much battery living in real - world use .

I ’ll have to wait to test that assault and battery life story once I have more prison term with one of theseMeteor Lakelaptops , but I do n’t surmise it will make a huge difference . Even the serious Windows laptop have somewhat mediocre battery life compared to thecompetition from Apple , and I doubt an NPU is enough to change that moral force .

That should n’t distract from the exciting AI apps we ’re begin to see on personal computer . From audaciousness to the Adobe suite to Stable Diffusion in GIMP , there are a long ton of agency you may now leverage AI . Most of those applications are either do by in the swarm or by your GPU , however , and the consecrate NPU is n’t doing much outside of background fuzz .

As NPUs become more muscular and these AI apps more effective , I have no doubt that NPUs will witness their place . And it makes sense that AMD and Intel are repose the fundament for that to happen in the future tense . As it stands , however , the big focus on AI performance is mostly pore on your GPU , not the efficient AI central processing unit handle your television call .