Shulou(Shulou.com)12/24 Report--

Intel today officially launched the first generation of Intel Core Ultra processors based on the Meteor Lake architecture, the first batch of which covers the Core Ultra 7 and 5 series. This time the Meteor Lake architecture encapsulates the new SoC module, the Intel 4 computing module, the new GPU unit and the IO module through Foveros 3D encapsulation technology, resulting in a significant improvement in energy consumption. Multi-level XPU design brings AI computing power as high as 34TOPS, laying the foundation for the full outbreak of AI PC.

Core Ultra processor platform is a collection of Intel's eight advanced technologies. First, 3D high-performance hybrid architecture, and 3D Foveros packaging technology; second, the use of Intel 4 process computing module, up to 6 performance cores, 8 energy efficiency cores, and 2 low power island energy efficiency cores, which can support a total of 22 threads; third, the main frequency is up to 5.1GHz; fourth, the integration of Intel Arc GPU graphics card, equipped with up to 8 Xe kernel; fifth, support multi-channel AI accelerated processing NPU Sixth, support the most 64GB LPDDR5/5X-7467 and 96GB DDR5-5600 memory; seventh, support Lightning 4 peripheral interface; eighth, integrated Intel Wi-Fi 6e wireless network card. Next, let's take a look at many of the new features of Intel Core Ultra architecture.

Architecture feature Meteor Lake is Intel's first processor product using Intel 4 process technology. Intel 4 is the most important link in Intel's "four years and five nodes" process development plan. The mass production of Intel 4 marks Intel's very smooth entry into the EUV era. At the same time, the mass production of Intel 4 also laid the foundation for the development of Intel 3 and subsequent Intel 20A and Intel 18A.

Intel 4 achieves twice the transistor density compared to Intel 7, while this generation of manufacturing process optimizes the CPU high-performance logic library, further achieving a higher density of metal layers and achieving more efficient power supply. From a process point of view, it has achieved a 20% improvement in energy efficiency compared to Intel 7 and Intel 4. For the parsing of Meteor Lake architecture here, you can refer to the previous content, which will not be dwelt on here.

One of the important design goals of Meteor Lake is to improve the energy efficiency of processors. The new P-Core and E-Core micro-architecture improves the performance and execution efficiency per watt. At the same time, through the separate modular design, the SoC module of the core Ultra processor introduces the low power island LPE-Core, while the memory controller, video codec, display processing unit and NPU are all integrated on the low power island to achieve a low power operation processing unit.

The modular design also realizes the separate power supply control of the module, which can effectively improve the energy consumption management of the whole processor. On the hardware, with the operating system, the hardware thread scheduler realizes more advanced third-order thread scheduling management, which effectively helps the operating system to realize the thread scheduling and execution of the logical processor core. From Raptor Lake to the current Meteor Lake, Intel hardware thread scheduler is optimized for processor characteristics in each generation, which ensures that Meteor Lake can achieve faster and more energy-efficient computing and running.

Intel Core Ultra processor in the SPEC 2017 related tests, Core Ultra 7 165H compared to the previous generation of i7-1370P, at 28W power consumption, Core Ultra has a pretty good performance. Compared with competitive products, Core Ultra can have higher performance with more power consumption. The entire CPU, especially in the lightweight PC, has an 11% lead in multithreading performance compared to the competitor in a similar power consumption range.

In the power consumption part, Core Ultra 7 165H runs the Netflix video recording power test, and we can see that the overall power consumption is about 1150mW. Due to the addition of the low power island of the core Ultra processor, it can save 25% power consumption compared to the previous generation of core i7-1370p.

In other scenarios, Core Ultra 7 165H saves more power than competitive products. For example, Windows idle standby has a very significant power saving compared with busy status, up to 79% power saving, such as live Netflix or 4K local video playback, and more than 40% power saving. General daily usage scenarios, Microsoft Teams or web browsing, also save about 7% of the power consumption.

The computing unit goes back to the computing module of Core Ultra. The two main cores are P-Core and E-Core. The E-Core micro-architecture here has been upgraded from the previous generation of Cracemont to Crestmont. Under the condition of ensuring the lowest power consumption, the improvement in architecture further improves the throughput capacity of E-Core and enhances the computing power of AI. In this way, the execution of the VNNI instruction set of E-Core has also been greatly improved.

