Intel's Arrow Lake-S won't be an AI powerhouse — 13 TOPS NPU is only slightly better than Meteor Lake, much less than Lunar Lake

News
By
published

NPUs are slowly filtering into the desktop segment, but mobile still gets the best NPU hardware for now.

Intel Meteor Lake
(Image credit: Intel)

Arrow Lake-S will be the first Intel desktop architecture with a neural processing unit (NPU), but it won't be as fast as people might expect. @Jaykihn on X reports that Arrow Lake-S will include an NPU that is only slightly more powerful than Meteor Lake's NPU, featuring just 13 TOPS of AI performance.

Arrow Lake-S's AI performance is well under Lunar Lake's 45 TOPS of NPU-only performance and nowhere near enough to qualify for Copilot+ certification. Arrow Lake-S would need at least 40 TOPS of AI performance from the NPU alone to qualify as a hardware-compatible product for Copilot+ PCs (if Microsoft ever allows desktops into the Copilot+ PC ecosystem).

The performance numbers from Arrow Lake-S NPU suggest that Intel utilizes a variant of Meteor Lake's NPU, which topped out at 10 TOPS of performance (no pun intended), possibly with a higher clock speed.

It is a mystery why Intel has decided to go with an NPU at all in Arrow Lake-S and use one that is not at least at the same performance level as Lunar Lake's NPU (after all, Lunar Lake is Arrow Lake's ultra-low power mobile counterpart — architecturally they are both similar). However, there are some takeaways we can surmise.

Having an NPU in a desktop environment is virtually useless; the main job of an NPU is to provide ultra-high AI performance with a low impact on laptop battery life. Desktops can also be used more often than laptops in conjunction with discrete GPUs, which provide substantially more AI performance than the best NPUs from Intel, AMD, or Qualcomm. For instance, Nvidia's RTX 40 series graphics cards are capable of up to 1,300 TOPS of AI performance.

As a result, putting a bleeding-edge NPU inside Arrow Lake-S would be virtually pointless and a complete waste of resources and money. However, there are still mobile variants of Arrow Lake that can take advantage of an NPU to consider.

Intel probably had to choose the best middle-ground between both use cases with Arrow Lake. A lower-powered NPU is perhaps cheaper to manufacture, which would be a better route than putting Lunar Lake's NPU in Arrow Lake. Another factor worth mentioning is that many of the Arrow-lake-equipped laptops will be sporting a discrete Nvidia GPU (inevitably), which lessens the importance of NPU performance when users can run AI-specific tasks on the Nvidia GPU (which has AI-specific tensor cores). This makes Intel's decision to put a weak NPU inside Arrow Lake even more logical.

This does not discount the idea of creating two (or more) dies for Arrow Lake, one without an NPU and one with an NPU aimed at desktop and mobile systems, respectively. Regardless, we will probably get an answer once Arrow Lake arrives later this year.

Aaron Klotz
Aaron Klotz
Contributing Writer

Aaron Klotz is a contributing writer for Tom’s Hardware, covering news related to computer hardware such as CPUs, and graphics cards.

See more CPUs News
31 Comments Comment from the forums
  • It is a mystery why Intel has decided to go with an NPU at all in Arrow Lake-S and use one that is not at least at the same performance level as Lunar Lake's NPU (after all, Lunar Lake is Arrow Lake's ultra-low power mobile counterpart — architecturally they are both similar).

    It's no mystery actually, because Arrow Lake-S lineup is most likely using an older NPU architecture, whereas INTEL used the NPU 4 arch version for the Lunar Lake mobile lineup (see image below). That's why there is a lot of difference in TOPS value between these two.

    Nonetheless, Arrow Lake S processors will have a combined 37 AI TOPS which puts it close to the Ryzen 8000G "Hawk Point" APUs which have a combined TOPS of 39.


    https://i.imgur.com/BZ9lVDL.png
    Reply
  • edzieba
    Oh no! Anyway...
    Reply
  • At this point though, none of these predictions and leaks can be fully trusted, and this poster has been tweeting almost every day with some new info, which in itself cannot be verified because there is no proof whatsoever to backup his claim.

    So, at this point I think these are just clickbait posts, and nothing else.

    Sorry, that's what one has to assume when some random user floods the internet each day with a new leaked info. For each tweet, we now get a new article which makes it even more awkward and funny.

    I wouldn't take this TOPS data value seriously though. Not unless confirmed via some official Intel slide/document.
    Reply
  • usertests
    This is a good thing, if businesses throw the dGPU-less mini desktops away faster to get to 40 TOPs for Copilot+... if there is demand for that. Then I'll buy them for cheep.
    Reply
  • thestryker
    The real question is why there would be any NPU in ARL-S given that it is a part of the SoC tile. They're certainly not just reusing the MTL tile since it had low power E-cores in it. It seems like a complete waste of silicon since they had to design the SoC tile for ARL-S in the first place. This would almost make sense if we were talking the ARL-P as Intel may not be putting the latest NPU in there and the IGP will be much more powerful.
    Reply
  • Well, on some other news, CPU-Z tool has just confirmed the nomenclature of the entire Arrow Lake S desktop processor lineup, and also Lunar Lake series.

    At least we can fully trust on this info since this is coming directly from the official CPU-Z change log.

    https://www.cpuid.com/softwares/cpu-z.html
    So for Arrow Lake-S, we have the following SKUs confirmed, but without any core counts though. If one would hazard a guess, then most probably, the Core Ultra 9 285K processor might be the flagship 24 cores part, sporting the "8 P + 16 E" core config:

    Core Ultra 9 285K
    Core Ultra 9 275,
    Core Ultra 7 265K
    Core Ultra 7 255,
    Core Ultra 5 245K
    Core Ultra 5 240.

