Intel Core i9-12900K review: Intel’s return to form

Intel’s “Rocket Lake” lineup released last year was essentially dead on arrival. Its marginally improved single-threaded performance was diminished by its unjustifiably higher cost, excess heat, the requirement for a new motherboard and lower core count on the flagship model.

Its main competitor, however, was winning. AMD’s Ryzen 5000 series processors, leveraging the Zen 3 architecture and TSMC’s 5nm transistors, became the go-to choice for system builders. Although Intel’s chips remained competitive in gaming with their high single-threaded performance, their lacklustre multithreaded performance hampered their appeal for productivity applications. In short, the company needed a win.

The culmination of Intel’s efforts is Alder Lake. With a hybrid design, new features, and a move onto the Intel 7 node, Alder Lake certainly looks promising. We took a look at the flagship Core i9-12900K to see how far Intel has come in just one year.

Table of contents

1. Feature overview
2. Chipset and motherboard
3. Test setup
4. Performance
5. Power and thermals
6. Overclocking
7. Conclusion

Intel Alder Lake 12900K features

Alder Lake is a fusion of a new architecture, more features, and a new node. Intel has always had different processor cores for its various product segments, but this is the first time two have existed in a single package.

A big part of this is thanks to the company finally shifting to the Intel 7 node transistor process, previously known as 10nm++. Leveraging its higher density and increased energy efficiency, Alder Lake naturally was able to stuff in more features.

At centre stage is its unique core complement. On the flagship Core i9-12900K, Intel paired eight “Golden Cove” performance cores with eight “Gracemont” efficiency cores, a design reminiscent of today’s smartphone system-on-chips (SoC).

There are plenty of improvements under the hood. The new cores feature deeper, wider processing pipelines, reworked and expanded cache designs, new instruction decoders, better branch prediction, and more. All the processors’ subsystems are held together by the compute, I/O and memory fabrics. Lastly, the new Intel Thread Director helps to most effectively allocate instructions to the appropriate cores.

Intel said during its Architecture day that Alder Lake’s P-cores offer 19 per cent higher performance on average over its 11th-gen “Cypress Cove” cores at the same frequency. For a deeper dive into the various core functions, read our breakdown coverage of the Intel Architecture Day.

Like Rocket Lake, Alder Lake also has “favoured cores” that can boost to higher frequencies due to their better quality. Two of the eight P-cores can boost to 5.2GHz while the rest can only reach 5.0GHz. However, there are no favoured E-cores; their maximum frequency is capped at 3.9 GHz by default.

Rocket Lake introduced memory gears that set the frequency at which the integrated memory controller (IMC) operates in relation to the frequency of the main memory. At Gear 1, the IMC matches the memory’s frequency in a 1:1 ratio. Gear 2 halves the IMC frequency to 1:2. The motivation behind this is that Gear 1 offers the best latency and Gear 2 allows for higher memory frequencies. Alder Lake takes this concept even further by introducing Gear 4, quartering the IMC’s frequency to 1:4. Unfortunately, our test platform’s Gigabyte Aorus Z690 Pro motherboard does not support Gear 1 for DDR5; Gear 2 is the highest setting it offers.

Intel’s Extreme Memory Profile (XMP) gives users an easy way to enable high-performance profiles based on tested settings. Alder Lake now supports up to five XMP profiles for enthusiast overclockers.

The 12900K’s features Intel HD Graphics 770. Based on the Intel Xe graphics cores, the HD Graphics 770 supports up to four DisplayPort or HDMI ports. Relying on integrated graphics isn’t ideal for such a high-end system but it will help to fill the gap for some people, given the state of the ongoing semiconductor shortage.

Chipset and motherboard

Alder Lake only fits into the new LGA1700 socket, which means buyers will also need to purchase a new motherboard. Because the socket is shaped differently than the LGA1200 socket used for Rocket Lake, users will also need to either get an adapter bracket for their current cooler or buy a new one. Luckily. Many CPU cooler manufacturers are selling or providing adapter kits for free.

Along with a new socket is the new Intel 600 series chipset. Motherboards with the new 600 chipsets have three big-ticket features: DDR5 memory, a wider Direct Media Interface (DMI), and PCIe 5 support.

Intel has started to transition to DDR5 for its future platforms. DDR5 offers increased memory chip density, lower operating power, and doubles the bandwidth of DDR4. For example, even a basic kit of DDR5 memory will start at around 4,800MHz; that’s the upper end of what DDR4 is capable of. They’ll also operate at 1.1V, as opposed to DDR4’s 1.2V. Note that DDR5 memory kits are generally more expensive than DDR4 at launch and users are unlikely to see massive performance gains just yet.

Alder Lake also brings PCIe 5 support. Granted, consumer devices are still a way off from fully saturating the bandwidth PCIe 4 provides, so it’s going to be a while before they can realize the benefits of PCIe 5.

