Intel Launches Core Ultra ‘Meteor Lake’ Chips In AI PC Push Against Rivals

Intel says the CPU, GPU and neural processing unit in the new Core Ultra chips for ultrathin laptops are made to accelerate a variety of workloads. The processors are expected to power more than 230 laptop designs from Dell, HP, Lenovo and a slew of other OEMs.

Intel is mounting a fresh challenge against rivals in the emerging AI PC category with its newly launched Core Ultra “Meteor Lake” processors for ultrathin laptops, packing a CPU, GPU and, for the first time, a neural processing unit to handle low-power AI tasks.

Launched on Thursday, Core Ultra is an updated brand name meant to represent Intel’s “most advanced processors” for PCs, and the first series of chips to carry the name are expected to power more than 230 laptop designs from Dell Technologies, HP Inc., Lenovo along with an additional 30-plus OEMs.

Select consumer laptops are available starting today, and commercial designs based on Intel’s vPro management and security platform are “coming soon.”

Intel CEO Pat Gelsinger has previously promised that the Core Ultra chips will “usher in the age of the AI PC,” which the chipmaker and other major PC players believe will transform the industry and drive a fresh wave of refresh opportunities, as CRN reported for AI PC Week.

“I actually believe that the real client use cases will come from commercial and enterprise use cases where companies are going to drive productivity gains,” Intel Chief Commercial Officer Christoph Schell said in an interview for the CRN series.

The Santa Clara, Calif.-based company is promoting the Core Ultra chips as part of its “AI everywhere” strategy, which involves accelerating AI workloads from the edge to the cloud and is becoming an important element of Gelsinger’s broader comeback plan.

“Intel is developing the technologies and solutions that empower customers to seamlessly integrate and effectively run AI in all their applications — in the cloud and, increasingly, locally at the PC and edge, where data is generated and used,” Gelsinger said in a statement.

Intel Core Ultra Basics: CPU, GPU And NPU

Within the Core Ultra series, the semiconductor giant claimed that the lineup’s flagship chip, the 16-core Core Ultra 7 165H, is the “most efficient x86 processor” running at 28 watts and provides “CPU core performance leadership” for ultrathin systems.

The company also said that the Core Ultra H-series processors—the most power-hungry in the family where its flagship chip resides—come with “world-class GPU performance for ultrathin PCs,” carrying the same Intel Arc brand as the company’s discrete graphics products.

The lower-power U-series chips, on the other hand, come with less graphics horsepower and carry the generic “Intel Graphics” brand for their integrated GPU.

The Core Ultra series marks the first time Intel has put an NPU inside a PC chip, following similar moves by Qualcomm, Apple and AMD, all of which currently have processors in the market with an NPU or equivalent component for low-power AI and machine learning workloads.

Like AMD, Intel sees the CPU, GPU and NPU as complementary accelerators for AI workloads with different requirements, but it brands the NPU as “Intel AI Boost.”

In Intel’s view, the CPU is best for low-latency AI workloads that require a quick response while the GPU is ideal for AI-accelerated content creation applications that need high-throughput processing. The NPU, on the other hand, is a low-power accelerator made to run and offload sustained AI workloads from the CPU and GPU to reduce battery use.

But while Intel is the last major chip company to commercialize a client chip that contains an NPU alongside the CPU and GPU, the x86 behemoth said its investment in consumer and commercial AI applications is “unmatched.”

The company also said it’s the leader for enabling “consistent AI performance” across the CPU, GPU and NPU, all of which support the 8-bit integer (INT8) and half-precision floating-point (FP16) data types commonly used by AI applications.

Intel Core Ultra Specs And Features

At launch, the Intel Core Ultra processors are split into two types: one that provides high performance in exchange for high power requirements and another that requires less power but provides less performance within ultrathin laptops.

On the higher performance side, the H-series consists of Core Ultra 7 and 5 chips that go up to 16 cores and a 5.1GHz frequency, feature an Intel Arc GPU and have a base power range of 28W, which OEMs can configure up to 64W or 115W.

Intel plans to launch a Core Ultra 9 chip in the first quarter of next year that will increase the maximum turbo frequency to 5.1GHz and the base power requirement to 45W. The only max turbo power configuration available to OEMs is 115W for this chip.

The U-series, on the other hand, consists of Core Ultra 7 and 5 chips that go up to 12 cores and a 4.9GHz frequency, feature integrated graphics with the generic Intel Graphics brand and have a power range of 15W to 57W.

The company plans to release additional Core Ultra 7 and 5 chips that come with a lower power range of 9 watts to 30 watts in the first quarter of next year.

For the first time in an Intel client processor, the company is making use of two low-power efficient cores, also known as E-cores, in addition to eight regular E-cores and the variable number of performance cores, also known as P-cores, across the stack.

All but two Core Ultra processors support a maximum memory capacity of 96GB of DDR5 with speeds up to 5600 MT/s or 64GB of LPDDR5 with speeds up to 7467 MT/s. The two upcoming Core Ultra 7 and 5 chips with a 9W base power, on the other hand, will only support a maximum memory capacity of 64GB of LPDDR5 with speeds up 6400 MT/s.

