Intel Teases Rocket Lake Core i9-11900K, Intends to Retake Gaming Crown With 19% IPC

Attachment 46441

Intel fired back at AMD's Ryzen 5000 today at CES 2021 with its Rocket Lake-S flagship, claiming the new chip takes back the gaming performance crown. Intel says that its 14nm Rocket Lake processors will come to market this quarter with a 19% IPC increase born of the new Cypress Cove architecture, matching AMD's IPC increase with its Ryzen 5000 series processors. Intel pairs this new backported architecture with its 14nm process, touting boost speeds up to a dual-core 5.3 GHz, all of which the company says will wrest the gaming lead back from the fastest gaming chip on the market, AMD's potent 5900X, and perhaps shake up our Best CPU for Gaming and CPU Benchmark Hierarchy in the process.

Intel also claims the new chips, courtesy of the new Xe Graphics architecture, will deliver a 50% gen-on-gen increase in integrated GPU gaming performance.

Intel has slowly teased bits of info about Rocket Lake, but here's probably the most important new bit of details: Intel says the 8C/16T Core i9-11900K reclaims the gaming performance crown from AMD's Ryzen 9 5900X at 1080p. The margins are slim, though. Intel ran these tests with an EVGA RTX 3080 XC3 graphics card, and the company's claimed lead ranges from 2% to 8% (roughly a 4% advantage overall). That paints a picture of a closely-contested battle in gaming performance between Rocket Lake and Zen 3, particularly at higher resolutions.

Intel's traditionally higher overclocking ceilings could prove to be an advantage against the Ryzen 5000 processors — it's important to remember that Intel tested the Rocket Lake chip at stock settings. If these slim deltas play out in our testing and the Intel chips overclock well, pricing might be the determining factor if you're on the hunt for a gaming processor in 2021. Unfortunately, Intel hasn't shared pricing or availability information yet (the chips are rumored to land in mid-March). Still, the company has detailed a few other new features for the Rocket Lake series.

Attachment 46442

Attachment 46443

Much of Intel's Rocket Lake disclosure covers information we already knew, but there are a few new tidbits. Intel's Core i9-11900K is the company's Rocket Lake flagship and comes with 5.3 GHz dual-core and 4.8 GHz all-core boost frequencies, matching the previous-gen 10900K. It's noteworthy that these are Thermal Velocity Boost frequencies that only activate if the processor is under a certain temperature limit, but most motherboard makers ignore those limits anyway. That means the chips will likely operate at these speeds regardless of temperature limits, at least on higher-end motherboards.

The 11900K comes with only eight cores and sixteen threads, a step back from the 10-core Comet Lake i9 models. Intel says the eight cores 'maximizes real-world performance,' but that decision also stems from the backported architecture.

"Backporting" is a method that allows Intel to take a new design built on a smaller process node, in this case 10nm, and etch it on an older, larger node (in this case, 14nm). Intel backported the 10nm Sunny Cove cores found in Ice Lake processors to the less-dense 14nm process to create the Cypress Cove architecture in Rocket Lake, leaving the chip designers with fewer transistors to make the new chips. As a result, Intel had to remove two cores; there simply wasn't enough room in the chip package for a larger die.

Intel claims that Rocket Lake's IPC and frequency improvements offset the removal of the two cores, implying the chip will offer the same performance in threaded work as the ten-core 10900K. The company has previously shared that the chips feature a 150W PL1 power rating (at the base frequency), a 25W increase over the 10900K, and an identical 250W PL2 (boost) rating. Intel also says the chips come with new overclocking features that it will explain in the future. We’re pretty sure that includes support for the new sub-ambient coolers from EKWB QuantumX Delta TEC and Cooler Master MasterLiquid ML360 Sub-Zero that Intel helped design to bring exotic cooling to the masses.

