Intel unveiled its new Intel Core Extremely processor—codenamed Meteor Lake—at its latest Innovation convention, which Moor Insights & Technique CEO and chief analyst Patrick Moorhead mentioned on this piece. To get extra background on the chip, I had the chance to journey to Malaysia just a few weeks in the past and get briefed about Meteor Lake. I additionally noticed how Intel has ready to construct this subsequent era of processors, and on this piece I need to share what I’ve discovered.
Many issues are altering inside Intel, together with how the corporate thinks about processors. A major factor of that is how the corporate configures its processors and the way that configuration is determined by new packaging and interconnects. That’s why dozens of press and analysts got here out to Penang, Malaysia, to see how Intel has carried out these capabilities at present and the way it’s already making ready to broaden them quickly.
Intel Core Extremely
Meteor Lake is constructed on what many within the business name a tiled structure, that means that the chip has many various tiles with completely different IP designs on them. On this case, Meteor Lake makes use of 4 tiles and a base die, together with an SoC tile, a graphics tile, a compute tile and an I/O tile. The completely different tiles are manufactured individually, then mixed utilizing Intel’s manufacturing and packaging applied sciences corresponding to Foveros.
This disaggregated manufacturing method signifies that Intel can at all times construct its SoCs utilizing essentially the most cost-effective and performance-optimized course of node for the IP on that tile. For instance, the compute tile (the CPU) makes use of the Intel 4 course of node, whereas the 3-D Foveros base die makes use of Intel’s 16 course of node. The graphics tile (also called the GPU) makes use of TSMC’s N5 course of node, whereas the SoC tile and I/O tile each use TSMC’s N6 course of node. This permits Intel to optimize every IP for the most effective yield and efficiency. Greater than that, I imagine this strategy reveals the corporate’s realization that it should adapt to market backlash round its personal course of stumbles lately, in addition to the fact that rivals like AMD have been utilizing compute tiles for some time.
CPU Cores
The brand new Intel Core Extremely processors now have three tiers of CPU cores, so clients can select from efficiency cores (P-cores), effectivity cores (E-cores) and low energy effectivity cores (LP E-cores). Prior to now, all of those cores would have resided on the compute tile. Intel has taken a distinct strategy with Meteor Lake, and two of the E-cores are low-power effectivity cores that reside on the low-power island contained in the SoC tile for even lower-power workloads. The P-cores are primarily based on the Redwood Cove microarchitecture, whereas the E-cores are primarily based on the Crestmont microarchitecture. This implies there aren’t important enhancements to IPC (iterations per clock) however relatively a powerful concentrate on effectivity. This additionally signifies that Intel’s Thread Director perform can now select throughout three completely different energy and efficiency curves to find out the place a workload is perhaps a great match, if it even finally ends up on the CPU in any respect.
GPU Cores
The GPU tile on the Intel Core Extremely makes use of Intel’s Arc graphics structure, placing it in a better class of graphics efficiency. This tile makes use of Intel’s new Xe-LPG graphics IP, which is a hybrid primarily based on its earlier Xe-LP (low-power) embedded designs. It additionally takes a lot of its efficiency options from Intel’s Xe-HPG (high-performance graphics) desktop and discrete graphics designs. Intel claims that this delivers one more 2x improve in GPU efficiency over Xe-LP in Tiger Lake, also called eleventh Gen Intel Core.
The Xe-LPG graphics tile is manufactured utilizing TSMC’s N5 course of node, a extra superior course of than the TSMC N6 node used for Intel’s present Alchemist household of discrete Arc graphics merchandise. This could yield even higher energy effectivity and areal density, that are important for a cellular type issue like Meteor Lake. In comparison with earlier years, extra of the graphics and show IP is shared throughout completely different tiles. For instance, the Xe show engine and Xe media engines aren’t on the graphics tile, however as a substitute reside on the low-power island of the SoC tile, which is smart on condition that each of these engines will likely be always-on. In different phrases, there could also be situations the place the GPU doesn’t must activate, however the show and media engine nonetheless do. The show PHYs are additionally situated on the I/O tile, which is smart as a result of the I/O tile is the place most issues will interface with the SoC and may be modified, relying on the SKU.
