Intel’s Rocket Lake Blasts Off With Fewer Cores, Higher Pricing

Intel provided a few benchmarks to underline its gaming leadership claims, but due to some disparities we noted with the test setups, you’ll need to take these with a heaping spoonful of salt.

Intel listed the PL1 rating, which is the power consumption measured at base frequencies, as 250W for all Rocket Lake processors in its gaming tests. This represents a 100W increase over Intel’s 150W PL1 spec (250W is the PL2 rating -power during boost), which serves as a form of cheaper-yet-sanctioned overclocking. Intel also didn’t specify the Tau (boost duration) used for the tests (we suspect it is unlimited), making these results suspect.

Meanwhile, the company assigned the competing Ryzen processor at its standard 105W PL1 and the Comet Lake models to a 150W PL1, meaning we’re looking at a lopsided battle where the newer Intel processor is free to draw more than twice the power with essentially an unlimited boost duration. We’ve included the test notes with Intel’s detailed configurations at the end of the article.

Intel outfitted all test systems with 32GB of memory set to the maximum supported speeds and with tight timings (14-14-14-34), an EVGA RTX 3080 XC3 Ultra, and an unspecified watercooler.

Intel compared the Core i9-11900K to the Ryzen 9 5900X, which currently tops our CPU benchmarks as the fastest gaming chip on the market, in a narrow selection of titles at 1080p High settings. Intel claims a 10% advantage over the Ryzen 9 5900X in the Total War Kingdoms Dynasty benchmark, 9% higher fps in Gears 5, an 8% advantage in GRID 2019, and an 11% lead in Microsoft Flight Simulator 2020.

Intel also presented gaming benchmarks showing the gen-on-gen performance gains of the Core i9-11900K against the 10900K, and the Core i5-11600K against the 10600K. As you would imagine, Intel recorded strong generational gains. However, while the Rocket Lake processors were apparently dialed into a 250W PL1, the Comet Lake chips are listed at their default 125W PL1 rating, meaning these results are likely skewed, too.

Intel’s final slide plays up its strong partnerships with several game developers, through which it helps optimize their game engines for Intel processors. Bear in mind that three of the four titles used for Intel’s gaming benchmarks are listed. For a perspective of the current state of gaming performance, we included a few of our own gaming benchmarks at the end of the album with the current Comet Lake and Ryzen lineups.

Intel 11th-Gen Core Rocket Lake Application Benchmarks

Intel also presented several benchmarks in productivity applications, but with a heavy focus on software that uses AI inference through targeted code optimizations that leverage Rocket Lake’s hardware-accelerated DL Boost and VNNi technology. As such, the Rocket Lake chips naturally have an easy advantage over competing chips that don’t support the technology.

Again, Intel provided benchmarks to show its own generational performance gains, and also to measure performance relative to the Ryzen 9 5900X. The Photoshop Photo Creation benchmark runs a colorizing workload on seven images and leverages Intel’s inference tech, while the Magix Vegas Pro video creation workflow utilizes DL Boost and OpenVINO, with both granting large performance gains over processors without hardware-accelerated AI features. MLPerf is also a perfect test to highlight hardware-accelerated advantages over competing chips.

Likewise, Nero’s photo tagger feature also leverages Intel’s DL Boost, but curiously, Intel only used it for comparisons to its own previous-gen models, and not the Ryzen 9 5900X.

These benchmarks do highlight impressive gains in a few applications that are optimized to exploit the advantages of Intel’s hardware-accelerated AI suite, but many of these programs only use inference for a subset of their overall capabilities, and industry uptake will take some time to expand out to other types of software.

We can’t say that these tests are indicative of the typical performance you can expect in most content creation and productivity applications, and Intel’s reluctance to share results in a broader spate of typical workloads isn’t very encouraging.

Intel’s Rocket Lake Die Shots and Comparisons

Here we can see our Core i9-11900K and Core i5-11600K samples next to two prior generations of the same series. The integrated heat spreader (IHS) and PCBs appear largely identical from the top, though the 9th-Gen processors have slightly slimmer ‘ears’ flanking the IHS for the socket retention mechanism to clamp down on. A closer look reveals that Intel has reduced the height of the IHS for the Rocket Lake chips by roughly 0.5mm, probably due to further die thinning or to reduce the amount of solder between the die and IHS, thus improving thermal dissipation.

This adjustment comes as a byproduct of Intel’s increased power consumption with each new generation of 14nm chips, which necessitates more aggressive techniques to dissipate thermal load. Intel had already reduced the thickness of the PCB with the 8th-gen models, leaving little room for additional thinning of the PCB, and then turned to thinning the die for the 10th-gen Comet Lake chips to help improve cooling. Rocket Lake’s addition of AVX-512 support, which trades higher power consumption for more efficiency, obviously exacerbates cooling issues.

