11th gen Core for desktop a.k.a. “Rocket Lake“. i9-11900K, the coming king of gaming PCs?
Intel is likely at the point where it got out of the worst of the “never-ending 14nm process” period now: the company has unveiled three families of 10nm mobile processors at CES 2021, some clocked up to 5 GHz. And in addition to this, Rocket Lake processors for desktop. Those are still 14nm products (the last ones), but they finally bring a new – and potentially very good – architecture. All these CPUs will be released in Q1 2021.
Let’s keep desktop processors asside for a moment and start with the mobile processors. The first to ship will be more powerful 10nm processors for laptops. In the summer, Intel released 15W to 28W quad-core Tiger Lake-U processors and now two more powerful versions are coming. The first of these are the Tiger Lake-H35 processors. These have actually been released last week and should start appearing in notebooks.
Tiger Lake-H35: four cores for thinner gaming notebooks
These processors are not brand new. They use the same silicon as Tiger Lake-U (the designation H35 is therefore slightly misleading). The difference is that Intel uses 35W TDP here, which allows for higher performance (optionally, they can be limited to 28 W, so in theory they can overlap with the fastest U-series chips).
These processors therefore have just four cores with eight threads and an integrated Intel Iris Xe GPU with 96 EU (768 shaders). Intel launched a total of three models, which you can see in the table below. The cheapest Core i5-11300H has a base clock speed of 3.1 GHz at 35W TDP, all-core boost of 4.0 GHz and a maximum boost for 1–2 active cores of 4.4 GHz. It also has its L3 cache cut down to 8 MB.
However, the next two faster models are the more interesting ones. These have a 12MB L3 cache; The Core i7-11370H has a base clock speed of 3.3 GHz, a maximum all-core boost of 4.3 GHz and maximum boost for 1–2 cores of 4.8 GHz. The maximum is therefore the same as the fastest U-series Tiger Lake, Core i7-1185G7.
And that’s where the fastes model comes in, the Core i7-11345H Special Edition – that is, this CPU should have a special status, we are not yet sure how widely will it be available. This i7-11375H SE has the same parameters as the i7-11370H overall, except that it has a 5.0 GHz boost for one core. This is achieved using the Turbo Boost Max 3.0 technology, so this clock is only available on the so-called favored core and not on all of them. In any case it’s Intel’s fastest 10nm processor in terms of single-threaded performance.
These processors have the same integrated chipset and connectivity (for example Thunderbolt 4/USB4) as Tiger Lake-U, memory can be DDR4-3200 or LPDDR4X-4266. However, the limitations of the PCI Express controller of the U-silion also remain, which distances these quad-cores from the regular H-series processors. For a graphics card, only a PCI Express 4.0 ×4 interface is available (which is actually intended to be mainly used for SSDs with U-series processors).
This limits its potential a bit, on the other hand the bandwidth is the same as with PCI Express 3.0 ×8, so these processors will still be appropriate for more portable gaming laptops. Intel gives the example of devices with a thickness of only around 16 mm. Of course, such lighter gaming laptops can also rely just on the integrated Iris Xe GPU. By the way, when paired with an Nvidia GeForce RTX 3000 graphics card, it will be possible to use the PCIe Resizable BAR feature (similar to AMD Smart Access Memory) to improve performance. Intel says it has implemented it in close collaboration with Nvidia.
According to Intel, over 40 laptops with these processors from various manufacturers should appear during the first half of this year.
8-core Tiger Lake-H models
While the Tiger Lake-H35 is not very different from the U version, there are regular Tiger Lake-H processors coming after that as well. But Intel has not yet completely revealed them at CES 2021, just teased them. However, they are also expected to hit the market later this quarter, so they are not that far away. The company is still hiding specific models and parameters though.
Tiger Lake-H will have eight cores and probably 45W TDP. The core architecture should be exactly the same, so processors could have up to 24 MB of L3 cache. The GPU has the same Xe LP architecture but only gets 32 EU/256 shaders. That’s because these processors are expected to usually be paired with a powerful dedicated GPU in a gaming notebook.
And Tiger Lake-H will already have a proper support for them: the processor provides a PCI Express 4.0 ×16 interface, i.e. the GPU connectivity as fast as on the best desktop platforms of today (AMD Ryzen 5000). In addition, there is another PCIe 4.0 ×4 for NVMe SSDs. The Resizable BAR feature should also work with these processors.
5 GHz on multiple cores simultaneously
We don’t know the performance and parameters, but Intel said that these processors will also reach up to 5.0 GHz clocks. And not just on one core, but with multiple cores active too. Single-threaded performance should be very good, but this time multi-threaded as well thanks to eight 10nm cores. Tiger Lake-H could be well competitive against mobile 7nm APU with eight Zen 3 cores, which AMD introduced a day after Intel had its keynote as the mobile Ryzen 5000 (and we’ll look at them later in another article). It remains to be seen which processor will be more energy efficient.