The Xe media engine inside Meteor Lake options AV1 {hardware} help, enabling 8K10 10-bit HDR encoding and decoding. The show engine brings energy optimizations with selective replace and {hardware} queueing, stopping pointless CPU and reminiscence calls. The show engine additionally permits for next-gen connectivity like HDMI 2.1 and DisplayPort 2.1, concurrently enabling as much as 8K 60 HDR or 4 4K60 shows. The Xe tile has two render slices with eight Xe cores, eight ray tracing items and 128 vector engines, amongst different options, with every Xe core having sixteen 256-bit vector engines. It should additionally convey many Xe-HPG inherited options, which ought to translate to fairly a little bit of efficiency optimizations. This consists of bringing Xe SS (supersampling) to Meteor Lake, which ought to significantly enhance gaming efficiency and battery life. This additionally consists of inheriting DP4A AI acceleration from Xe-HPG.
AI Acceleration — NPU
The neural processing unit (NPU) inside Meteor Lake is just one of many elements of the entire SoC that may execute AI workloads. It’s Intel’s first built-in NPU, which is devoted to executing AI workloads and doing so in a power-efficient manner. With SoCs gaining AI capabilities, we see the business shifting in direction of a diversified AI execution mannequin the place some AI duties occur on the NPU, whereas others occur on the GPU, on the CPU and even throughout a number of cores concurrently. We’ve already seen a lot of that taking place within the cellular section, primarily since Google’s Neural Networks API (NNAPI) primarily works on a mixture of CPU + GPU at present.
As proven within the chart above, there are completely different advantages to operating issues on a mix of cores, relying on the workload. The NPU inside Intel Core Extremely is essentially the most highly effective the corporate has designed for a shopper product and can allow Intel and its builders to drive AI at scale on-device. Additionally, as a result of the NPU is probably going for use regularly and for longer intervals of time, impartial of the CPU or GPU tiles, it additionally resides on the SoC tile with the media and show engines in addition to the low-power E-cores.
Wi-Fi 7 and Thunderbolt 4
Intel didn’t speak a lot about this, however Meteor Lake goes to be the corporate’s first product to supply Wi-Fi 7. It is a large deal as a result of, nearly concurrently, Qualcomm has introduced a lot of its personal Wi-Fi 7 merchandise; this implies we are going to lastly get actual quantity of Wi-Fi 7 merchandise in This fall of 2023 and the beginning of subsequent yr. Wi-Fi 7 is a giant deal as a result of it takes benefit of extra spectrum, additional expands latency enhancements over Wi-Fi 6 and aggregates extra spectrum to enhance throughput, which might enhance battery life.
The Intel Core Extremely options the Thunderbolt 4 connector, which Intel introduced in 2020 and has develop into ubiquitous for delivering exterior storage and show capabilities in laptops and bettering the docking expertise. That mentioned, Intel additionally simply introduced Thunderbolt 5, which brings important enhancements in bandwidth and energy supply. I want that Intel may squeeze Thunderbolt 5 into Meteor Lake, however plainly must wait till subsequent yr since Thunderbolt 5 solely simply obtained introduced as properly. Thunderbolt 4 continues to be nice for docking and connectivity, however I really feel Thunderbolt 5 may give {that a} large increase with added energy and throughput.
Meteor Lake and the AI PC
Intel Core Extremely is a transparent shift from Intel by way of the way it designs and manufactures processors to handle the expansion of AI. Intel continues to be on a protracted journey of regaining process-node management, however the firm has taken a web page out of its rivals’ e book and disaggregated its compute into completely different tiles and, with applied sciences like Foveros, can achieve this with out a lot of a penalty to energy or efficiency.
Meteor Lake has demonstrated that Intel can take its many various IPs and optimize them to match what shoppers and OEMs need. The largest challenges now will likely be getting builders onboard with toolsets like OpenVINO for AI inference and for Microsoft to push the envelope on AI with Home windows to make the most of all of the {hardware} capabilities constructed into Intel Core Extremely. Even ignoring AI, Meteor Lake nonetheless brings extraordinarily aggressive CPU, GPU and connectivity selections, so even should you aren’t that eager on the AI increase we’ve seen this yr, there are nonetheless loads of advantages to be gained by upgrading. That mentioned, I do imagine that Meteor Lake’s AI capabilities will likely be essential to allow the AI PC of the longer term, which I feel we are going to begin to see from Microsoft and PC OEMs very quickly.
