Introduction
We’re not gonna waste anytime with all the details of this CPU. You can read about the architecture, the Z690 chipset, the hybrid approach and all of that in our early preview article. What we have today is Intel’s attempt at claiming back its performance crown: the 12th-gen Intel Core i9-12900K. Featuring Intel’s new hybrid architecture that mixes eight(8) high-performance CPU cores built on Golden Cove microarchitecture and is mixed in with eight(8) more efficiency cores based on the Gracemont architecture and built on Intel’s 10nm Enhanced SuperFin aka Intel 7. So if you’re not up to speed with all of that and want to educate yourself on what happened since March 30, 2021 (yes, 11th-gen was this year), then read the article below:
If you just want the short version of the story: Intel has lost the performance crown to AMD and are very aware of it and they want it back. They want it badly but want to do it their own way. The solution: hybrid cores. Intel is still chasing frequency and their 5Ghz dream, while realized, is still not an all-core situation and their race to increase core count is still plagued with extreme heat and power draw. Spoiler: many of these issues persist but Intel manages to offset it to a point.
But the main fact of the matter here is if the new flagship from Intel able to at least give them a fighting chance? In this review we’ll look closely at the performance of the Core i9-12900K and see if the new strategy works while they rebuild their company’s former luster.
Again, this article will not discuss the details and will dive straight to the actual review. Please refer to this article for a brief on the nitty-gritty details on the Intel 12th-gen Core CPUs, the Z690 chipset and what’s new with this generation.
Photos
The Core i9-12900K is an LGA1700 CPU and requires taller socket than the LGA1200 from previous generations. That said, the CPU has also extended and has become taller vertically. Much of the design is the same with a single monolith die underneath the IHS.
The LGA1700 socket found on 600-series chipset boards will have a new socket design and will be somewhat stronger-feeling when it comes to retention strength. The CPU socket itself has a larger recessed area covered by a lip around the socket.
Test Setup & Methodology
All tests are performed in the latest version of Windows 11. For earlier reports of AMD suffering performance issues, please refer to this update published by AMD which notes that the issue has been resolved.
All systems tested use the same version of the application and no data from previous reviews are used for this test. All games are updated to their latest version and are set to details indicated in their charts. Back2Gaming prefers running games in max details with motion blur off.
Test System: Intel (10th/11th) | |
CPU | Intel Core i9-11900K, Intel Core i9-10900K |
Motherboard | ASUS ROG MAXIMUS XIII EXTREME |
RAM | G.Skill TridentZ Royal DDR4-4000 C17 16GBx2 |
GPU | ZOTAC RTX 3080 Ti AMP EXTREME |
Storage | KLEVV CRAS 920 2TB M.2 NVMe |
Cooling | NZXT Kraken X73 RGB |
Power Supply | FSP HydroG Pro 1000w |
Test System: Intel 12th | |
CPU | Intel Core i9-12900K |
Motherboard | ASUS ROG MAXIMUS Z690 HERO |
RAM | Kingston FURY Beast DDR5-5200 CL40 |
GPU | ZOTAC RTX 3080 Ti AMP EXTREME |
Storage | KLEVV CRAS 920 2TB M.2 NVMe |
Cooling | NZXT Kraken X73 RGB |
Power Supply | FSP HydroG Pro 1000w |
Test System: AMD | |
CPU | AMD Ryzen 9 5950X, AMD Ryzen 7 5800X |
Motherboard | ASUS ROG Crosshair VIII Formula |
RAM | G.Skill TridentZ Royal DDR4-4000 C17 16GBx2 |
GPU | ZOTAC RTX 3080 Ti AMP EXTREME |
Storage | KLEVV CRAS 920 2TB M.2 NVMe |
Cooling | NZXT Kraken X73 RGB |
Power Supply | FSP HydroG Pro 1000w |
Special thanks to ASUS, GIGABYTE, Kingston, FSP and ZOTAC for the hardware used in this test. Special thanks to Cybenetics Labs, NVIDIA, PassMark ,OCCT and CapFrameX for the testing equipment and software we use in this review.