According to Z690’s block diagram above, only the 20 PCIe lanes from the CPU are PCIe 5. The ones routed from the chipset are still PCIe 4 and PCIe 3.

More users today are purchasing high-end PCIe SSDs, so to support these high-end storage devices at their full speeds, Intel has doubled the DMI lane width on the motherboard from four to eight. DMI is the data interface between the processor and the chipset, otherwise known as the north and south bridges. The new revision, called DMI 4.0, supports 2GT/s per lane for 16GT/s in total.

There’s no lack of connectivity options either. The Z690 chipset offers up to 12 PCIe 4 lanes, 16 PCIe 3 lanes, and eight SATA 6 ports. It also supports four 20Gbps USB 3.2 Gen 2×2 lanes ports, up to ten 10Gbps USB 3.2 2×1 ports, ten 5Gbps USB 3.2 1×1 ports, and 14 USB 2.0 ports. As always, motherboard manufacturers will select the appropriate ports according to the product tier.

Test systems

Intel Core i9-12900K

Intel Core i9-11900K

Intel has renamed the processor power levels in this generation, officially moving away from using TDP to represent power consumption and thermal dissipation. The number has become increasingly convoluted over the years. Between the various processor and cooler manufacturers, it seems like everyone has a different formula for calculating TDP. The new definitions are as follows:

TDP is now called Processor Base Power (PBP). It’s defined as the time-average power dissipation that the processor is validated to not exceed during manufacturing while executing an Intel-specified high complexity workload at Base Frequency and at the junction temperature as specified in the Datasheet for the SKU segment and configuration.

PL2 is now called Maximum Turbo Power (MTP): It’s defined as the maximum sustained (>1s) power dissipation of the processor as limited by current and/or temperature controls. Instantaneous power may exceed Maximum Turbo Power for short durations (<=10ms). Maximum Turbo Power is configurable by system vendor and can be system-specific.

In its reviewer’s guide, Intel’s testbench set the PBP equal to MTP of 241W, essentially giving the chip an unlimited boost duration for as long as it stays within a temperature and current limit. This ran counter to previous generations whereas the company recommended a turbo boost duration.

Performance

I tested the 12900K at four power levels: 125W (PBP), 241W (MTP), the motherboard’s MCE settings, and a manual overclock.

The 11900K has all of its boost modes enabled, that’s Adaptive Boost, Thermal Velocity Boost, Turbo Boost Max, and Turbo Boost 3.0. Adaptive Boost and Intel Thermal Velocity Boost are not verified for Alder Lake; it only features Turbo Boost Max 3 instead.

Cinebench R20

Cinebench R20 paints a snapshot of the processor’s performance using Maxon Cinema 4D’s Physical rendering engine by measuring how quickly it can render a 3D scene. The instructions use AVX.

Unsurprisingly, the 12900K’s extra eight cores gave it the clear lead over the 11900K in the multicore test. But even in the single core test, the 12900K’s Golden Cove P-cores are around 20 per cent faster than the 11900K’s Cypress Cove cores. The efficiency of its new architecture and the Intel 7 node is on full display here.

Cinebench R23

Cinebench R23 is an upgraded iteration of Cinebench R20. For this bench, instead of rendering the scene just once, we run the test for 10 minutes to gauge the processor’s average performance over a prolonged heavy workload.

We saw a similar result here. At its PBP of 125W, the 12900K was 46 per cent faster than the 11900K with its power limits removed. When just one core was active, neither processor touched their power limits, thereby repeating the 20 per cent gap we saw in Cinebench R20.

SPECworkstation 3

Developed by the Standard Performance Evaluation Corporation, SPECworkstation 3 benchmarks a system using a variety of professional applications. The seven tests used for this review are all CPU-dependent. I’ll be following up this review with a description of their respective workloads.

Once again, the 12900K pulled ahead of the 11900K across the board. The greatest difference was seen in Calculix, a test that uses the finite element method to simulate the internal temperature of a jet engine, where the 12900K more than doubled the 11900K’s score. The closest the 11900K came to matching the 12900K was in Octave, an open-source programming language used for developing machine learning and numerical methods, though it still trailed by around 10 per cent.

UL PCMark 10

UL PCMark 10 measures overall system performance by simulating diverse day-to-day use cases such as web conferencing, writing, web browsing, media editing and more. It’s designed to touch on almost all aspects of the device.

Because purely CPU-dependent tests only account for a small portion of the score, the 12900K’s lead wasn’t as dramatic. With that said, it still had a solid edge in specific categories such as video sharpening, deshaking, and spreadsheet recalculation tasks. The largest gap was seen in digital content creation, where the 12900K at 125W led the stock 11900K by 18 per cent, and 21 per cent when both chips had their power limits removed.

