Lunar Lake’s NPU performance tops Meteor Lake as Intel over the Moon • The Register

WhatsApp Group Join Now
Telegram Group Join Now
Instagram Group Join Now

Intel Vision Intel claims that its upcoming Lunar Lake CPUs will have more than 100 Tera Operations Per Second (TOPS) of AI performance – 45 of them from its Neural Processing Unit (NPU).

Intel CEO Pete Gelsinger has previously claimed that Lunar Lake (and its high-performance counterpart, Arrow Lake) will triple the AI ​​performance of Meteor Lake – powering Intel’s current “AI PC” offering. to give Since Meteor Lake has a total of 34 TOPS from NPU, GPU, and CPU cores, that would put Lunar Lake at around 102 TOPS – meaning that “over 100 TOPS” is no more than three digits.

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

In a presentation at its Vision conference on Tuesday, Intel didn’t specify which tech improved the NPU’s performance so much — but upgrading to a new node (rumored to be Intel’s 18A or TSMC’s N3B) might be one. is an element.

Specifically, this 45-TOPS NPU meets Microsoft’s definition of an “AI PC,” which requires 40 TOPS of NPU performance.

The Lunar Lake NPU paired with the Ryzen 8040 Hawk Point APUs is significantly faster than the latest AMD XDNA NPU, which clocks in at 16 TOPS, slightly outpacing Hailo’s 40-TOPS Hailo-10 add-in chip, and only the NPUx Matches. Qualcomm’s Snapdragon X Elite

Pat Gelsinger admires Lunar Lake for the stars – click to enlarge. Source: Intel Vision

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

Most organizations in this space default to INT8, but we asked Intel to clarify its numbers. INT4 trades more tops for less accuracy, but using INT4 to measure NPU performance isn’t necessarily a bad thing – it will reduce the memory requirements for running native AI models. But that would mean that comparing the performance of Meteor Lake and Lunar Lake isn’t really apples to apples.

Of course, graphics remain a significant portion of Lunar Lake’s total TOPS, which should account for around 55 TOPS of GPU-derived power. That’s double the performance of Meteor Lake’s integrated Arc GPU, which means that Intel is also making a lot of progress in graphics with Lunar Lake – at least for AI workloads.

Gelsinger claims that Intel’s next generation of CPUs, dubbed Panther Lake, will double AI performance again — possibly to a total of 200 or more. Given that Lunar Lake’s AI performance is split almost equally between NPU and GPU, it’s possible that Panther Lake features an NPU capable of hitting 100 TOPS. That’s assuming Intel sees the NPU as a key component driving AI performance. ®

WhatsApp Group Join Now
Telegram Group Join Now
Instagram Group Join Now

Leave a Comment