Arithmetic
Superpi
SuperPI is a benchmarking standard used for simple system benchmarking to extreme, world-record performance overclocking. SuperPI operates in the simple premise of computing the value of Pi to a very large decimal. The number is user-selectable and we choose the maximum 32M value as our test. This test is single-threaded. SuperPI is free software and is available for anyone to download.
wPrime
wPrime is a multi-threaded benchmarking software that also crunches numbers, particularly calculating a set of square roots using Newton’s method. wPrime is aimed to tax all threads and cores in a system. It is also free to download for anyone.
Note: The Core i9-12900K runs on the E-cores by default in my test setup. I changed the workload priority to realtime which shifted the load only to the P-cores. You are seeing 24-threads running on the 16-thread P-cores only in this result.
Rendering
Cinebench R20
Cinebench R20 and R23 are currently the most popular performance benchmark for both Intel and AMD. Used heavily in promotional and marketing materials, Cinebench has gained massive traction due to its marketing use and many users gauge performance from it. At its core, it is the benchmark for Maxon’s Cinema4D to test system hardware performance, both multi-core and single-core use.
Cinebench R23
Corona Render
The benchmark tool for Corona, a renderer used with Autodesk 3DS Max, Maxon’s Cinema 4D, etc. It is a purely CPU-based renderer. The benchmark is a very old version but is kept for relative performance comparison.
Blender Benchmark
Blender Benchmark is the benchmark tool for Blender used to compile data from all benchmark runs to help further develop Blender. The benchmark has a selection of scenes to render including the famous BMW test, Classroom and other Cycle scenes. We choose Pavilion Barcelona only for this test.
Encoding and Compression
7-zip
7-zip is a popular compression and archiving utility. Performance-wise, it has a benchmark that measure system performance for compression and decompression measured in MIPS or millions instructions per second. Basically how fast your PC zip and unzips a 7z archive.
AV1 Encoding
AV1 encoding has been recently getting more attention so I reckon since we’ve been doing H265 for a while, a newer standard shouldn’t hurt. We use AV1-SVT standard profile in FastFlix to encode a 250MB 4K60 150Mbits 8-bit footage from a Panasonic GH5 to 1080p30. The same source is used for all our media encoding tests.
H265 Encoding
H264 and H265 encoding tests are done in Handbrake with the default Matroska profile for 1080p30.
H264 Encoding
Office & Productivity
Microsoft Office
Potentially the most used desktop software in the world next to the Windows OS, Microsoft Office is synonymous to the office experience. This test is a series of tasks like copying, pasting, cutting, adding images, resizing, saving, etc. Common tasks most Office users take for granted but done in succession to gauge responsiveness and system performance.
Photo & Video Editing
UL Procyon for Photo & Video Editing
UL Procyon is an application-based benchmark that uses the actual software to test performance. UL Procyon runs tasks and scores the system overall performance on specific tasks. UL Procyon currently has benchmarks available for Adobe Lightroom Classic, Adobe Photoshop and Adobe Premiere Pro.
PugetBenchmark for Adobe Photoshop
PugetSystems is a US-based system builder that provides high-performance computers built for workstation solutions. They developed an in-house benchmarking tool for various apps including Adobe Photoshop, Premiere Pro, After Effects, Davinci Resolve and a few others. These benchmarks put a professional-level workload on systems and is scored based on a reference system.
The current version PugetBenchmark for Photoshop is scored against a Core i9-9900K system while the Premiere Pro benchmark is scored against a Ryzen 9 5900X. PugetBench is a plugin to the software its named after and requires actual Adobe software to function. For a detailed description of the benchmark, see here (Ps) and here (Pr)
PugetBenchmark for Adobe PremierePro
Note: Adobe Photoshop 2022, Adobe Lightroom Classic 2021, and Adobe Premiere Pro 2021 were used for all testing.
AI & Machine Learning
Image Upscaling
Machine learning, deep learning, AI, we’ve been hearing them more and more and as we their applications grow, they permeate our daily tasks more and more. As someone who owns a large library of files, new and old I personally own software who advertise themselves as AI. I fondly remember TopazLabs Vivacity back in the 2010s which I heavily used in a lot of my designs then. The company has now shifted to AI-powered software offerings for image purposes.
