With their 11th generation processors, Intel is finally introducing a new architecture to replace the Skylake arch that has been dragged along forever. At the same time, they're adding support for PCI-Express 4.0 and new instructions, like DLBoost for deep learning and AVX-512 for SIMD calculations. The Core i9-11900K is the company's flagship, and it's probably the first time in the history of processors that the core count is going down. While the Core i9-10900K had ten cores, the Core i9-11900K "only" has eight. Eight is still plenty as we saw throughout our benchmarks, but it's still a surprising development. There are several underlying reasons: the 14 nm process, iGPU requirement, larger iGPU, larger Cypress Cove cores, limited space on LGA1200—Intel really didn't have much choice, they had to compromise somewhere. They made no compromises with clock speeds, on the other hand. Owing to various boosting algorithms, the Core i9-11900K boosts up to 5.3 GHz—a lot. Thermal Velocity Boost now works much better and is sustained for longer, and overclockers also get some tweaking knobs to adjust its operation, as opposed to the black box previously.

With our Rocket Lake reviews we're also introducing our new CPU Test Suite, which is now comprised of 38 (!) benchmarks that cover the whole spectrum of workloads, from consumer to scientific, content creation, and enterprise. The applications tested are a healthy mix of single-threaded, lightly-threaded, and fully multi-threaded workloads—just like you would encounter in real life. Averaged over all these tests, we find the Core i9-11900K beating the Core i9-10900K by around 3%. Really impressive if you consider this is an 8-core vs. 10-core duel. When looking at individual applications, we see impressive gains due to the new architecture. In other apps, the results are less impressive, though. It seems Cypress Cove is sometimes able to make up a two-core deficit against Skylake, but not always. When comparing against AMD's recent Zen 3 releases, the Core i9-11900K can't impress. Even in the equal match-up against the 8-core Ryzen 7 5800X, the i9-11900K loses by 3%. Against the 5900X and 5950X, the gaps are big, 18% and 28% respectively. However, application choice matters a lot. Certain workloads that do scale somewhat, only to a few cores, mostly from the "creators" ecosystem, do run very well on Rocket Lake and can often beat much more powerful AMD processors. Just like other Intel CPUs, raising the power limit can make a huge difference—we gained an average of 7.2% in applications, which is very nice.

What could bring big wins for Intel is the newfound love for AVX512 and DLBoost—extensions that have been available for years but never made it to the desktop. At this time, software support for either of those instruction sets is extremely limited, and they are not useful. I am convinced that they can offer tangible benefits once adoption rates go up, though. Remember AVX—everybody said it's a useless tech that's not needed as "we already have SSE"; today, a lot of apps and games use AVX, also because of excellent compiler support. Using these new instructions is often as simple as checking a box that tells the compiler it may optimize with AVX instructions—that's it. The hard work will be done by the compiler, you don't have to mess with hand-coded assembly instructions. Today, all this doesn't matter as consumer won't be needing AVX-512 for a couple of years at least. That said, Rocket Lake can be a cost-effective option for researchers and industry professionals who want to use these new instructions to speed up their calculations, but don't want to pay up for the expensive Xeons.

Gaming performance is surprisingly low with Rocket Lake. Everybody expected it to beat Comet Lake, especially in gaming. My numbers show the opposite. The Core i9-10900K is clearly faster than its 11th generation counterpart despite PCI-Express Gen 4 support on the latter. That the Zen 3 architecture is able repel the Rocket Lake attack is a huge win for AMD—gaming performance is still better on AMD. Once you go to higher resolutions, differences between processors become tiny, a percent here or there really makes no difference. Spend wisely—if you save money on the CPU, you might have enough left over to be able to afford one of the new GPUs.

Just like their other recent desktop processors, Rocket Lake is fabricated on Intel's 14 nanometer process. The company does have a working 10 nm process, which is in production for Ice Lake, and everybody is wondering why they didn't use it for Rocket Lake. Silicon manufacturing has extremely long lead times, so I suspect they decided on 14 nm well before AMD's Zen 3 came out to make sure their older fabs continue to see full use. The excellent yields on the mature 14 nm process certainly played a role in this business decision, too. The drawback of this aging technology is that power consumption is much higher than what AMD offers.

Energy efficiency has definitely suffered with Rocket Lake even compared to Comet Lake, which wasn't impressive either. Looking at our power consumption results, we have a total energy of 18.4 kiloJoules to complete one Cinebench run for the 11900K, against 7.8 kJ for the Ryzen 5900X, which is less than half. AMD's Ryzen 9 5950X is only a third of that, with 6.4 kJ. This "total energy consumed for Cinebench" metric is important because it takes the processor's speed into account, not only momentary power usage in watts.