The P-Core micro-architecture has been upgraded from the previous generation of Golden Cove to Redwood Cove, which greatly improves energy efficiency on the premise of further improving performance. P-Core and E-Core together can make the product design at its best, whether it's a very thin notebook or a high-performance notebook. LPE-Core low power island is located in the SoC module, the architecture and E-Core architecture is the same, are based on Crestmont architecture, so LPE-Core can provide complete computing power at lower energy consumption.

The micro-architecture evolution of E-Core continues to improve the power of IPC and make its computing power higher. In addition, Intel has improved its peak prediction ability, whether it is E-Core or P-Core, to make its instruction execution more efficient. E-Core, P-Core, and LPE-Core are all provisioned through Intel Thread Scheduler.

The ability to execute VNNI instructions has been further improved on E-Core in preparation for AI acceleration. Both E-Core and P-Core support the same instruction set. If tasks can be executed on P-Core, they can be executed on E-Core or even LPE-Core. Whether it is integer operations, floating-point operations, including a variety of AI tasks are possible. In this way, no matter any computing task, including AI running, can be controlled and dynamically assigned by Intel thread scheduler, real-time conversion can be done on E-Core, P-Core and LPE-Core, when performance needs response speed, it moves up to P-Core, and when power consumption needs to be reduced, it moves up to E-Core or even LPE-Core. Therefore, instructions can be successfully executed on all Core.

P-Core is a traditional large core. While continuously improving performance, throughput, internal bus, external bus, bandwidth, including all buffers, Intel further optimizes performance efficiency and reduces the power consumption of P-Core execution on Meteor Lake. Meteor Lake has a great advantage over the previous generation in energy consumption control. The whole P-Core, E-Core and LPE-Core are distributed dynamically and in real time under the Intel thread scheduler. Intel thread scheduler has evolved to the third generation, it is highly integrated with the operating system, external software adaptation has been done, so Meteor Lake has been able to provide better energy efficiency and power consumption performance.

Intel Core Ultra processor multithreaded performance has been greatly improved, compared to Core i7-1370P, Core Ultra 7 165H has 8% performance improvement. Compared with competitive products, 11% of the performance is in the lead.

Single threaded part. Compared with the competitive products, Core Ultra has a good performance lead, compared with the competitive products, the performance lead is about 12%. Of course, the dominant frequency will be slightly higher on the Intel 13th-generation Core processor, compared with the Core Ultra, but it still has a very good lead over the competition.

With similar hardware configurations, Intel Core Ultra 7 155H has a clear lead over previous i7-1360p and competitive products in UL Procyon video clips, PugetBench for Premiere Pro and PugetBench for Lightroom, with 31 per cent performance leadership in video editing, 41 per cent performance leadership in Premiere Pro and 19 per cent performance leadership in Lightroom.

GPU unit take a look at the GPU section, this generation of Intel Core Ultra integrates Intel Arc GPU, unlike the 13th generation, this generation uses a new Xe LPG architecture and supports DX12 Ultimate features, that is, set display also supports new graphics features such as hardware ray tracing and mesh shading.

In terms of codec capability, Arc GPU not only supports H.265 and H.264, but also supports AV1 coding, up to 8K 10bit HDR. In the part of display engine, this generation of set display supports 4 display channels at the same time. From the display interface specification, HDMI2.1 and DP 2.1are all supported by Core Ultra. The maximum bandwidth of these two specifications is 20Gbps, and eDP 1.4b is also supported.

Two other advanced features are the support of the DP4A engine and XeSS,DP4A AI instructions to speed up AI computing faster. XeSS supersampling improves the fluency and clarity of games through AI.

Specifically, Core Ultra has a 100% improvement over the previous generation of i7-1370p in the game Baird Gate 3. In other mainstream games, such as "biochemical crisis", "isolated Island shock", "Watchman Vanguard" and so on, FPS has greatly improved compared with the previous generation. These games are mainly on ultra-thin laptops, several mainstream games on the market, all tested in 1080p picture quality.

In the comparison of several other PC products, we see that Core Ultra 7 165H has an average lead of 5% over the competition.

This time, the XeSS supersampling technology based on AI has also brought an improvement on the drawing paper for the game. We know that if you want to improve the picture quality in the game, the frame rate will be sacrificed and the fluency will be greatly reduced. Without changing the hardware configuration, XeSS first degrades the resolution of each frame through its own AI instructions, such as rendering from 1080p to 540p, which ensures higher frame rate fluency, then renders every low-pixel frame first, and then upgrades to a higher pixel resolution through XeSS technology. In this way, the resolution is raised to a higher level while ensuring the same frame rate. It ensures not only the fluency of the frame rate, but also the pixel quality of the picture, because the original target pixel required by the user is degraded first, and after the degradation, the still high frame rate can be guaranteed, and then the pixel quality is upgraded back through XeSS technology, which ensures the picture quality of the game on the premise of ensuring the frame rate.

