With the release of its new Core Ultra chips for laptops, Intel has announced the beginning of a new era of “artificial intelligence computers.” Intel’s chips, codenamed Meteor Lake, contain an NPU, or neural processing unit, to accelerate artificial intelligence tasks.
However, in addition to the AI accelerator, the new chips have many other conceptual innovations. In the case of Meteor Lake, Intel is moving away from a monolithic die layout in favor of a tile structure with several crystals combined using Intel Foveros technology. Intel is also using TSMC chips in its processors for the first time.
- CPU crystal: Intel 4
- 3D Faveros matrix: Intel 16
- GPU crystal: TSMC N5
- SoC crystal: TSMC N6
- I/O crystal: TSMC N6
First-generation Core Ultra processors support the latest connectivity standards, including WiFi7, Thunderbolt 4 (but not Thunderbolt 5), and DisplayPort 2.1. In addition, these processors work with DDR5-5600 and LPDDR5(x)-7467 RAM.
The Compute chip contains high-performance P-Cores with Redwood Cove architecture and energy-efficient E-Cores (Crestmont) that share an L3 cache, as well as a separate dual-core LP E-Core cluster located on the SoC chip.
Intel admits that Meteor Lake will not be a record-breaker in terms of CPU performance. As indicated on the manufacturer’s slides, the new Core Ultra 7 165H will have an advantage over the Core i7-1370P (Raptor Lake) under multi-threaded workloads, but single-threaded performance will be slightly lower. However, in both cases, the performance of Intel chips should be higher than that of AMD’s Ryzen 7 7840U. Although it should be noted here that the relative results are based on the SPECrate2017_int_base test.
The most important update concerns integrated graphics. Intel is updating the Xe-LP architecture to Xe-LPG, which is the basis for the Alchemist series of discrete GPUs (XeHPG), and will offer a similar set of features and capabilities. The GPU chip supports DirectX 12 Ultimate, hardware ray tracing acceleration, XeSS scaling, and advanced encoding capabilities such as AV1, H265, and VP9.
The integrated GPU architecture is based on 8 Xe cores that will be accelerated to 2.35 GHz. Intel provides some internal benchmarks that claim the new graphics will be up to two times faster than Xe-LPG in Intel’s 13th Generation Core CPUs while consuming 28W. Although, as you can see in the chart, the average advantage is somewhat more modest.
Intel assures that Meteor Lake chips are ready to compete with the integrated Radeon 780M graphics based on RDNA3 in terms of GPU performance. So in Full HD mode with average graphics quality, Core Ultra 7 165H can be 5-10% faster than Ryzen 7 7840U. For sure, there will be nuances in the details, and the balance of power will depend on the specific project. But this is a remarkable progress in a competition that has long been dominated by AMD solutions.
Support for XeSS intelligent scaling technology will also be very useful for increasing fps. The average performance increase of 39% can already allow you to try even quite resource-intensive games without a discrete graphics card.
One of the key features of Meteor Lake is the presence of a hardware AI accelerator. According to Intel, the NPU module provides processing performance of AI algorithms up to 34 TeraOPS. The new chips will be 70% faster in generative AI workloads compared to Raptor Lake.
Intel Core Ultra processor lineup
With the introduction of Meteor Lake, the company is introducing an updated processor labeling system. The Ultra identifier will be used for the new chips. During the announcement, 11 models were presented – five of them belong to the Core-H series and six to the more economical Core-U models.
The flagship model, Core Ultra 9 185H, has 16 cores and 22 threads, including six P-Cores (up to 5.1 GHz), eight E-Cores (up to 3.8 GHz), and two LP E-Cores separately located on the I/O chip. At the same time, the chip is equipped with a GPU with 8 Xe-core cores with a maximum acceleration of up to 2.35 GHz. This model is characterized by a base power (PBP/PL1) of 45 W and a maximum turbo power (MTP/PL2) of up to 115 W.
The Core Ultra 7 165H and Core Ultra 7 155H have a similar configuration to the Core 9, but with different clock speeds and power specifications. The 165H model reaches a maximum processor clock speed of 5.0 GHz, while the 155H reaches 4.8 GHz. GPU operating frequencies are 2.3 and 2.25 GHz, respectively. These chips are designed for a TDP of 28W by default, but Intel allows maximum power up to 64W or even 115W, depending on the laptop implementation.
The Core Ultra 5 135H and Core Ultra 5 125H processors are lightweight versions with four high-performance P-cores instead of six in the Core Ultra 7/9. The number of E-Cores and LP E-Cores remained unchanged, although the operating frequencies of all computers are slightly lower. Also, the number of Xe cores of the graphics core, which operates at frequencies up to 2.2 GHz, has been reduced to 7.
Intel has also announced more economical Core Ultra 100U chips with a base TDP of 15 W and the possibility of increasing it to 57 W. All CPUs in this line will have a functional formula of 2+8+2 cores. That is, the number of productive cores has been reduced to two, while the number of energy-efficient cores remains unchanged. The GPU is also simpler, with 4 Xe cores and operating frequencies of 1.75-2 GHz. In the new U-line, the manufacturer also provides even more energy-efficient CPU versions with a basic package of 9 W and a maximum of 30 W. To reduce power consumption, the operating frequencies of the CPU/GPU and supply voltages of the units will be limited. It should be noted that regardless of the configuration, all the announced processors are equipped with an AI NPU accelerator.
Intel’s official specifications indicate that the flagship of the series, Core Ultra 9 185H, as well as the 9-watt Core Ultra 7 164U and Core Ultra 5 134U will be available in Q1 2024, while the rest of the models are already available to manufacturers and should appear on store shelves as part of new laptops.
Loading comments …