The Rocket Lake-S chips are backward compatible with most 400-series chipsets, and the PCIe 4.0 connection will work on existing motherboards that support the interface. Rocket Lake-S isn't compatible with the H410 and B460 chipsets. Most Comet Lake chips are forward-compatible with the new 500-series motherboards that will debut today, the lone exception being Celeron models with 2MB of CPU cache.

Rocket Lake also brings the debut of Intel's Xe LP graphics for its desktop chips. Intel says it added a 'third more' EUs to the chips to boost performance up to 50% over the previous-gen UHD 630 graphics. It isn't clear if the increased EU count is over the existing UHD 630 graphics, which would mean we're looking at chips with a maximum of a lackluster 32 EUs, or if it represents an increase over the Tiger Lake chips, which would beef up the Rocket Lake processors with 128 EUs.

We hope it's the latter, as 32 EUs would hamper gaming performance significantly. On the other hand, doing 128 Xe EUs on 14nm would use up a considerable amount of die space, and most desktop users (at least in the DIY space) will simply use a dedicated GPU anyway. Notably, Intel's test notes say the margin of error for its performance claim is +/- 15%, so we'll have to wait for real-world testing. Intel's Xe graphics also bring in-built AV1 decode acceleration, which is helpful because the codec reduces bandwidth up to 20% for video streaming (such as 4K and 8K content).

Intel also bumped up memory support from DDR4-2933 to DDR4-3200, matching AMD's Ryzen, and added support for 20 lanes of PCIe 4.0, a needed addition that comes two long years after AMD led the industry with the first PC chips to support the interface. Intel also reworked the internal PCIe subsystem to accommodate the x4 direct connection (the chips now support 20 lanes) for M.2 SSDs to the CPU. Intel also widened the DMI 3.0 connection (the pathway that connects the CPU and chipset) from four lanes to eight, doubling throughput up to a theoretical ~7.86 GB/s.

Rocket Lake's wider x8 DMI connection is only active on 'select' 500-series chipsets, so the chip defaults to a x4 connection on B560 and H510 motherboards. Comet Lake chips also only use a x4 connection on all 500-series motherboards, and the same x4 connection applies for Rocket Lake processors in 400-series motherboards.

The wider DMI connection should help with bottlenecks for devices attached to the chipset, like SSDs in RAID. However, the most recent unofficial information points to PCIe 3.0 support for devices connected to the chipset, and not PCIe 4.0. We do know that the chipset now features an integrated USB 3.2 Gen 2x2 controller, doubling support over the existing interface to 20Gbps.

Intel also demoed the new Hitman 3, an Intel-sponsored title that supports a new feature that tests your CPU and adjusts settings 'on Intel hardware,' dialing up features like destruction physics, NPC density and 3D spatial audio through the engine's 'brick' system. As a result, this system turns on automatically for PCs that have a CPU that passes a certain performance threshold.

Intel claims the 11th-Gen Rocket Lake i9-11900K offers up to 7% more performance in Hitman 3 than the 10th-Gen Comet Lake 10900K, but didn't specify if the performance jump was due to the new features, or if that stems from the natural benefits of Rocket Lake's higher performance. Hitman 3 comes to market this month.

Oddly, Intel also touted its 'new' Always-On QuickSync Video. This feature allows Xe Graphics to run concurrently with a discrete GPU so you can offload some workloads, like video streaming with QuickSync, to alleviate the burden on your discrete GPU. This feature has already been around for several years, but it required BIOS support, and you had to manually enable the integrated graphics unit after you installed a video card. Intel says the technique is now a supported and validated configuration that's enabled by default in the BIOS.

Rocket Lake also supports Intel's Deep Learning Boost (DLBoost) and VNNI features, which require support for AVX-512 instructions, to speed up AI workloads. Intel has helped enable the software ecosystem to better take advantage of the explosive performance benefits. Adobe's recent suite of updates is a prime example of the improvements we can expect as AI becomes more broadly adopted for PC workloads.

Intel hasn't shared detailed specifications, launch dates, or pricing, though the company did say that it was launching eight new high-performance desktop PC chips. We'll update as more information becomes available.