Generative Data Intelligence

Intel says Lunar Lake’s NPU is 4x faster than Meteor

Date:

Intel Vision Intel claims its forthcoming Lunar Lake CPUs will have over 100 Tera Operations Per Second (TOPS) of AI performance – 45 of them from its neural processing unit (NPU).

Intel CEO Pat Gelsinger has previously claimed that Lunar Lake (and its higher-performance counterpart, Arrow Lake) would have triple the AI performance of Meteor Lake – Intel’s current offering to power the “AI PC.” Since Meteor Lake has up to 34 TOPS total from its NPU, GPU, and CPU cores, that would put Lunar Lake at roughly 102 TOPS – implying that “over 100 TOPS” isn’t too much beyond triple digits.

Intel’s breakdown of Lunar Lake’s TOPS by component reveals a 45 TOPS NPU, with the lion’s share of the remaining 55 TOPS or so likely coming from the GPU. For comparison, 11 of Meteor Lake’s 34 TOPS are from the NPU – meaning Lunar Lake’s NPU has a little over four times the processing power of the part found in Meteor Lake processors.

In a presentation at its Vision conference on Tuesday, Intel didn’t specify what tech delivered such a substantial improvement in NPU performance – but upgrading to a new node (rumored to be either Intel’s 18A or TSMC’s N3B) is probably one factor.

Notably, this 45-TOPS NPU satisfies Microsoft’s definition of an “AI PC,” which calls for 40 TOPS of NPU performance.

The Lunar Lake NPU is substantially faster than the latest AMD XDNA NPU paired with Ryzen 8040 Hawk Point APUs, which clocks in at 16 TOPS, slightly edges out Hailo’s 40-TOPS Hailo-10 add-in chip, and just matches the Hexagon NPU on Qualcomm’s Snapdragon X Elite.

lunar lake

Pat Gelsinger praises Lunar Lake to the stars – Click to enlarge. Source: Intel Vision

Intel did not specify the data type it used to calculate Lunar Lake’s TOPS count.

Most orgs in this space default to INT8, but we’ve asked Intel to clarify its numbers. INT4 trades more TOPS for less accuracy, but using INT4 to measure the NPU’s performance wouldn’t necessarily be a bad thing – it would lower memory requirements to run local AI models. But it would mean the performance comparison of Meteor Lake and Lunar Lake is not really apples to apples.

Of course, graphics remain an important part of the total TOPS for Lunar Lake, which should have roughly 55 TOPS of GPU-derived power. That’s over double the performance of Meteor Lake’s integrated Arc GPU, implying that Intel is also making substantial strides in graphics with Lunar Lake – at least in respect to AI workloads.

Gelsinger claimed Intel’s next generation of CPUs, dubbed Panther Lake, will again double AI performance – presumably to a total of 200 or so TOPS. Given that Lunar Lake’s AI performance is almost equally divided between the NPU and the GPU, it’s possible that Panther Lake may feature an NPU capable of hitting 100 TOPS. That’s assuming Intel sees the NPU as the key part driving AI performance. ®

spot_img

Latest Intelligence

spot_img

Chat with us

Hi there! How can I help you?