In this benchmark we four(4) 2MP photos of my MG Freedom Gundam taken with my Sony Ericsson W900i back in 2008 and upscaled to 24MP (4x) in TopazLabs GigaPixel AI using the Art & CG model.
Shameless plug, but I love Topaz Labs and if you do photo and graphics work, please try GigaPixel AI and also checkout Video Enhance AI. They have free trials and the standalone apps are perpetual license with support for 1 year of updates.
Image Identification
While search for a duplicate file finder software, I stumbled upon Nero who is apparently still alive. They have a software called Nero AI Photo Tagger which identifies photos based on their content through machine learning. It can classify specific animals, specific group events like parties, concerts, stage presentation, group shots, portraits, etc. as well as general products.
We take 5067 images of varying sizes starting from 80KB all the way to large photos 16MB in size. The software outputs time taken to finish the task.
CrossMark and 3DMark
CrossMark
CrossMark is a cross-platform benchmark aimed to provide comparative data for different platforms. It uses application models and gauges system performance and responsivess to generation a score.
3DMark
UL Benchmark 3DMark is a gaming benchmark aimed to test systems gaming performance. The new Time Spy module is made for DirectX 12 while Fire Strike is a DirectX 11 scene.
Gaming (720p)
Testing games in 720p is a theoretical test to induce a CPU-limited scenario by pumping frames and giving us a good image of how a CPU would perform once it reaches these numbers. A system incapable of handling 720p numbers would theoretically not be able to attain anything higher with the resolution cranked up. That said, this serves as a purely artificial testing scenario.
Gaming (1080p)
Gaming (1440p)
Gaming (4K)
Power Consumption
Power draw is capture using Powenetics power draw. You can find out more about Powenetics in this article. Big thanks to Aris from Cybenetics Labs for this allowing us to recreate and use this equipment.
Power loads are rendered thru OCCT to create single-threaded, multi-threaded light work and multi-threaded stress loads.
In a more granular look, Alder Lake actual looks very good. Removing all power leaving only a GTX 1050 Ti, a USB keyboard with even the LAN port disconnected and no RGB, we can see the Core i9-12900K almost zeroing out in CPU power draw:
Now this very curious? Why is our system power (total wallpower) at 40w while there is only ~1w on the CPU, 3W on the GPU (no PCIe cables on this GTX 1050Ti) and only 5w on SATA power for 3x120mm fans? I wired upn the 24-pin tester to see. I also changed to a non-RGB board to check.
Here is the 24-pin pulling around 20-40W of power in idle so we decided to check if the E-cores pull power from the EPS power:
Stressing the E-cores only with wPrime as shown in the image above, we definitely only have E-core loads here so let’s see how the max power for the E-cores are:
What the hell is this motherboard powering with that 30 watts. No RGB, no nothing. I actually am not sure so that will be covered in a future article.
Temperature
All our test systems are tested with the same NZXT Kraken X73 RGB and use the same application method using Thermal Grizzly Kryonaut.
Conclusion
You may be looking for overclocking testing and allow me to say that given the 6 days we’ve had with this CPU and having to work on other things, we had to give up on some aspects we had to cover. One of this was overclocking as it presented a challenge. Intel was quite vague on what to expected on the overclocking results but early leaked benchmarks suggested 5.4Ghz non-AVX but with very high temperatures. That being said, just reaching 5.3Ghz proved to be a challenge and is something I’m still looking to explore in terms of the CPU averages and voltages. I do feel that overclocked settings need to at least show the performance improvement in numerous benchmarks to prove viability. A screenshot of me showing 5.3Ghz at 1.5v would be quite deceptive and is overselling Intel’s promises.
