Intel’s Rocket Lake is headed to the launch pad, and it couldn’t come at a better time. AMD’s Zen architecture has thundered into the desktop PC market, steadily taking chunks of market share in a segment long dominated by Intel. Now AMD is on the cusp of releasing its Zen 3 chips, upping the ante in the suddenly brutally competitive desktop PC market.
Intel’s response will come in the form of Rocket Lake, but unlike the forward progress we see with the company’s 10nm Tiger Lake chips for laptops, the chips are 14nm’s last hurrah on the desktop. That marks the seventh and final iteration of the longest-lived leading-edge node in Intel’s history.
Intel’s 14nm process may be dated, but the company’s incessant optimizations have wrung out far more extra performance than anyone could have imagined when it debuted back in 2015. Rocket Lake is widely expected to match or exceed the 5.3GHz boost speeds we’ve seen with its Comet Lake processors.
But that’s the boring part of the Rocket Lake story. Rocket Lake could bring Intel’s first new microarchitecture to the desktop PC in five years. Intel has been hamstrung by its inability to economically produce its 10nm node, largely because it tied new chip architectures to new process nodes. As a result, the company wallowed on Skylake-based microarchitectures since 2015. Intel reconsidered this practice and announced in 2018 that it would make its future architectures portable between nodes in a process called backporting, meaning it’s plausible that a design originally intended for 10nm could come back to the 14nm node.
Test results submitted to publicly-accessible databases reveal that Intel is likely to backport a new chip design to 14nm, allowing it to bring a new architecture to the desktop market. A new architecture should offer higher instruction per cycle (IPC) throughput, and paired with the expected blistering 5.3 GHz ( ) clock speeds, Intel could deliver larger-than-normal performance gains that help assert its gaming dominance.
Intel’s 10nm delay also left it vulnerable on the graphics front, too. Courtesy of the integrated graphics engine present on almost all of its chips, Intel is the world’s largest GPU producer. However, Intel’s integrated graphics are woefully inadequate for gaming, and we haven’t seen a meaningful iGPU performance boost for desktop PCs since Kaby Lake arrived back in 2016. This changes with Rocket Lake, which is thought to come with Intel’s performant 12th-gen Xe graphics. There’s a bit more to that, which we’ll cover below, but the net-net is an expected doubling of graphics performance over the current-gen chips, bringing 1080p-capable gaming (albeit with reduced fidelity) to Intel’s mainstream chips.
We already know with certainty that Rocket Lake will come with support for the PCIe 4.0 interface, which provides twice the bandwidth as PCIe 3.0. That addresses a key weakness exacerbated by AMD’s now year-long advantage with its uncontested leadership position in I/O connectivity.
So what’s the catch? Early indicators seem to point to an eight-core ceiling for intel’s new chips, which certainly feels inadequate in the face of AMD’s 16-core Ryzen 9 3950X, not to mention Intel’s own Comet Lake roster that tops out at ten cores. If true, that means either Intel is betting heavily on its ability to realize large IPC gains from a new architecture, or that it will have a split product stack with Alder Lake hybrid chips coming in later to address the higher core count segment.
However, Rocket Lake still remains shrouded in mystery. Despite plenty of evidence that the chips do exist, Intel hasn’t made any official announcements. The company has even acknowledged the existence of its next-next-gen 10nm Alder Lake processors and announced their launch by the end of 2021, all while not revealing any details of Rocket Lake, its final 14nm refresh generation. However, there’s no doubt that Rocket Lake exists, so let’s cover what we know so far.
Intel 11th-Gen Rocket Lake At a Glance
- Support for PCIe 4.0
- 14nm processor
- New microarchitecture for the desktop
- AVX-512, Thunderbolt 4 support
- Rocket Lake is projected to land at CES 2021, but it could have slipped to Q1.
- Intel 12th-gen Xe LP Graphics provide step-function increase in integrated graphics performance
Intel 14nm Rocket Lake Release Date
As noted above, Intel hasn’t officially given a release date for the Rocket Lake-S processors. However, the company encountered problems with the i225 ‘Foxville’ 2.5 GbE PHY, which is responsible for providing wired 2.5 GbE connectivity for Rocket Lake-compatible 400-series motherboards, leading it to issue an advisory bulletin for the incoming fix.
That bulletin was purportedly leaked and posted, and it included the expected ETA for the hardware stepping fix. The bulletin lists that the corrected PHY is due in the second half of 2020, which the slide says is “aligned with RKL-S production timeframe.” That indicates that, at least at the time, the Rocket Lake-S desktop chips were slated to begin production in the second half of 2020.
Production schedules are subject to change based on a myriad of factors, and they also only denote when the company begins mass-producing the chips – not when they arrive at retail. Intel is largely thought to have the Rocket Lake launch planned for CES 2021, a relatively fast release after the Comet Lake processors launched a mere three quarters prior. We could also see the reveal at CES, followed by a launch in Q1 2021.