Moor Insights & Technique offers or has offered paid (want providers to know-how firms, like all tech business analysis and analyst corporations. These providers embrace analysis, evaluation, advising, consulting, benchmarking, acquisition matchmaking, and video and talking sponsorships. The corporate has had or presently has paid enterprise relationships with 8×8, Accenture, A10 Networks, Adobe, Superior Micro Units, Amazon, Amazon Internet Providers, Ambient Scientific, Ampere Computing, Analog Units, Anuta Networks, Utilized Mind Analysis, Utilized Micro, Apstra, Arm, Aruba Networks (now HPE), Atom Computing, AT&T, Aura, Avaya Holdings, Automation Anyplace, AWS, A-10 Methods, Bitfusion, Blaize, Field, Broadcom, C3.AI, Calix, Cadence Programs, Campfire, Cisco Programs, Clear Software program, Cloudera, Clumio, Cohesity, Cognitive Programs, CompuCom, Cradlepoint, CyberArk, Dell, Dell EMC, Dell Applied sciences, Diablo Applied sciences, Dialogue Group, Digital Optics, Dreamium Labs, D-Wave, Echelon, Elastic, Ericsson, Excessive Networks, Five9, Flex, Fortinet, Foundries.io, Foxconn, Body (now VMware), Frore Programs, Fujitsu, Gen Z Consortium, Glue Networks, GlobalFoundries, Revolve (now Google), Google Cloud, Graphcore, Groq, Hiregenics, Hotwire International, HP Inc., Hewlett Packard Enterprise, Honeywell, Huawei Applied sciences, HYCU, IBM, Infinidat, Infoblox, Infosys, Inseego, IonQ, IonVR, Inseego, Infosys, Infiot, Intel, Interdigital, Intuit, Iron Mountain, Jabil Circuit, Juniper Networks, Keysight, Konica Minolta, Lattice Semiconductor, Lenovo, Linux Basis, Lightbits Labs, LogicMonitor, LoRa Alliance, Luminar, MapBox, Marvell Expertise, Mavenir, Marseille Inc, Mayfair Fairness, MemryX, Meraki (Cisco), Merck KGaA, Mesophere, Micron Expertise, Microsoft, MiTEL, Mojo Networks, MongoDB, Movandi, Multefire Alliance, Nationwide Devices, Neat, NetApp, Netskope, Nightwatch, NOKIA, Nortek, Novumind, NTT, NVIDIA, Nutanix, Nuvia (now Qualcomm), NXP, onsemi, ONUG, OpenStack Basis, Oracle, Palo Alto Networks, Panasas, Peraso, Pexip, Pixelworks, Plume Design, PlusAI, Poly (previously Plantronics), Portworx, Pure Storage, Qualcomm, Quantinuum, Rackspace, Rambus, Rayvolt E-Bikes, Purple Hat, Renesas, Residio, Rigetti Computing, Ring Central, Salseforce.com, Samsung Electronics, Samsung Semi, SAP, SAS, Scale Computing, Schneider Electrical, SiFive, Silver Peak (now Aruba-HPE), SkyWorks, SONY Optical Storage, Splunk, Springpath (now Cisco), Spirent, Splunk, Dash (now T-Cellular), Stratus Applied sciences, Symantec, Synaptics, Syniverse, Synopsys, Tanium, Telesign,TE Connectivity, TensTorrent, Tobii Expertise, Teradata,T-Cellular, Treasure Information, Twitter, Unity Applied sciences, UiPath, Verizon Communications, VAST Information, Veeam, Ventana Micro Programs, Vidyo, Volumez, VMware, Wave Computing, Wells Fargo, Wellsmith, Xilinx, Zayo, Zebra, Zededa, Zendesk, Zoho, Zoom, and Zscaler.