That said, we decided to focus on stock performance. As the performance standard that everyone will most likely be looking at when building a similar flagship-tier Intel rig by next-gen, the Core i9-12900K manages to re-establish Intel’s performance lead for this time. Sadly, this is in exchange for the same power draw levels we’ve seen since the 9900K along with the same heat output. Intel does offset this with their hybrid cores, which as we’ve seen in our direct power monitoring, can put CPUs to nearly 0 watts on the EPS power. This does still warrant further exploration as system power will still be around 40-50w on modest systems. Now testing this has also proved a challenge as calling forth the E-cores is quite challenging for actual real-world usage. Intel may have their Thread Director tool to demo usage, but us real-world reviewers ain’t got access to that so sadly we really can’t pin down actual power draws on the E-cores although in my testing, full-core loading on the cores draws 40w on the EPS power.
Ok so we’ve missed out on bullet #2 above as overclocking isn’t quite being the best for me right now. So let’s talk about content creation. In our real-world use benchmarks, all the scores point to Intel taking the lead. Procyon deals with consumer video editing while PugetBenchmarks deals with professional use. The results we have point us to the direction that for professional use, the Core i9-12900K is just a smidge faster than the 5950X while for consumer use, its definitely favoring the Intel CPU. This is inversely true for photo work. Lightroom Classic workload benefit very little and while still faster, isn’t by a very large margin but on a professional environment with large images, the CPU shines in graphics and photo editing.
Perhaps the greatest asset that the Core i9-12900K has over the competing flagship is responsiveness. It thoroughly beats other CPUs we have in our Office testing and responsiveness scores from CrossMark put it ahead by a significant amount.
Which points us to gaming. We actually have a dedicated gaming article for this CPU but we will need quite a significant amount of time to finish up as we plan to look at a big bunch of games on this CPU compared with the others. That said, our choice games shows the 12900K beating out or tying the current gaming CPU leader, the AMD Ryzen 9 5950X, handily in the majority of our tests.
Versus the AMD Ryzen 9 5950X, should the Intel Core i9-12900K retails for $600, it will still be cheaper $300 versus the Ryzen 9 5950X at current market price. Where AMD has Intel beat though is platform cost. A Z690 motherboard currently starts at $200 and nets you the basic of what a Z690 board offers: PCIe Gen5 slot, Gen5 M.2, 2.5GbE LAN, at least 3x M.2 NVMe Gen4 slots, a phat VRM solution, a neat looking board but then it just goes up from there. Here in the Philippines, only ASUS has announced pricing for their Z690 lineup and at more than $2200 for the MAXIMUS Z690 GLACIAL, that’s a huge chunk of change and a lot of people may have something against this is but again keep in mind that even the cheapest Z690 boards will have Gen5, Gen4, USB3.2 Gen2x2 and a lot more but it will still be pricier than a B550 for a Ryzen motherboard.
So with all of that said, should one move to the Intel Core i9-12900K? Right now, exactly November 4, 2021 at 8:30PM, the prices of GPUs are still skyhigh with RTX 3080s in the $2000 mark. And I highly urge you to invest in a GPU for your gaming needs. Its a drop-in upgrade that will benefit you even if you have a bottom-rung CPU. If you do already have a shiny, new graphics card and want to build a gaming rig around it with the latest and greatest, then by all means the Core i9-12900K is easily the best CPU for gaming available right now. Despite the power readings in our benchmarks, the CPU barely uses 80w in our esports titles and Witcher 3’s CPU-intensive area only averages around 100w in our testing. And again, we are fully equipped to this at the precise of a level with all of our tools. The cost-savings you do get, get put in the motherboard expense and there are a lot of nice-looking motherboards for Z690 with some nice features which we’ll talk about in our motherboard reviews.
So in conclusion, Intel can breath a sigh of relief for a few months as AMD readies their Zen3+ or Zen4 CPU. Whichever they release, Intel should be poised to optimized their new hybrid architecture and with news of an HEDT variety already swirling, its definitely going to be exciting watching AMD and Intel tussle it out once again on a more equal playing field, this time around though, AMD isn’t the one tugging the power lines as the power-hungry beast of the Bulldozer days, as Intel now carries that weight but what they do have is the advantage of further maturing their process and architecture. Still, focusing on what we have now, the Intel Core i9-12900K is an interesting CPU and as an enthusiast it does get me excited along with what it brings with on the platform.
As we say in Filipino, “Masarap ulam ni Intel ngayong gabi” as they take back their performance crown.