UL Procyon – Office Productivity Benchmark

Procyon is a new benchmark used for evaluating system performance using a range of real-world software. Its Office Productivity Benchmark suite measures the system’s performance in Microsoft Office 2021.

The 12900K appears more impressive than the 11900K on paper, but in reality, neither processor bottlenecked user experience in the Microsoft Office suite.

Puget Systems – DaVinci Resolve

DaVinci Resolve is a popular video editing application that’s a direct competitor to Adobe Premiere. Puget Systems’ Resolve benchmark tests the CPU, GPU, and a blend of both in a variety of scenarios. The CPU tests focus on exporting a set of 4K and 8K media using the H.264 and DNxHR codecs. CPU performance also plays a role in the Fusion test, which encompasses a mix of 3D, green screen and motion graphics workloads.

On average, the 12900K was 20 per cent faster over the 11900K.

Puget Systems – Lightroom Classic

Adobe Lightroom Classic is a staple photo editing software for many professional photographers. The Puget Systems’ benchmark tests the application’s performance using RAW images taken on three different professional cameras.

Switching to the 12900K increased the Lightroom score by around 45 per cent. It was faster in all metrics, but Image preview build time saw the most dramatic reduction as the 12900K managed to cut it down by more than a third compared to the 11900K. The 12900K also demonstrated impressive performance in JPEG exports by completing the same task in half the time as the 11900K.

7-Zip

7-Zip is one of the world’s most popular archiving tools. It also includes a benchmark suite that tests the processor’s compression and decompression capabilities, measured in millions of instructions per second. The tool also scales well with high core counts.

Again, no contest here; the 12900K trounces the 11900K. Because 7-Zip returns a score within a minute of starting the benchmark, it doesn’t show the effects of the frequency dropoff when the stock 11900K reaches Tau (56s by default). Its sustained performance is likely worse than the score shown here.

POV-Ray 3.7

The POV-Ray raytracer has helped to create countless gorgeous graphics for 30 (!!) years. Its built-in benchmark ray traces a pre-defined scene, producing an output in pixels per second. It features both single and multithreaded benchmarks that scale well across multiple cores.

At 125W, the 12900K was already 56 per cent faster than the 11900K with its power limit removed. When dialled to its peak, the 12900K widened that gap to a whopping 80 per cent. It also led in single-threaded performance by 14 per cent versus the no-power limit 11900K at 125W.

Blender

Blender is an open-source 3D graphics design tool for creating animation and visual effects. It features two rendering engines: the Cycles raytrace-based engine and the Freestyle line-based engine. The Blender benchmark renders seven pre-defined scenes.

Blender not only responds well to core scaling but also frequency. The 12900K completed the renders between 17 and 50 per cent faster than the 11900K at 125W, and almost halved the time across in all scenarios after we raised the power limit to 241W. Although the 11900K saw major gains when we removed its power limit, the 12900K didn’t budge much as it was already reaching the limit of the cooler and therefore had no thermal headroom to increase its performance much further.

XTU Benchmark 2.0

Intel’s Xtreme Tuning Utility overclocking software offers a benchmark tool that gives a quick glance at the processor’s performance, much like Cinebench. The benchmark takes under a minute to complete on a high-end desktop processor.

The results speak for themselves.

Openbenchmark – Stockfish

Stockfish is the most popular chess engine today. It evaluates a given chess position by calculating all possible future moves. Its performance, measured in nodes per second, indicates how many moves the processor can calculate. A faster processor allows the engine to take more positions into consideration, thereby increasing its confidence in selecting the best move.

The 12900K led the stock 11900K by 68 per cent and the unlocked 11900K by 44 per cent at 125W. Dialling the 12900K’s power limit to 241W resulted in another 12 per cent increase over PBP.

Openbenchmark – SVT-AV1, 1080p Bosphoros, preset 4 (medium)

The SVT-AV1 test uses the less common AV1 codec to encode a YUV video file. It’s another heavily multithreaded test that scales with core count.

In this test, the 12900K’s extra cores gave it a 29 to 38 per cent advantage at 125W and unlimited power levels respectively.

Openbenchmark – x264

VideoLAN, the organization behind the popular VLC media player, developed this multithreaded H.264 encoding benchmark using its x264 encoder. Despite various newer codecs emerging throughout the years, including the H.265 codec, H.264 remains the most popular and prevalent streaming codec today.

At 125W, the 12900K is 69 per cent faster than the stock 11900K and 49 per cent faster than the unlimited 11900K.

Power and thermals

Spoiler: the flagship Alder Lake guzzled power and ran hot, very hot.

Intel pegs Alder Lake’s PBP at 125W and 241 MTP. Normally, I’d set PL1 and PL2 to these values as the stock configuration. But as previously mentioned, Intel had set PBP equal to MTP by default. I’ve therefore adopted this new setting as the “stock” setting for this review. I’ve also tested the processor’s thermals, power consumption and performance at its BPB.