Jasper Lake: 10nm Atom with Tremont architecture is here
The last mobile processors we need to discuss are Jasper Lake. This is also a completely new chip – the very first 10nm SoC of the cheap and at the same time power-efficient line of Atom processors with “small cores”. Jasper Lake delivers four cores using the Tremont architecture, manufactured on a 10nm process of the same type as the one used for Ice Lake processors (it is not the more powerful SuperFin variant from Tiger Lake).
Intel did not devote much time to Jasper Lake, which doesn’t surprise too much though, previous generations of these cheap processors were often launched with close to zero publicity. Intel has now announced Jasper Lake as a solution for Chromebooks, but as usual, they should probably be used in traditional laptops, mini PCs and mini-ITX boards too. Intel already considers them as launched to market, but the arrival of particular devices based on these chips might take a few weeks or months.
Jasper Lake comes in a total of six models – three 6W ships for laptops and three 10W chips for desktops. 10W models should consume up to 20 W while boosting, so these processors joins the list of Intel products on which TDP does not equal the maximum power draw. You can see the parameters of the models in the tables below.
A Tremont core has significantly higher performance than the previous Goldmont+ architecture used in 14nm Gemini Lake Atoms/Celerons/Pentiums that are currently sold. The IPC should increase by up to 30%; according to Intel, the performance of processors should be up to 35% better than that of Gemini Lake. The processors support DDR4-2933 and LPDDR4X-2933 memory.
Integrated graphics should also be improved. For Pentium Silver models, it has 32 EU (256 shaders), but for quad-core Celerons, it is cut to 24 EU (192 shaders) and for dual-core to only 16 EU (128 shaders). The architecture is Gen11, like in Ice Lake processors. So this is not yet the latest Xe architecture from Tiger Lake processors. Therefore, AV1 decoding is likely not yet supported either.
On the next page, we’ll take a look at Rocket Lake desktop processors and the first showing of the next-gen Alder Lake CPU.
Intel is likely at the point where it got out of the worst of the “never-ending 14nm process” period now: the company has unveiled three families of 10nm mobile processors at CES 2021, some clocked up to 5 GHz. And in addition to this, Rocket Lake processors for desktop. Those are still 14nm products (the last ones), but they finally bring a new – and potentially very good – architecture. All these CPUs will be released in Q1 2021.
Now let’s finally talk the last 14nm processor to be released by Intel (or hopefully the last). Ironically, these could could actually be some of the most interesting ones. Rocket Lake, or the 11th generation Core for desktop, has not yet been released, but Intel has previewed it and revealed most of the details about the fastest SKU: Core i9-11900K. These processors should also be released this quarter, probably in March.
Finally a new desktop architecture
Rocket Lake is a line of processors for the LGA 1200 socket, expected to offer six or eight cores – there will be no deca-cores. The processor uses the architecture from 10nm Ice Lake and/or Tiger Lake processors, modified to be manufactured on a 14nm process – the resulting core is called Cypress Cove and is equipped with 512KB L2 cache (so the architecture might be closer to Ice Lake. According to Intel, its IPC – i.e. performance at a given clock – is 19% higher than that of Skylake, which promises very good single-threaded performance as well as performance in games. The core also brings support of AVX-512, VNNI and DL Boost instructions into mainstream desktops, for the first time.
The integrated graphics does come from Tiger Lake processors though – it’s Iris Xe, but only with 32 EU (256 shaders). However, support for AV1 video playback is retained. The news is that this GPU will always be active, even if you install a dedicated graphics card alongside. This keeps Intel’s Quick Sync multimedia acceleration available for PCs with a dedicated GPU. On the software side this should be handled similarly to how GPU switching works in notebooks.
Intel has not yet revealed all models and parameters and focused only on the most powerful version of the Core i9-11900K. However, below you can see a table of models leaked from unofficial sources (the numbers that are more likely to be true have been marked in bold).
The Core i9-11900K will be an octa-core with 16 threads and 16MB L3 cache, with 125W TDP. The PL2 power draw in boost should be up to 250 W and the standard time for which this will be sustainable will in turn be the same as for Comet Lake – 56 seconds (but motherboards will again be able to override this duration to “infinite”). The processor will have an all-core boost of up to 4.8 GHz – only 100 MHz lower than the Core i9-10900K in the Comet Lake generation. But the single-core boost will be the same – 5.3 GHz. Both frequencies are with Thermal Velocity Boost applied, so they are dependent on temperature being low enough, so when the chip heats up above a certain limit, the frequency will probably decrease. The same is already the case with the i9-10900K.