After opening XeSS, the increase will be different in different games. Some popular games, such as "Killer", "Dragon 7", "call of Duty", "Tomb Raider" and "Cyberpunk" have significantly improved the frame rate with the help of XeSS.

In Ghost Walker 2, combined with Core Ultra and Arc graphics cards and XeSS technology at 1080p, the same resolution is three times higher than that of the previous Core processor, and it will save power at the same frame rate.

AI processing power AI is undoubtedly the hottest topic this year, Intel Core Ultra processor can be said to be a new generation of platform for AI PC scenarios. Core Ultra has three XPU--CPU, GPU and NPU, all of which can be used to do AI processing. At present, the overall computing power of Core Ultra can now reach 34TOPS. Intel recommends that users choose different architectures to handle AI workloads, and this time Core Ultra introduces a new NPU, which is characterized by very efficient and power-efficient handling of AI workloads because it is a processor designed specifically for AI workloads.

When opening a video conference or watching streaming media, because the meeting time is longer and the viewing time is longer, the efficiency improvement can often achieve more obvious battery savings, so NPU is more suitable for long-term use. For example, CPU,CPU also has a special AI acceleration instruction set, such as VNNI, which can be used specifically to speed up AI workload processing. CPU is recommended for running delay-sensitive applications such as voice control and speech recognition. Compared with other NPU and GPU, CPU takes the shortest time to start. GPU can also handle AI, which is characterized by large bandwidth throughput. When users need a large amount of data, they can use GPU to do it. Through the cooperation of 3 XPU, we can reasonably deal with complex tasks and get the best energy efficiency performance.

Intel has done a lot to promote the development of AI PC. At present, Intel plans to put more than 100 million computers with AI accelerator, that is, NPU, on the market by 2025. At present, there are a large number of partners in the Intel ecosystem developing AI applications, or using AI to reshape existing applications. According to data provided by Intel, Intel currently has more than 100 ISV partners and more than 300 applications with AI features. In addition to hardware, Intel also has a unified software interface such as OpenVINO, which encapsulates different underlying hardware architectures into one through OpenVINO, and can even deploy AI load from CPU to NPU. This is also a framework widely welcomed by the industry.

Core Ultra provides good XPU computing power. Compared with other hardware platforms, Intel Core Ultra brings a huge improvement in AI performance, both mainstream software and local Stable Diffusion tools have obvious advantages.

The NPU in Intel Core Ultra is characterized by low power consumption and efficient processing. During Zoom video conferencing, 38% power consumption can be saved by transferring these special effects from background removal to NPU. UL Procyon AI Benchmark also has 2.5 times the savings in power consumption, which makes sense for notebook platforms.

Core Ultra processor with OpenVINO can support a variety of data types. At this stage, many complex models often need to be quantified when running to the front end, and it is difficult for FP32 to run to the front end, so now it can be quantified, and Intel Core Ultra can also support it very well. For example, Intel Core Ultra for 8-bit plastic surgery, 16-bit floating point and other quantitative support is very perfect, whether CPU, GPU or NPU, can well support these quantitative data types.

At present, there are many locally generated AI models on the market, such as BERT, LLaMA2, GLM 2, GLM 3, etc., and Ventures LCM, etc., Intel Core Ultra is well supported. Through Intel Core Ultra, it is possible to quickly implement generative AI applications locally, which can be implemented on notebooks without bulky independent graphics cards, greatly reducing the threshold for the use of AI.

At present, Intel has worked with a lot of ISV, through tools like OpenVINO, these ISV applications can effectively take advantage of Intel's latest features, such as integrated graphics card, NPU. In addition, many domestic large language model manufacturers cooperate with Intel under the premise of compliance, and adapt to the Core Ultra platform. In terms of AI cooperation, Intel partners are also growing rapidly, and the number of these ISV is expected to grow to 100 by the first half of 2024.

Intel Core Ultra processor product parameters next, take a specific look at the features and parameters of Intel Core Ultra processor products. The new Core Ultra processor product provides a maximum processor specification of 6x8x2. Intel hardware thread scheduler is also optimized according to the new hardware scheduling mode.