At the 125W PBP, the 12900K peaked at around 80°C in Cinebench R23 with the P-cores hovering at around 4GHz and E-cores at 3.2GHz.

At 241W MTP, the 12900K maintained 4.8GHz on the P-cores and 3.7GHz on the E-cores with the Noctua NH-D15. The package temperature hovered in the 90°C range under load, and even sporadically spiked to 100°C. This means that SFF builds and overclockers need a solid AIO or a custom water loop to tame the beast.

When I handed off the power controls to the motherboard’s MCE, the Gigabyte Aorus Pro pushed both BPB and MTP to 4,096W, essentially removing the power limit. Since the 12900K already reached the junction temperature at 241W, I wasn’t expecting Gigabyte’s Multi-Core Enhancement (MCE) feature to work any magic without a better cooler.

Contrary to my expectation, leaving thermals as an afterthought did raise the multithreaded performance slightly. The 12900K sustained a 4.9GHz on the P-cores and 3.9GHz on the E-cores, drawing nearly 300W at its peak. The frequency boost translated to 27,768 points in Cinebench R23’s multicore run, a five per cent increase over what it had achieved at 241W. The single-core performance also saw a small jump to 2,034 but the difference was negligible.

As expected, temperatures shot through the roof with MCE on. The package temperature reached a scorching 101°C and the core temperature bobbed between 90 and 101°C, raising serious concerns of throttling and degradation.

Overclocking

Since the 12900K already hit the mid-90s in temperature at 241W, there was little overclocking headroom. Regardless, I wanted to see how far I could push the chip.

Having two different core architectures led to some interesting overclocking decisions on a limited thermal budget. Nudging the Vcore to 1.22V, I achieved a modest overclock of 5GHz on all P-cores. I did, however, have to suppress the E-core frequency from 3.9GHz to 3.7GHz to keep things stable. I also had to dial down the max frequency of the favoured cores by 200MHz.

Temperatures were stratospheric, as expected. Even the modest overclock sent the package temperature to 101°C, way too hot for day-to-day use. Peak power draw was around 280W. Still, I’m confident that with a high-end 240 or 360mm AIO, users can easily achieve much more impressive results if they don’t mind the power consumption.

Conclusion

For the first time, Intel has a competitive 16-core chip in its consumer segment. Although it isn’t all performance cores, the 12900K drove high-single core performance which benefits today’s games by delegating background tasks to the E-cores, thereby freeing up resources for the P-cores to do the heavy lifting more effectively.

And the performance gains are very impressive. It feels like Intel has finally taken a step forward in advancing its microarchitecture. In almost all multicore benchmarks, the Core i9-12900K showed double-digit gains compared to the last gen’s Rocket Lake. The 12900K also displayed admirable performance in single-core performance, often beating the 11900K by between 10 and 20 per cent. For Intel fans holding out to upgrade because of the disappointing performance of the past few generations, Alder Lake is the chip they’ve been waiting for.

Moreover, Alder Lake finally seems to push back AMD’s reign in the consumer processor space. The 12900K’s performance makes it equal to AMD’s 5900X and 5950X, at nearly half the price of the 5900X. Despite its high power consumption and toasty temperatures, the 12900K now offers competitive multi-threaded performance without compromising single-threaded performance. With such value, the 12900K earns an easy recommendation for gaming and heavy productivity.

Adopting Alder Lake means embracing its new platform. On one hand, users are investing in a new platform with a bevy of new technologies. On the other hand, they’ll have to wait for some of these features to become mainstream. Further, users would not only need to invest in a new motherboard but also DDR5 memory for the full performance benefits. As with early adopters of any technology, there’s a premium attached to even a DDR5 RAM kit. And while most motherboard manufacturers offer DDR4 variants of their 600 series motherboards, buyers must decide if it’s worth investing in a platform that’s in the process of being phased out.

Lastly, Alder Lake needs Windows 11 to unlock the Thread Director thread scheduling magic. Microsoft’s latest Windows revamp still has wrinkles to iron out. The bigger issue is driver and application support. Before purchasing, make sure all the apps and plug-ins you use are supported on the new OS. In addition, based on other reviews, Thread Director does not appear to greatly enhance performance, or at least not yet. I’ll likely follow up with another article that compares the 12900K’s performance on Windows 10 and Windows 11.

I’m also curious to see what Intel has planned for the mobile front. The E-core is a clear statement of efficiency, density, and performance. At the Intel Architecture Day event, the company extolled the scalability of its processing cores, as well as its fabric and packaging technology. It will be interesting to see how Intel plans on mixing and matching the P and E-cores for mobile, and what efficiency gains they can achieve.