Intel has managed to port a core with a 19% higher IPC to a 14nm process without losing any clock speed – this shows that the clock potential of 14nm technology is really top notch after the years of tweaking it has seen. But the area of the chip and energy efficiency is, naturally, a big burden compared to more modern technologies, so it is still absolutely necessary to move on from 14nm.
| SKU | Cores | L3 | Clock | Max. Boost | All-core Boost | GPU | TDP |
| i9-11900K | 8 (16 t.) | 16 MB | 3.5 GHz | 5.3 GHz | 4.8 GHz | Yes | 125 W |
| i9-11900KF | 8 (16 t.) | 16 MB | 3.5 GHz | 5.3 GHz | 4.8 GHz | No | 125 W |
| i9-11900 | 8 (16 t.) | 16 MB | ? GHz | ? GHz | ? GHz | Yes | 65 W |
| i9-11900F | 8 (16 t.) | 16 MB | ? GHz | ? GHz | ? GHz | No | 65 W |
| i7-11700K | 8 (16 t.) | 16 MB | 3.6 GHz | 5.0 GHz | 4.6 GHz | Yes | 125 W |
| i7-11700KF | 8 (16 t.) | 16 MB | 3.6 GHz | 5.0 GHz | 4.6 GHz | No | 125 W |
| i7-11700 | 8 (16 t.) | 16 MB | 2.5 GHz | 4.9 GHz | ? GHz | Yes | 65 W |
| i7-11700F | 8 (16 t.) | 16 MB | 2.5 GHz | 4.9 GHz | ? GHz | No | 65 W |
| i5-11600K | 6 (12 t.) | 12 MB | ? GHz | 4.9 GHz | 4.6 GHz | Yes | 125 W |
| i5-11600KF | 6 (12 t.) | 12 MB | ? GHz | 4.9 GHz | 4.6 GHz | No | 125 W |
| i5-11400 | 6 (12 t.) | 12 MB | 2.6 GHz | 4.4 GHz | 4.2 GHz | Yes | 65 W |
| i5-11400F | 6 (12 t.) | 12 MB | 2.6 GHz | 4.4 GHz | 4.2 GHz | No | 65 W |
Intel has shown benchmarks according to which the Core i9-11900K should beat the Ryzen 9 5900X (a 12-core Zen 3) by 2–8% in 1080p gaming. Of course, these are official results that need to be taken with a grain of salt, as they can be cherry-picking. But Intel has the advantage of monolithic silicon with lower memory latency on their side, so it is not unexpected for them to gain the upper hand in gaming performance. In any case, real performance will only be clear once there are independent reviews.
During the presentation, Intel also showed a demo in Metro Exodus, where the i9-11900K achieved about 6.2% higher average FPS compared to the Ryzen 9 5900X: 156.54 FPS against 147.43 FPS.
New motherboards and compatibility just with H470 and Z490?
Rocket Lake processors now support PCI Express 4.0 – they also provide PCIe 4.0 ×16 for a graphics card, as well as PCIe 4.0 ×4 for SSDs. Official DDR4-3200 support is added, too. This will work on 500-series chipset boards, but after upgrading the BIOS, it will also be possible to use PCI Express 4.0 on motherboard with Z490 chipset as long as they were designed for this in advance.
The new 500-series chipsets have already been officially released and such motherboards should start appearing in the coming days and weeks. Their advantage is that they support SuperSpeed USB 20Gbps (USB 3.2 Gen 2×2). With the exception of H510 that is, H510 is limited to just base 5.0Gb/s USB 3.0 ports. As a novelty, the B560 and H570 chipsets will now support XMP profiles and DDR4 overclocking, so they will be much better for gaming computers, as faster memory (along with overclock CPUs through increasing their PL2 and TDP) will partially replace the capability to full overclock Intel CPUs, which remails exclusive to Z590.
Another big change is in the connection of the chipset to the processor. It is confirmed that Intel has increased throughput between chipset and Rocket Lake CPUs. Instead of DMI with four lanes (which is equivalent to PCIe 3.0 ×4), Rocket Lake connects to the 500-series chipsets over eight lanes, equivalent to PCIe 3.0 ×8. Thus, bandwidth increased from 4 GB/s to 8 GB/s (in real life it is less due to overheads).
Rocket Lake to not work on boards with H410 and B460 chipsets
But this seems to be a source of some incompatibility. According to Intel, Rocket Lake processors will eventually only work on older Z490 and H470 chipsets for those who will be upgrading from Comet Lake – because these chipsets have newer 14nm silicon. However, H410 and B460 chipsets use a different, older 22nm silicon and, according to official information, are simply not compatible. So on LGA 1200 boards using H410 or b460 it will likely not be possible to use Rocket Lake CPUs.
We’ll see if motherboard manufacturers manage to work around this limitation somehow, but if you’re planning to buy a computer based on the LGA 1200 platform now and you’re not going to buy a 500-series board, keep this issue in mind. The new B510 and B560 chipsets should not have this problem.
First preview of Alder Lake
Intel also showed an ES sample of the Alder Lake processor in action – these will be Intel’s first desktop processors with 10nm chips, alreadyusing DDR5 memory and new LGA 1700 platform. The processor was not visible, but Intel showed an operating computer running with it. The processor was cooled by a stock cooler which suggests it was a 65W CPU (but this does not rule out it can reach higher power draw during boost).
The Alder Lake processors and LGA 1700 platform are expected to be available later this year, so the Rocket Lake, to be released in March, might be be replaced relatively quickly: Intel has once again confirmed that it plans to release Alder Lake in the second half of this year. The CPU core used in Alder Lake is again supposed to bring a significantly higher IPC, so we can expect an improvement in single-threaded and gaming performance even beyond what Rocket Lake is bringing to table.
English translation and edit by Lukáš Terényi