Depending on OEM enablement and operating support, the processors support either Wi-Fi 7 or Wi-Fi 6E connectivity. They also support Thunderbolt 4 connectivity, with up to 40 Gbps of bi-directional throughput per port, as well as Bluetooth 5.4.

Intel Core Ultra H-Series SKU Table

Intel Core Ultra 9 185H: 16 cores/22 threads; 6 P-cores/8 E-cores/2 LP E-cores; 24M Intel Smart Cache; 5.1GHz P-core max turbo clock; 3.8GHz E-core max turbo clock; Intel Arc GPU with 2.35GHz max clock and 8 Xe cores; 2x Gen3 compute engines for Intel AI Boost NPU; 45W base power; 115W max turbo power. Q1 2024 availability.
Intel Core Ultra 7 165H: 16 cores/22 threads; 6 P-cores/8 E-cores/2 LP E-cores; 24M Intel Smart Cache; 5GHz P-core max turbo clock; 3.8GHz E-core max turbo clock; Intel Arc GPU with 2.3GHz max clock and 8 Xe cores; 2x Gen3 compute engines for Intel AI Boost NPU; 28W base power; 64/115W max turbo power.
Intel Core Ultra 7 155H: 16 cores/22 threads; 6 P-cores/8 E-cores/2 LP E-cores; 24M Intel Smart Cache; 4.8GHz P-core max turbo clock; 3.8GHz E-core max turbo clock; Intel Arc GPU with 2.25GHz max clock and 8 Xe cores; 2x Gen3 compute engines for Intel AI Boost NPU; 28W base power; 64/115W max turbo power.
Intel Core Ultra 5 135H: 14 cores/18 threads; 4 P-cores/8 E-cores/2 LP E-cores, 18M Intel Smart Cache; 4.6GHz P-core max turbo clock; 3.6GHz E-core max turbo clock; Intel Arc GPU with 2.2GHz max clock and 7 Xe cores; 2x Gen3 compute engines for Intel AI Boost NPU; 28W base power; 64/115W max turbo power.
Intel Core Ultra 5 125H: 14 cores/18 threads; 4 P-cores/8 E-cores/2 LP E-cores, 18M Intel Smart Cache; 4.5GHz P-core max turbo clock; 3.6GHz E-core max turbo clock; Intel Arc GPU with 2.2GHz max clock and 7 Xe cores; 2x Gen3 compute engines for Intel AI Boost NPU; 28W base power; 64/115W max turbo power.

Intel Core Ultra U-Series SKU Table

Intel Core Ultra 7 165U: 12 cores/14 threads; 2 P-cores/8 E-cores/2 LP E-cores, 12M Intel Smart Cache; 4.9GHz P-core max turbo clock; 3.8GHz E-core max turbo clock; Intel Graphics with 2GHz max clock and 4 Xe cores; 2x Gen3 compute engines for Intel AI Boost NPU; 15W base power; 57W max turbo power.
Intel Core Ultra 7 164U: 12 cores/14 threads; 2 P-cores/8 E-cores/2 LP E-cores, 12M Intel Smart Cache; 4.8GHz P-core max turbo clock; 3.8GHz E-core max turbo clock; Intel Graphics with 1.8GHz max clock and 4 Xe cores; 2x Gen3 compute engines for Intel AI Boost NPU; 9W base power; 30W max turbo power. Q1 2024 availability.
Intel Core Ultra 7 155U: 12 cores/14 threads; 2 P-cores/8 E-cores/2 LP E-cores, 12M Intel Smart Cache; 4.8GHz P-core max turbo clock; 3.8GHz E-core max turbo clock; Intel Graphics with 1.95GHz max clock and 4 Xe cores; 2x Gen3 compute engines for Intel AI Boost NPU; 15W base power; 57W max turbo power.
Intel Core Ultra 7 135U: 12 cores/14 threads; 2 P-cores/8 E-cores/2 LP E-cores, 12M Intel Smart Cache; 4.4GHz P-core max turbo clock; 3.6GHz E-core max turbo clock; Intel Graphics with 1.9GHz max clock and 4 Xe cores; 2x Gen3 compute engines for Intel AI Boost NPU; 15W base power; 57W max turbo power.
Intel Core Ultra 7 134U: 12 cores/14 threads; 2 P-cores/8 E-cores/2 LP E-cores, 12M Intel Smart Cache; 4.4GHz P-core max turbo clock; 3.6GHz E-core max turbo clock; Intel Graphics with 1.75GHz max clock and 4 Xe cores; 2x Gen3 compute engines for Intel AI Boost NPU; 9W base power; 30W max turbo power. Q1 2024 availability.
Intel Core Ultra 7 125U: 12 cores/14 threads; 2 P-cores/8 E-cores/2 LP E-cores, 12M Intel Smart Cache; 4.3GHz P-core max turbo clock; 3.6GHz E-core max turbo clock; Intel Graphics with 1.85GHz max clock and 4 Xe cores; 2x Gen3 compute engines for Intel AI Boost NPU; 15W base power; 57W max turbo power.