Integrated with the new Xe LPG engine, there has been a big jump in the set show. The low-power island of the modular design has a built-in NPU. On the memory specification, it supports LPDDR5/5X-7467 and DDR5-5600 specifications, with 1 × 8 PCle 5.0high-speed channel and 20 channels equivalent to the original CPU directly connected PCIe. It has 4 lightning 4 interfaces and 8 PCle channels equivalent to the original chipset, supporting Wi-Fi 6e and the latest Wi-Fi 7 wireless network card. At the same time, there is also a built-in image processing unit to provide a high-quality camera solution.

Intel has been providing high-speed connection solutions for notebook platforms, including Lightning 4, Wi-Fi 6e and the latest Wi-Fi 7. Intel Killer network solution, based on the new artificial intelligence connection optimization solution, can ensure the smoothness and stability of network connection. At the same time, Intel wireless network card includes Bluetooth part, supports the new 5.4 Bluetooth specification, and can provide low-power, high-quality audio connection.

The full family of Intel Core Ultra processor models and parameters are as follows. Intel will be the first to launch Intel Core Ultra H at this month's event, and other models will be released in early 2024. We can see that the highest model of Intel Core Ultra processor launched this time is Core Ultra 7 165H, and its parameters are more detailed than previous generations. Its basic power consumption is 28W, and the maximum Turbo power consumption is 65W or 115W. It also lists the main frequency information of P-Core and E-Core in detail. P-Core Rui frequency is 5.0GHz core core frequency is 3.8GHz.

After the official launch of Intel Core Ultra, starting early next year, it will launch more than 230 Meteor Lake-equipped computer products in partnership with more than 35 OEM, and will be released globally through more than 30 top retailers.

New Intel Evo Certification Standard

After talking about processors, let's take a look at the further refinement of Intel's Evo certification standard. Intel Evo certification is an excellent mobile notebook certification standard established by Intel, which leads the PC ecosystem to promote platform innovation and provide users with a better notebook experience by working with the world's top computer manufacturers in the industry. Notebooks that have passed the strict certification of Intel Evo platform will carry the Intel Evo logo to facilitate consumer identification and purchase. To some extent, Intel Evo certification is a system that tries to improve the user experience from all angles.

The release of Intel Core Ultra processor also means another iterative improvement of Intel Evo certification standard, further deepening the experience details of notebook.

Specifically, it includes fast wake-up capabilities, different levels of platform-related technologies based on performance and response, practical battery life, fast charging connection technology, appearance innovation, cross-device collaboration, and sustainability.

Here, Intel Evo certification pays more attention to the fact that mobile performance indicators will not be affected. Specifically, for battery life, Intel Evo aims to have a battery life of more than 10 hours in a real FHD scenario. In addition, there is also a new indicator for fan noise, which requires it to be cold and quiet. Intel Evo requires less than 1.5 seconds in terms of fast response time. Then there is fast charging, which requires recharging for half an hour and a battery life of four hours. In addition, it needs to meet the requirements of environmental protection, and it needs to reach the silver level of EPEAT or pass the TCO certification.

In terms of intelligent connection, it consists of four parts. The first is VCX, which is used to improve the imaging quality of the camera; one is NPU, which was mentioned in detail earlier; the second is Wi-Fi 6e smart connection and support for ICPS; and the last is Microsoft-based MSE.

This time Intel Core Ultra also supports multi-device collaboration technology Intel Unison, this time Intel Unison extends to Android and iOS platforms, and supports screen expansion and file and image sharing on laptops and mobile devices. In the future, a general control function will be added, which can control different devices across devices through the mouse. Swift connect is a fast connection, but also a portable connection technology, which can quickly access devices

For peripherals, Intel also has a special certification system to enable partners to meet the requirements of Evo. For example, the new shape design, as well as lightning dock, display screen, memory, keyboard and mouse headphones, AP equipment and so on all have special products that comply with Evo certification.

Summing up this release of Intel Core Ultra processor also allows us to see that AI PC is accelerating into our daily work and life. The brand-new processor form, through the CPU, GPU, NPU three-layer AI computing network, Meteor Lake pushes the AI acceleration ability of the client processor to a new level. With the support of such a computing network, the local large language model and AIGC-related topics can jump away from the cloud computing power, making AI further deepen in the field of edge computing.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.