In either case, Intel officially announced that its 10nm hybrid Alder Lake-S chips would launch in the second half of 2021. That timeline indicates that Rocket Lake will likely serve either as a short-lived stopgap or part of a split product stack for the mainstream desktop, but it’s clear Intel can’t wait seven quarters to launch a replacement for Comet Lake. Especially with AMD’s Zen 3 barreling towards the desktop.
Intel 11th-Gen Rocket Lake Specifications and Performance Benchmarks
A test submission to the publicly-accessible 3DMark benchmark database revealed the highest core-count Rocket Lake model we’ve seen to date. This model comes with eight cores and 16 threads and is listed with a 3.2 GHz base frequency and 4.3 GHz boost. However, this early electronic sample (ES) chip was tested on an Intel reference validation platform (RVP), indicating the silicon is still under development. As such, we fully expect the frequencies to change before the chips hit retail. In fact, later results have surfaced with the chips sporting a 5.0 GHz boost.
Intel’s hyper-optimized 14nm process currently boosts to 5.3 GHz on a single core with the 10-core Core i9-10900K, and we expect an eight-core Rocket Lake model to meet or exceed that watermark.
Test results have also emerged for six-core 12-thread Rocket Lake processors, but there are no indications of Rocket Lake chips with more than eight cores. The Rocket Lake processors are said to have a 125W maximum TDP rating, which aligns well with the standard power envelopes we see with Intel’s flagship chips. With that in mind, it certainly feels like a regression for Intel to step back to eight cores, as Comet Lake-S (CML-S) sports up to 10 cores within the same TDP envelope. Given that Rocket Lake will come with a new microarchitecture, it appears that Intel is betting on drastically-improved instruction per cycle (IPC) throughput to offset the lower core counts.
The Rocket Lake-S processors will undoubtedly support the PCIe 4.0 interface. Our own sources told us that Intel had planned to add support for PCIe 4.0 to the Comet Lake chipset, and motherboard makers even designed their products to support the interface. However, issues with the chipset late in the production cycle led Intel to scuttle those plans, instead settling for limited PCIe 4.0 support for the PCIe slots only.
Support for the PCIe 4.0 interface is a curious addition to a platform launched with PCIe 3.0 Comet Lake chips that aren’t compatible with the interface, but those motherboards did ship to the market. Motherboard makers say this is to provide PCIe 4.0 support for “future” Intel processors that will drop into the platform, and our sources indicate those chips are the Rocket Lake models. We’ve also seen confirmation of Rocket Lake’s PCIe 4.0 interface via a benchmark submission to the 3DMark database.
The biggest series of clues comes from a leaked slide outlining the Rocket Lake platform. However, this slide is unconfirmed, so view it with the appropriate level of skepticism.
The slide outlines that the chips support 20 lanes of PCIe 4.0, with four dedicated to direct-attach NVMe storage – just like with the Tiger Lake processors. Intel has also broadened the DMI interface, which connects the processor to the platform controller hub (PCH), from a x4 connection to x8. The slide doesn’t indicate if this bus operates at PCIe 3.0 or 4.0 transfer speeds, but it will certainly help address a massive bottleneck for devices attached to the chipset, like SSD in RAID. However, the lanes that hang off the chipset are still at PCIe 3.0, which tracks well with reports that Intel encountered problems with its chipset design.
The slide notes that Rocket Lake-S will come with a new core architecture, and the purported support for AVX-512 instructions certainly furthers that assertion. We’ll dive into that more in the architecture section.
The slide lists the Rocket Lake processors with the Intel Xe graphics architecture, adding fuel to the rumors that Intel will use a secondary chiplet for the Xe graphics on Rocket Lake processors. The Linux-focused Phoronix website has also spotted code submissions that add support for Intel’s Xe LP graphics for the Rocket Lake processors, all but confirming the switch to a new graphics engine (though the form they come in remains open for debate).
Intel certainly has a broad pallet of chip packaging technologies that can enable a chiplet-based design, and EMIB is already proven in shipping designs. That line of thought is furthered by a series of leaked Intel roadmaps that listed Rocket Lake chips with 10nm graphics for mobile products and 14nm graphics for desktop processors. Shifting to a chiplet design would be a significant departure from Intel’s standard operating procedure, which focuses on chips with a monolithic die for the desktop. Still, the company has repeatedly outlined that chiplet-based designs are in its future plans. Intel also says the EMIB interconnect is mature enough for economical mass production. That said, it is entirely possible that any Rocket Lake chiplet designs are destined for the mobile market only, or that these chips will serve as specialized SKUs that replace the retired Kaby Lake-G models that came with an integrated graphics chip. The roadmaps do indicate 14nm graphics for the Rocket Lake desktop chips and 10nm chiplets for mobile chips only, making this a likely strategy.
Subsequent benchmark submissions have pointed to Rocket Lake having 32 execution units that run at 1.15 GHz on early silicon. Intel’s Gen12 Xe LP graphics are projected to be twice as fast as the existing Gen9.5 graphics, so we can expect a drastic improvement to graphics performance.