Intel Core Ultra CPU, GPU Performance Claims

For Intel’s performance claims, the company focused on the Core Ultra 7 165H, the most powerful and power-hungry chip currently available.

When it comes to the CPU, Intel said the chip provides up to 11 percent faster multi-threaded performance than AMD’s Ryzen 7 7840U at the same power. The Ryzen 7 chip has a default thermal design power of 28 watts that can be configured anywhere from 15 watts to 30 watts.

In the same graph, Intel showed that the Core Ultra 7 165H can also provide more performance than Apple’s M3 chip at roughly 23W and Qualcomm’s Snapdragon 8cx Gen 3 at 15W.

In terms of single-threaded performance, Intel said the Core Ultra 7 165H is 12 percent faster than the Ryzen 7 7840U. But by the company’s own admission, its previous flagship chip for ultrathin laptops, the Core i7-1370P, is 21 percent faster than AMD’s chip in this area, making its new top performer slower than the previous generation.

Compared to Intel’s previous-generation Core i7-1370P, the Core Ultra 7 165H uses 25 percent less power for Netflix video playback because of the latter’s low-power E-cores.

When put head-to-head with the Ryzen 7 7840U, the Core Ultra 7 uses 48 percent less power for Netflix streaming, 44 percent less power for 4K local video, 79 percent less power for Windows when the desktop idles and 7 percent less power for Microsoft Teams using the three-by-three video call grid and 7 percent less power for web browsing, according to Intel.

In comparison to the Ryzen 7 7840U in the content creation arena, Intel said its second-best chip available now, the Core Ultra 7 155H, is 31 percent faster for the UL Procyon video editing benchmark, 41 percent faster for the Adobe Premiere Pro PugetBench video editing test and 19 percent faster for the Adobe Lightroom PugetBench tool.

When it comes to gaming performance enabled by the GPU, Intel said the Arc GPU inside its Core Ultra 165H provides up to two times faster performance compared to the previous-generation Core i7-1370P depending on the game.

The company also compared the gaming performance of two laptops with the Core Ultra 7 165H against two notebooks with the Ryzen 7 7840U. While one Core Ultra 7 laptop and one Ryzen 7 laptop tied for performance, the other Core Ultra 7 system outperformance the first Ryzen 7 computer by 5 percent and the other Ryzen 7 machine by much more.

Intel said the latest version of its upscaling technology, XeSS, can improve gaming performance even more, bringing the average frames per second of the title Ghostrunner 2 from 62 without upscaling turned on to 82 with the technology activated.

Core Ultra AI Performance Claims, ISV Support

With the Core Ultra processors, Intel is promising to enable hundreds AI features from ISVs and power more than 100 million AI PCs through 2025.

The company previously announced that it is working with more than 100 ISVs to enable over 300 AI-accelerated software features on Core Ultra-based PCs as part of what it has called the “industry’s first” AI PC Acceleration Program.

The program provides ISVs with a bevy of engineering, design and marketing resources to take advantage of Intel’s chips and software tools for AI applications on PCs, and it includes a mix of big and small ISVs, including Adobe, Zoom Video Communications and Cisco Systems.

The applications with AI-enabled features supported by the Core Ultra chips include the Microsoft 365 office application suite, the Audacity audio editing tool, the GIMP image editing tool and a new generative AI tool called Superpower.

With Intel’s focus on AI acceleration across the CPU, GPU and NPU, the company demonstrated how each component can significantly accelerate workloads.

Using the A1111 web interface for the Stable Diffusion text-to-image generation, the Core Ultra 7 165H is 70 percent faster than the previous-generation Core i7-1370P, thanks to the application’s built-in GPU offload capabilities, according to Intel. Compared to the Ryzen 7 7840U, the Core Ultra 7 165 is 3.2 times faster.

As for other GPU-enabled AI capabilities, Intel said the Core Ultra 7 165H is 5.4 times faster for GIMP’s Stable Diffusion plugin, 70 percent faster for Adobe Premiere Pro’s color grade/scene edit/export workflow and 20 percent faster for DaVinci Resolve’s render/AI mask/export workflow compared to the Ryzen 7 7840U.

When it comes to the NPU’s offload capabilities, Intel said that accelerator enables the Core Ultra 7 165H to use 38 percent lower power in Zoom video calls versus the Core i7-1370P. And with the UL Procyon AI benchmark, the NPU enables 2.5 times greater power efficiency.

Intel said the Core Ultra processors support versions of more than 80 AI models, including BERT, Stable Diffusion, Llama and Dolly. Among these supported model iterations is a 7-billion-parameter Llama 2 model.

What makes Intel’s support for AI applications possible is its OpenVINO toolkit,” which “enhances AI performance by routing workloads to the right compute engine” using “minimal code changes,” according to the company.