    These are the Lunar Lake SKUs:

    Core Ultra 9 288V ;
    Core Ultra 7 268V,
    Core Ultra 7 266V,
    Core Ultra 7 258V,
    Core Ultra 7 256V ;
    Core Ultra 5 236V,
    Core Ultra 5 228V,
    Core Ultra 5 226V     (this specific part has a typo error in it's name)

    https://i.imgur.com/aLA4rmA.png

    Version 2.10 for windows® x86/x64 AMD Ryzen 9 9950X (16C/32T), 9900X (12C/24T), Ryzen 7 9700X (8C/16T) and Ryzen 5 9600X (6C/12T) Granite Ridge (Zen 5).AMD Ryzen AI 9 HX 375 (4x Zen 5 + 8x Zen 5c), Ryzen AI 9 365 (4x Zen 5 + 6x Zen 5c) Strix Point APUs.AMD Ryzen 9 8945H, Ryzen 7 8845HS (Hawk Point).Intel Core Ultra 9 285K & 275, Core Ultra 7 265K & 255, Core Ultra 5 245K & 240 (Arrow Lake).Intel Core Ultra 9 288V ; Ultra 7 268V, 266V, 258V, 256V ; Ultra 5 236V, 228V, 2266V (Lunar Lake).AMD Radeon RX 7600 XT (Navi 33 XT).New Chinese translation.
    Reply
  • thestryker
    Metal Messiah. said:
    There are Lunar Lake SKUs:

    Core Ultra 9 288V ;
    Core Ultra 7 268V,
    Core Ultra 7 266V,
    Core Ultra 7 258V,
    Core Ultra 7 256V ;
    Core Ultra 5 236V,
    Core Ultra 5 228V,
    Core Ultra 5 226V (this specifics part has a typo error in it's name)
    It's looking like you were spot on before about the 6/8 at the end signifying DRAM amount.
    Metal Messiah. said:
    Core Ultra 9 285K
    Core Ultra 9 275,
    Core Ultra 7 265K
    Core Ultra 7 255,
    Kind of curious what the significance of the second number of the model being different is if anything. It could just be a differentiation to make it less confusing for customers (eg thinking a 13900 is as fast as a 13900K), but companies aren't really known for caring a whole lot about that.

    I'm also curious if the tiled architecture will mean the death of the F SKU or if Intel will release some without an IGP. I cannot imagine the packaging failure rates on just the IGP will be high enough on their own to make for a whole F SKU line.
    Reply
  • thestryker said:
    The real question is why there would be any NPU in ARL-S given that it is a part of the SoC tile. They're certainly not just reusing the MTL tile since it had low power E-cores in it.

    Exactly my thoughts. I was also about to point this out.

    It makes little sense to put a SoC tile on a desktop part. So having an NPU seems a bit off to me. Like I mentioned above, I have doubts on these tweets being legit or not. I'm not buying all this leaked Arrow Lake info as of now.

    Will wait for some official leaked slide or document instead.

    But in any case, FWIW, the same leaker also claimed that Arrow Lake "refresh" CPUs which might incorporate a bigger and faster NPU within the SOC tile. The leaker states that this NPU will be an upgraded version from the one featured on the upcoming Arrow Lake-S.

    So with the upgraded NPU, Intel Arrow Lake refresh chips could see a larger die size, expanding by 2.8mm compared to current-gen Arrow Lake chips.

    While the die size will be larger, the package size will be the same. So the chips won't require major socket changes (BGA and LGA). They will retain the compatibility but motherboard vendors will have to enable Fast Voltage Mode (FVM) on the VccSA rails.

    Makes sense, eh ? To me, it doesn't. But here is the tweet anyway.

    1808621251965169774View: https://x.com/jaykihn0/status/1808621251965169774

    EDIT:

    Ouch, Break Time. Head spinning fast. Grabs a cappuccino ! 🥱
    Reply
  • thestryker
    Metal Messiah. said:
    Makes sense, eh ? To me, it doesn't. But here is the tweet.
    I saw that one earlier and almost laughed out loud. There's a zero percent chance that Intel would put out a socketed CPU that may or may not work in existing motherboards. If it was dependent on something implemented at the board level it would be a mandatory part of the specification.
    Reply
  • usertests
    Metal Messiah. said:
    It makes little sense to put a SoC tile on a desktop part. So having an NPU seems a bit off to me. Like I mentioned above, I have doubts on these tweets being legit or not. I'm not buying all this leaked Arrow Lake info as of now.
    Intel's tile strategy is all over the place, just look at Lunar Lake reducing the tiles and ditching LP E-cores.

    I'm not sure it doesn't make sense to have a "SoC tile" on desktop? Can you expand on that? Because if they are moving to tiles on desktop from now on, they could subdivide the chip in all sorts of ways. All we have to go on for now are Meteor Lake and Lunar Lake.

    NPU on desktop is going to be a thing simply to make sure it's everywhere as a minimum spec, and for power efficiency as long as it can get better TOPS/Watt than the iGPU. Seems absurd but it could be an "ESG" type of thing that governments want, and Intel will surely have 35W TDP 'T' variants of these chips being stuffed into OptiPlex, ThinkCentre, etc. business systems, giving them all NPUs.

    They also may end up with some SKUs that have no iGPU but do have an NPU. You'd almost certainly want to use the discrete GPU for AI in that case, but as long as it's there, it might see some use somewhere. You could also use it in the background while leaving the GPU completely for something else like gaming.

    Metal Messiah. said:
    So with the upgraded NPU, Intel Arrow Lake refresh chips could see a larger die size, expanding by 2.8mm compared to current-gen Arrow Lake chips.
    Arrow Lake Refreshin' the NPU?! What happened to 8 P-cores + 32 E-cores?
    Reply