As expected, Intel is using their full arsenal of boosting algorithms to increase the processor's clock frequencies far beyond the base frequency of 3.5 GHz, which seems to be more of a guaranteed minimum, possibly also for legal reasons, than a frequency you'll ever encounter in real-life. Actually, the i9-11900K holds its boost frequencies very well, much better than the i9-10900K. In our Comet Lake reviews, I complained that Thermal Velocity Boost was active only for split seconds, so it never really made much of a difference. With Rocket Lake, TVB works very well and definitely helps with low-threaded workloads. It's a shame that this feature is only available on so few processor models.

Kind of at the last minute, Intel added "Adaptive Boost Technology," a new boost method that helps in scenarios where the CPU is loaded with multiple threads that use up a lot of processor time. Previously, a hard limit existed for "8 cores active," for example; 4.7 GHz on the i9-11900K. With Adaptive Boost active, the processor can now clock up to 5.1 GHz on all cores if given enough headroom. Why is this so awesome? Because it uses the default voltage-frequency curve, which means it's guaranteed to be stable at all times, and it doesn't void the warranty, either. We tested this feature and found that it requires a power limit increase, and even then the difference is only around 1%. The price you're paying is extremely high energy usage and temperatures. In Blender, we saw temperatures skyrocket to 110°C with a Noctua NH-U14S. No doubt there's plenty of headroom to tweak things, and everybody will have to get to know the new platform and its kinks first before making the most out of it.

Rocket Lake finally brings with it PCI-Express 4.0 support, which is an essential capability gamers demand these days even though it doesn't make that much of a difference, neither for graphics nor storage. With the much more modern platform, AMD is obviously capitalizing on Intel's shortcomings, even if it's just marketing. While many pieces of the Rocket Lake puzzle are now running at PCI-Express 4.0, the chipset still puts out Gen 3 lanes, which means you're limited to one PCIe Gen 4 M.2 NVMe SSD; the other slots support Gen 3 only. Praise also goes out to Intel for letting us use faster memory not only on Z590, but also on lesser chipsets, which are often good enough and have motherboards that are priced much more reasonably.

What I do like is that integrated graphics are included on many Rocket Lake processors—plenty of scenarios don't need a discrete graphics card, which makes for possible cost savings, too. While we haven't had the time to test the Iris Xe integrated graphics solution based on the Gen12 Xe LP architecture, this solution has many on-paper improvements. Intel claims it's 50% faster and supports the latest HDMI 2.0 with HBR3, which supports 4K @ 60 Hz for non-gaming entertainment. The iGPU also offers hardware acceleration for AV1 and HEVC video formats, which should make these very potent chips for HTPCs.

What really displeases me is how tacked together and unfinished the whole Rocket Lake platform feels. The BIOSes have numerous bugs that are compeltely obvious to anyone using them for more than 10 minutes. Maybe this is not Intel's fault, but since AMD introduced AGESA, a common-base software stack, things have gotten much better for the red team. Ryzen Master has also improved so much and is on a completely different level than XTU. POST times have always been good with Intel, but I'm now sometimes sitting at A2 (VGA) for 20 seconds, and there's occasionally a double boot when changing a BIOS setting, items we criticized AMD for in the past. This whole experience reminds me of the first generation of Ryzen. With Rocket Lake, Intel is introducing a new memory controller that offers a setting of "Gear 1" and "Gear 2." In Gear 1, the memory controller will run at the same clock as the memory, and in Gear 2, it runs at half the clock. Gear 2 is designed for ultra-high memory overclocks, or so I thought. My sample doesn't even run DDR4-3800 in Gear 1, so it seems only up to DDR4-3200 MHz is guaranteed for the 11900K(F), and the other SKUs are guaranteed only 2933 in Gear 1. Running Gear 2 adds a latency penalty—I was curious how much of one, so I ran our whole test suite at 3200 CL14 with Gear 1 and Gear 2. The difference is around 1% on average, with much bigger swings depending on the application. Surprisingly, 3200 CL14 Gear 1 is actually faster than 3800 CL16 Gear 2. The whole BIOS situation is quite fluid at this time, so I hope Intel figures out some additional memory optimizations because I really don't want to go back to the days of raising BCLK just to achieve the desired memory speed.

At $550, the Core i9-11900K is roughly priced the same as the Core i9-10900K. At that price point, there is simply no way this is a worthy investment. You're much better off spending that money on the Ryzen 5900X, which costs $550, too. Huge competition also comes from Intel's Core i9-10850K, which costs only $380 and basically matches the Core i9-11900K. Another interesting alternative is AMD's Ryzen 7 5800X, which isn't as overpriced as the 5900X, yet offers better gaming performance and enough application performance for nearly all usage scenarios. If Intel can bring down pricing of the Core i9-11900K, it definitely has the potential to become an interesting option, I did a quick poll here with the team, and the price points that came back were between $380 and $420. Do check out our other review of the Core i5-10600K, which we also published today. At $275, it is priced much more competitively and offers excellent performance in both apps and gaming. I really wanted to like the Core i9-11900K because what we need is competition or AMD will keep raising prices until they become the next Intel. With AMD staying ahead in many respects, especially energy efficiency, Intel will hopefully make its next processor on 10 nm.