Flipping the chips over reveals the same LGA pad arrangement for both Comet Lake and Rocket Lake, which is expected given that they drop into the same LGA1200 socket. We also notice a different arrangement of surface mount devices (SMD) in the center of the chips.

We aren’t prepared to potentially sacrifice our sample by doing a risky delid yet, but luckily a daring enthusiast recently delidded an Core i7-11700K processor and shared the results at the Overlock.net forums. Unfortunately the processor didn’t live through the procedure, but we can get a good sense of the size of Rocket Lake’s die through a photo lineup compiled from Der8auer’s excellent die analysis article (via @harukaze5719).

Intel has confirmed that all Rocket Lake-S chips come with the same eight-core die, with two cores disabled for the six-core models. Surprisingly, based on our rough projections, Rocket Lake’s eight-core die is about 34% larger than the ten-core Comet Lake die.

Generation	Model	Cores	Die Size
Rocket Lake-S	Core i7-11700K	Eight	276.4 mm2
Comet Lake-S	Core i9-10900K	Ten	206.1 mm2
Coffee Lake-S Refresh	Core i9-9900K	Eight	180.3 mm2
Coffee Lake-S	Core i7-8700K	Six	153.6 mm2

Intel chalks much of this disparity up to GPU and CPU cores that are physically larger than those found on Comet Lake, a byproduct of backporting from 10nm to 14nm, along with the increased number of graphics EUs. Intel could have shrunk or removed the integrated graphics engine to cram in more CPU cores to meet its power, performance, area, and cost (PPAC) targets, but integrated graphics are a key requirement for the high-volume OEM systems market that tends to leverage on-chip graphics rather than discrete GPUs.

Intel’s 14nm process is ultra-mature, so we assume yields are well in line and the majority of its chipmaking equipment is depreciated. However, fabbing a larger die exposes the chips to a greater chance of defects, thus complicating the binning process, and also results in fewer die per wafer. All of these factors increase cost, which could help explain Intel’s higher pricing for its highest-binned SKUs.

Intel 11th-Gen Core Rocket Lake Overclocking

Intel recently discontinued its Performance Tuning Protection Plan (PTPP), an extended warranty plan that offered protection from damage that occurred during overclocking. That led to plenty of commentaries that overclocking is dying, but Intel says differently and has increased its overclocking capabilities, like adding support for memory overclocking on lower-end 500-series B- and H- chipsets.

Intel has also revamped its eXtreme Tuning Utility (XTU) with a new, fresher look and added new features to enable Rocket Lake’s new overclocking options. That includes new overclocking offsets, like a separate AVX-512 offset and the ability to set voltage guardbands for the different flavors of AVX. Intel also added an option to completely disable AVX support, though that feature is primarily geared for professional overclockers.

Dan Ragland, Intel’s principle overclocking engineer, says that most users can expect similar overclocked CPU frequencies to the 10th-Gen Comet Lake processors. That equates to a common ceiling of a 5.2 GHz all-core overclock (with AVX offsets applied) on most Rocket Lake chips, though the silicon lottery does apply. With more robust cooling, Ragland claims users with cherry chips can add a few more hundred megahertz to that total, albeit on a limited number of cores.

Although Rocket’s overclocking ceilings are very similar to Comet Lake, the increased IPC should result in larger performance gains, comparatively. Like the Comet Lake chips, Rocket also supports per-core frequency and hyper-threading control (enable/disable) to help eke out more overclocking headroom.

Intel also offers revamped memory overclocking features, like the Gear 1 and Gear 2 modes we covered earlier. Aside from being used as a somewhat questionable tactic to segment Intel’s product stack, these settings also allow overclockers to reach higher memory frequencies.

Intel has also added support for real-time memory frequency adjustments, though motherboard support will vary by model and vendor. This feature allows you to shift from, say, DDR4-2933, for example, to DDR4-3200 from within Windows 10 without rebooting. Intel also continues to support its existing mechanism for live memory timing adjustments from within the operating system, giving users a plethora of on-the-fly memory overclocking options. Overall, Ragland says he thinks that Rocket Lake’s memory controllers, which are entirely new, coupled with the expanded memory overclocking features, will help topple many of the existing world records.

Intel’s overclocking headroom, while shrinking, still tends to be a big advantage over AMD’s chips. Intel also helped develop new sub-ambient coolers from EKWB QuantumX Delta TEC and Cooler Master MasterLiquid ML360 Sub-Zero to bring extreme cooling to the masses, but these are somewhat expensive and exotic solutions for very specific uses, which you can read about here

Intel 11th-Gen Core Rocket Lake Xe Graphics