Even though the competitor AMD continues to advance its 7 nm silicon, Intel has so far held onto a larger 14 nm node for its desktop chips and only migrated to 10 nm on mobile devices last year. Critics have complained of Intel's stubbornness, but the company recently released a 10-year roadmap outlining a bullish cadence of innovations with promising long-term prospects.
Looking ahead to the end of the decade, Intel plans to achieve a much smaller 1.4nm node by 2029 – or just a tenth the size of its current desktop node. Intel hopes that its aggressive development trajectory will help it regain silicon leadership, which will be important to compete with AMD's Ryzen processors, and drive to move from companies like Apple and Microsoft to ARM-based processors.
So far 2020: Ice Lake and Comet Lake
This year, Intel continues to introduce the 10th generation processor family that began with last year's 10nm Ice Lake debut on thin and light laptops. Working with Ice Lake, Intel is also driving its Project Athena initiative to standardize small, portable laptops that have been certified by Intel.
One of the biggest improvements to Intel's Ice Lake processors was the integration of integrated Gen11 graphics, also known as Iris Plus.
In 2020, Intel will also launch new versions of its 14nm architecture. The 10th generation Comet Lake H processors were introduced on larger, more powerful laptops. Intel took a similar approach to desktops by introducing Comet Lake S-Chips.
In addition to these standard annual updates, Intel has also officially launched its Lakefield hybrid processors, which will be available later this year on devices with new form factors and are likely to be part of the company's Project Athena initiative. These experimental new chips are expected to be available on foldable devices such as Lenovo's ThinkPad X1 Fold, announced at CES in January, and on dual-screen devices such as Microsoft's Surface Neo.
Unlike traditional Intel chips, Lakefield uses a combination of architectures to optimize performance and battery life. The design is no different from the big one. SMALL ARM chip design because Intel wants to combine Ice Lake's Sunny Cove microarchitecture on the powerful Core i3 and Core i5 processors with energy-efficient, atom-based Tremont cores.
2020 is coming: Tiger Lake, Xe graphics and more
Intel typically launches its next generation of mobile processors in the fall. This year it is the 11th generation Tiger Lake line. Tiger Lake could debut on September 2 at a planned Intel event with 10nm ++ design, although this has not yet been confirmed. Compared to a 14nm design, Intel claimed that its 10nm-based nodes offer 2.7x density scaling. These nodes are based on Fovero's first generation 3D stacking and second generation EMIB packaging designs.
And although this wasn't officially announced, Intel had already confirmed some of the features, including support for the new Thunderbolt 4 standard USB4 and the new Gen12 graphics. Gen12 graphics, also known as Intel Xe, are based on the same GPU architecture that Intel uses for its discrete DG1 graphics card. Gen12 is expected to offer twice the performance of Gen11, which will help Intel stand up to the upcoming 7nm Navi graphics architecture from rival AMD.
A recent SiSoftware benchmark found that Gen12 graphics could fall under the Intel Iris Xe branding. The benchmark also confirmed that Intel's integrated GPU is clocked at 1.3 GHz and contains 96 EUs. In a separate Twitter post, Intel chief strategist Ryan Shrout showed the possibilities of Gen12 graphics. In a video uploaded to the social networking site, Shrout showed a laptop with Gen12 graphics running at 30 frames per second, with Battlefield V playing at high settings.
Intel previously confirmed in a presentation on the company's roadmap that Tiger Lake will be equipped with a new processor core. The processor will likely be based on an expanded 10nm node (10nm +) and will have new Willow Cove cores. Compared to the 10th generation Ice Lake, according to Tom's hardware, Tiger Lake should arrive with up to 50% more L3 cache than Ice Lake.
While Intel will only use its Tiger Lake silicon for laptops and gaming notebooks, Comet Lake will be the company's Rocket Lake processor on the desktop. Confusingly, Intel's Rocket Lake should also be available on laptops. On the desktop, the 11th generation Rocket Lake S will finally migrate from its aging 14nm Skylake architecture, which will now be in its seventh iteration. Intel's leaked roadmap confirmed backporting opportunities, and industry insiders believe that desktop silicon could use a Willow Cove design backport that debuted at Tiger Lake.
Intel promises better performance with Rocket Lake and it is believed that the CPU can achieve an IPC increase of approximately 25% compared to Comet Lake. Although speculation is set to hit Rocket Lake in 2020, given Intel's already busy cadence for release, it wouldn't surprise us if the processor started in early 2021.
A leaked film obtained from Videocardz showed that Rocket Lake may be the first Intel CPU to support PCIe 4.0. It also uses a premium GT1 version of the Intel Xe graphics. Rocket Lake is speculated to use Intel's hybrid microarchitecture design, which combines a 10nm processor with an integrated 14nm GPU architecture.
Graphics performance is up for debate, however, as Twitter user @chiakokhua said Rocket Lake's Gen12 graphics will only contain 32 execution units compared to the 96 EUs on Tiger Lake. Intel's silicon also supports discrete (rather than integrated) Thunderbolt 4 support.
Oddly, unlike Comet Lake, which works at its maximum with a 10-core Core i9 design, Rocket Lake is believed to use an eight- and 16-thread architecture, according to Wccftech, and the lower number of cores could help Intel make a large one IPC provide profits with this generation.