Other purported features include enhanced display capabilities with integrated HDMI 2.0b and DisplayPort 1.4a, 12-bit AV1 and HEVC compression, USB 3.2 Gen 2×2 (20G), new overclocking features, and USB audio offload. The chips also support the Thunderbolt 4 interface, but this doesn’t come as an integrated implementation as we see with Ice Lake and Tiger Lake. The slide also lists “increased” DDR4 speed via a dual-channel memory controller.
The test submissions point to DDR4-2933, which is the same speed as Intel’s Comet Lake chips. Naturally, that could increase during the optimization phase of the design process.
Intel Rocket Lake Architecture
It’s clear from the test submissions that the Rocket Lake processors come with a revamped cache hierarchy, and support for AVX-512 further points to a new microarchitecture for the desktop. Add in that the internal PCIe subsystem has been reworked to accommodate a x4 direct connection from the CPU to NVMe storage, and that the DMI lanes have been increased from four to eight, and it’s clear that Intel has made significant alterations to the design.
Intel’s Sunny Cove is the first core microarchitecture that is portable between nodes, while all other following architectures will also be portable. In the more immediate timeframe, Intel has either its Sunny Cove or Willow Cove cores to choose from for Rocket Lake.
Sunny Cove debuted with Ice Lake processors and imparted an 18% IPC gain over Skylake, at least according to Intel. Geekbench test submissions list Rocket Lake’s cache configuration at the same 2MB of L3 cache per core as Sunny Cove, and the L2 is also the same 512KB per core. That suggests Rocket Lake will have Sunny Cove cores, but it isn’t quite that simple.
Intel’s addition of a x4 direct connection to the CPU for NVMe devices didn’t come with Ice Lake, which features Sunny Cove cores. However, that could be a broader optimization, much like the widened DMI connection, that Intel added to an existing Core architecture.
Naturally, we expect there will be targeted optimizations to whatever architecture Intel uses to tailor it for desktop PC use. There are unconfirmed rumors that Intel will use the Willow Cove cores and brand them as Cypress Cove to denote the difference between the 10nm and 14nm versions. That does make some sense, but Intel hasn’t mentioned ‘Cypress Cove’ branding in any form on its roadmaps, while Willow Cove is front and center.
For now, the jury is out on whether Intel will use Sunny Cove, Willow Cove, or some custom combination of the two. Now, Intel doesn’t claim any IPC gains for the mobile Tiger Lake processors over the existing Ice Lake processors, instead saying it tuned the architecture for higher frequencies rather than a big IPC gain.
That means, regardless of which architecture (or derivative thereof), we’re still looking at the ~18% IPC gains that Ice Lake delivers over Skylake. Our sources say Rocket Lake’s architecture imparts IPC gains in the ~15% range, but do take the IPC predictions with a grain of salt, as that can vary by workload and measurement methodology.
Intel Rocket Lake Motherboard
Can you use your 400-series motherboard for Rocket Lake processors? Yes.
The Rocket Lake processors will drop into the Socket 1200 interface, so current-gen 400-series motherboards will support the chip. Many Z490 motherboards support the PCIe 4.0 interface, which motherboard vendors indicate is to provide support for the ‘future’ processors that are undoubtedly the Rocket Lake family.
However, in keeping with Intel’s traditional launch strategy, we expect that new 500-series motherboards, like Z590, will come to market with at least some new features, like expanded PCIe 4.0 connectivity, to accompany the Rocket Lake processors. We don’t expect a revolution in terms of features with 500-series motherboards, though.
Socket 1200 motherboards will be short lived. According to Intel’s own documentation, the hybrid Alder Lake-S processors that arrive in the latter half of 2021 will drop into Socket 1700. That means there will be no forward compatibility for Socket 1200 motherboards with future Intel processors.
Intel 11th-Gen Rocket Lake Pricing and Availability
Intel hasn’t officially announced the Rocket Lake processors, so there are no firm indicators of its pricing strategy. We do know that the company has steadily reduced pricing-per-thread as it attempts to fend off AMD’s value-centric Zen assault.
However, AMD made some changes to its pricing scheme with the Zen 2 generation. These chips came with higher recommended price points than AMD’s previous-gen chips as the company positions itself as a premium chip supplier as opposed to its long history as the value alternative. The company has also discarded one of the goodies we’ve become accustomed to – the higher-end Ryzen XT models came without bundled coolers. We could see the company take a similar approach with Zen 3 processors.
That means Intel may not have to be as competitive on the pricing front, so we could see per-core and per-thread pricing remain relatively static with the 11th-Gen Rocket Lake processors.
Some unconfirmed signs point to Intel using a next-gen packaging technique, likely EMIB, for some models. That could cause an incremental price increase – advanced packaging techniques do trend towards the expensive – but it remains to be seen if that will have a notable impact on end-user pricing. Not to mention that it seems unlikely EMIB-connected chips will come to the mainstream desktop, instead being relegated to mobile or specialty products.
We’ll update this article as more information becomes available.