Intel to deliver a 5X jump in AI in just two chip generations — Panther Lake in 2025 will double the AI performance over Arrow Lake and Lunar Lake

CPU
(Image credit: Shutterstock)

During Intel's Q4 earnings call, CEO Pat Gelsinger (via Seeking Alpha) revealed that the company is targeting a double uplift in AI performance for Panther Lake over next-generation Arrow Lake and Lunar Lake processors, which will compete against the best CPUs. Gelsinger also confirmed that Xeon Clearwater Forest processors are already in the fabs.

"The Core Ultra platform delivers leadership AI performance today with our next-generation platforms launching later this year, Lunar Lake and Arrow Lake tripling our AI performance. In 2025 with Panther Lake, we will grow AI performance up to an additional 2x," stated Gelsinger in Intel's Q4 earnings call.

Although Panther Lake will bear an impressive amount of AI performance compared to Meteor Lake, most of this 5X performance improvement comes from Arrow Lake and Lunar Lake. The two CPUs will be three times faster than Meteor Lake, and then Panther Lake will double its performance, leading to 5X better performance than Meteor Lake.

Gelsinger also reiterated the 2024 launch date for Arrow Lake and Lunar Lake and that Panther Lake would still arrive in 2025. The fast-paced release schedule (assuming Intel can stick to it) combined with these aggressive performance gains in AI every single generation means Intel will likely become a potent contender in consumer AI hardware. It certainly makes the 60% gain AMD achieved from its Ryzen 7040 series Phoenix APUs to its Ryzen 8040 series Hawk Point APUs a bit quaint.

"We are first in the industry to have incorporated both gate-all-around and backside power delivery in a single process node, the latter unexpected two years ahead of our competition. Arrow Lake, our lead Intel 20A vehicle will launch this year. Intel 18A is expected to achieve manufacturing readiness in second half '24, completing our five nodes in four year journey and bringing us back to process leadership. I am pleased to say that Clearwater Forest, our first Intel 18A part for servers has already gone into fab and Panther Lake for clients will be heading into Fab shortly," confirmed Gelsinger.

That AI performance is all on top of what is expected to be at least decent uplifts in traditional CPU performance and efficiency, thanks in part to Panther Lake's usage of the Intel 18A node. 18A should deliver a significant upgrade over Arrow Lake's 20A process and Lunar Lake's TSMC 3nm process, but it's hard to say by how much yet. It's also difficult to say how impactful architectural improvements will be.

Meanwhile, Clearwater Forest has already been taped out and in fabs. According to a new Linux patch, the Xeon chips will reportedly leverage Darkmont E-cores.

Given Panther Lake was only officially unveiled last September, other details on the CPU are scarce. It is more likely to be a successor to high-power Arrow Lake and an efficiency-focused Lunar Lake. It was first pitched as a Lunar Lake successor. Still, Gelsinger's comparison of Panther Lake to Arrow Lake and leaks suggesting there may be a Panther Lake-S imply that the upcoming CPU will handle the whole market, just as Alder Lake and Raptor Lake did in 2021 and 2022.

Matthew Connatser

Matthew Connatser is a freelancing writer for Tom's Hardware US. He writes articles about CPUs, GPUs, SSDs, and computers in general.

  • jasonf2
    Well the good news is that there might be a chip by late 2025 that will be able to run the minimum hardware specs for Windows 12 OS AI mode.
    Reply
  • endocine
    Intel: "at some point in the future our products may be competitive, with something, at least that's what we are claiming even though we can't deliver anything on time"
    Reply
  • gg83
    Do we think Intel was able to dump enough r&d to be able to surpass tsmc? They have been slacking for a while now. I hope it was on purpose.
    Reply
  • wr3zzz
    Articles like this really need to be better edited. While 5X gain sounds impressive the real world impact won't matter much if it's something like gaming GPU going from 2fps to 10fps while in 720P.

    So what would the 5X "AI" performance gain do for me?
    Reply
  • bit_user
    "We are first in the industry to have incorporated both gate-all-around and backside power delivery in a single process node, the latter unexpected two years ahead of our competition.
    I think this is a transcription error. It'd make much more sense if he said "... the latter an expected two years ...".
    Reply
  • bit_user
    wr3zzz said:
    Articles like this really need to be better edited. While 5X gain sounds impressive the real world impact won't matter much if it's something like gaming GPU going from 2fps to 10fps while in 720P.
    The GPU analogy is a good one.

    Consider that Meteor Lake's NPU isn't even as fast or capable as its 128 EU iGPU:
    Source: https://www.tomshardware.com/news/intel-details-core-ultra-meteor-lake-architecture-launches-december-14
    While it would be impressive to boost it 5x, that's probably not even going to equal the performance of an Arc A770 that's not even using its XMX units. Still, good for an integrated IP block, but not stellar.

    Also, don't plan on using it for training. I expect it'll be heavily biased towards inference.
    Reply
  • bluvg
    If it's 3x to Arrow/Lunar Lake, then Panther Lake is additional 2x, wouldn't that be 6x Meteor Lake? The other way of reading it--(3xMTL) plus (2xMTL)--is kind of tortured wording, or maybe he just spoke inaccurately. "An additional 200%" seems clearer if he meant 5x MTL.
    Reply
  • bit_user
    bluvg said:
    If it's 3x to Arrow/Lunar Lake, then Panther Lake is additional 2x, wouldn't that be 6x Meteor Lake? The other way of reading it--(3xMTL) plus (2xMTL)--is kind of tortured wording, or maybe he just spoke inaccurately. "An additional 200%" seems clearer if he meant 5x MTL.
    My guess was that they were rounding up, in each case, but the final speedup rounds to 5x. For instance, if the speedup in Arrow/Lunar lake is 2.8x, and then you get a 1.8x speedup to Panther Lake, that would result in a final speedup of 5.04x, which rounds down to 5x.

    Still, it is a little weird of him to say it like that. Especially for a former engineer. Maybe he had been saying "nearly 3x" and "nearly 2x", but at some point the "nearly" got dropped?
    Reply
  • yeyibi
    Small letter: It will support 5 extra GPU slots for nvidia cards*.

    *5 nvidia cards will be able to run ChatGPT8, but we will charge a subscription by pretending that it needs the cloud to run.
    Reply