Despite maintaining compatibility with the LGA1200 socket, upgradeability is still under discussion as Intel plans to launch its new 500 chipset alongside Rocket Lake. The processor series will include 15-watt Rocket Lake-U and 45-watt Rocket Lake-H for mobile devices, and desktop Rocket Lake-S will have TDPs between 35 and 125 watts.
Intel's discrete DG1 graphics card, based on the same integrated Gen12 graphics architecture, was launched at CES earlier this year. Intel chief architect Raja Koduri had previously indicated a start in June. However, this schedule was likely disrupted by the global coronavirus pandemic.
Consumers looking for a suitable rival for the graphic stronghold of AMD and Nvidia have to look elsewhere since DG1 is not intended for the consumer market. However, we still expect Intel to launch its first discrete graphics card for end users in late 2020.
2021: A hybrid architectural approach
In 2021, the transition from Intel to the 12th generation begins with Alder Lake. On the desktop, the 10nm Alder Lake is expected to hit the market later this year. This is the fourth iteration of Intel's 10 nm node. Above all, Alder Lake S will introduce a new 10nm silicon architecture for desktops based on Lakefield's hybrid architecture variant of ARM's big.LITTLE design.
It's unclear why Intel is migrating to this architecture at this point – desktops aren't limited by the battery life requirements of Lakefield phones – but one possibility is that Intel can use more cores in its CPU layout. Rumor has it that Alder Lake S will combine Intel's more powerful Willow Cove or Golden Cove cores with the low-performance Tremont or Gracemont cores, while the exact design is still speculative.
There will be multiple configurations of Alder Lake that span desktop and mobile devices and support different TDPs. On the desktop, Alder Lake S could arrive in a configuration with eight large cores and eight small cores or six large cores and no small cores. Both configurations come with premium GT1 Xe-based graphics.
It is also believed that the processor initiates support for DDR5 memory and is compatible with Intel 600 series motherboards. According to Aloc Lake, Intel is to combine the 10 nm processor with integrated 14 nm Intel Xe-based graphics in a mixed chiplet strategy, according to Videocardz.
In the high-end desktop area (HEDT), Intel may not have a processor until 2021, so the upcoming Ryzen 4000 Threadripper from AMD could remain unchallenged for a while. This means that by 2021 there will be no Cascade Lake-X successor to hit the market in late 2019.
2022: move to 7nm
Beyond Alder Lake, Meteor Lake is the last important code name that we have in Intel's architectural plan. Meteor Lake S will be the first 7nm desktop processor from Intel, which means that switching from Intel to 7nm computers will be several years behind AMD. Like Alder Lake, Meteor Lake will be a heterogeneous silicon as the company continues to develop a design that resembles ARM's big.LITTLE architecture and works on both desktops and mobile devices.
Compared to 10 nm, Intel's 7 nm process delivers double density and uses FIBOS and Embedded Multi-Die Interconnect Bridge (EMIB) packaging of the next generation. The company expects to also be able to use planned internal optimizations. And since Intel claims that its 7nm process offers superior transistor density over TSMC's 5nm fabrication, Meteor Lake S could potentially arrive with even more cores. The processor is manufactured using extreme ultraviolet lithography or EUV.
The chipset will use Intel's Golden Cove cores. Meteor Lake S is likely to continue to be compatible with Alder Lake's LGA1700 jacks, but details beyond are not known.
2023 to 2029: trip to 1.4nm
Although we have no leaked product names, it is believed that Intel will likely further optimize its 7nm process in 2023 and move towards the 5nm process in 2024 to end the decade at 1.4nm. According to the schedule, this means that Intel returned to a two-year cadence with the launch of the 10nm ice lake in late 2019.
Intel describes the two-year cadence of the release of new nodes as the optimal cost-performance path. However, the company also leaves room to get more performance out of existing nodes through backporting, as explained on a slide at the IEEE International Electron Devices Meeting that Anandtech received.
With each main process node, Intel will also work on extended + and ++ versions that will get even more performance out of their investment with at least two optimizations. Currently with 10nm + we see 10nm ++ and 10nm +++, while 7nm 2022 will get 7nm + and 2023 7nm ++. The 5nm from Intel will be optimized to 5nm + in 2024 and 5nm ++ in 2025 and so on until we arrive at 2nm ++ in 2029.
These iterations are performed annually, and according to the release, Intel will have overlapping teams to ensure that process node development overlaps. Due to the overlap, the ++ optimized node can be started next to the next main node. This means that the optimized node can include some advantages such as higher clock rates and better yields. According to this schedule, we can expect 2021 7 nm, 2023 or 2024 5 nm, 2025 3 nm, 2027 2 nm and 2029 1.4 nm, Wccftech reported. The cadence of these major versions would be aggressive for Intel because it took a long time for the company to switch completely to 10 nm on mobile devices and desktops.
Intel's mention of backport opportunities on the slide is interesting, which is currently rumored for the architecture of Rocket Lake. For example, backporting could allow Intel to use a 7nm design on a 10nm +++ node or a 5nm design on a 7nm ++ node. Intel with Rocket Lake is believed to use 10nm Willow Cove cores on a 14nm architecture.
Anandtech noticed that Intel was also talking about new materials and designs, so we could start seeing nanosheets and nanowires when Intel took advantage of FinFET beyond the 7nm node that TSMC's 5nm node should correspond to process.