Starting with the top-of-the-line Core i9-12900K CPU, we can finally show you how Intel's new 12th generation Alder Lake processor architecture fares. The chip costs $ 650, which means that Intel has positioned its new flagship desktop CPU right between the Ryzen 9 5900X and 5950X, which currently costs $ 520 and $ 750, respectively.
Today it's all about real world performance as we've already covered the 12900K's specifications.
The Core i9-12900K is an interesting beast that packs 8 power cores along with 8 efficient cores for a total of 24 threads, which seems like an odd number and that's because these efficient or "E-cores" don't support Hyper-Threading .
The P cores clock up to 5.2 GHz, the E cores are limited to 3.9 GHz. There is 30 MB L3 cache, a total of 14 MB L2 cache, 16 PCIe 5.0 lanes from the CPU and four PCIe 4.0 lanes. Another new feature is DDR5 memory support, although Alder Lake also supports DDR4. However, you need to choose which memory technology you want to pair your 12th generation processor with as you cannot mix the two.
|Intel Core i9 12900K||Intel Core i7 12700K||Intel Core i5 12600K||Intel Core i9 11900K||Intel Core i7 11700K|
|MSRP $||$ 650||$ 450||$ 320||$ 540||$ 400|
|Release date||November 2021||March 2021|
|Cores / threads||16/24||12/20||10/16||8/16|
|Base frequency||2.4 / 3.4 GHz||2.7 / 3.6 GHz||2.8 / 3.7 GHz||3.5 GHz||3.6 GHz|
|Maximum turbo||3.9 / 5.2 GHz||3.8 / 5.0 GHz||3.6 / 4.9 GHz||5.3 GHz||5.0 GHz|
|L3 cache||30 MB||25 MB||20 MB||20 MB||16 MB|
|Storage||DDR5-4800 / DDR4-3200||DDR4-3200|
|power outlet||LGA 1700||LGA 1200|
Standard memory support includes either DDR4-3200 or DDR5-4800 and today we're going to be testing both, although we won't care about DDR5-4800 and instead, to maximize what this new memory technology really offers, we've outfitted our Alder Lake CPU with G.Skills Trident Z5 DDR5-6000 CL36 memory.
For the mainboards we use the MSI Z690 Unify for DDR5 tests and the MSI Z690 Tomahawk WiFi DDR4 for our DDR4 test system. Both motherboards work exceptionally well, and we appreciate the fact that MSI nailed their BIOS from the start. All application and game data was collected using the Radeon RX 6900 XT graphics card.
Due to Alder Lake's hybrid core design, the 12900K, along with most 12th generation processors, requires Windows 11 and its improved thread scheduler for optimal performance. So we tested with a fresh install of Windows 11 and updated all other benchmark data for Ryzen and other CPUs with the latest Microsoft operating system.
The Asus ROG Crosshair VIII Dark Hero motherboard runs on the Ryzen test system with the latest BIOS update and of course all current Windows updates and drivers. Getting all of this data ready in time for Alder Lake launch was a mammoth task, and yes, there is more to come as we work on the tests for the Core i7-12700KF, followed by the Core i5-12600K test, and a lot more.
We start with Cinebench and can tell you straight away that these results are very impressive. The 12900K manages to make the Ryzen 9 5950X appear slow and increase multi-core performance by 13%.
We're also looking at a 76% increase in performance over the 11900K (!)
We strongly recommended avoiding the 11900K like the plague and these results only confirm our original stance.
An increase of 76% is a massive generation leap and yet Intel is charging less than 20% surcharge, so the 12900K already looks solid.
It's interesting to note that DDR5 did nothing to improve the 12900K's performance in this test, but that's not particularly surprising as Cinebench isn't a particularly memory-sensitive test.
For this test, Cinebench R23 uses the P-cores and as you can see, these large cores are fast, really fast. Compared to the Core i9-11900K, we see an 18% increase in single-core performance and a 23% increase over AMD's Zen 3 architecture.
7-Zip results are a little strange, but we checked them again to be sure. Here we see a massive difference in performance for the 12900K with DDR4 and DDR5 memory in the 7-Zip compression test. With DDR4-3200 memory, the 12900K was comparable to the 10900K, which made it a lot slower than the 12- and 16-core Ryzen processors.
However, using DDR5-6000 memory boosted performance by a whopping 50%, making the 12900K ahead of the 5950X, which is pretty incredible.
We don't see the same improvement in decompression performance and here the 12900K is pretty average, miles behind AMD, despite being a little over 40% more powerful compared to the 11th generation.
In the Corona benchmark, we see that the 12900K is slightly slower than the 5950X, which takes 8 to 12% longer to complete the rendering, depending on the memory used, with DDR5-6000 improving performance by 4%. Compared to the 5900X, the 12900K reduced the render time by at least 8%, so that it fits well between the two Ryzen CPUs, where the RRP places it.
The performance in Adobe Premiere Pro 2021 is pretty strong, roughly on par with the 5950X when paired with DDR4 memory. Then we look at a small 5% performance boost with DDR5 memory, and while that's enough to put the 12900K ahead of the 5950X, it's nowhere near enough to warrant the current DDR5 memory commands.
This strong single-core performance is fully visible in Photoshop 2022, with the 12900K's DDR4 configuration increasing performance by 14% over the Ryzen 9 5950X and 5900X.
With DDR5 memory, the score is then improved by a further 7%, making the 12900K the dominant desktop processor in this application.
The results of Adobe After Effects 2022 are interesting. Here, the 12900K in conjunction with DDR4 memory offered a performance increase of 11% over the 5950X, while the DDR5-6000 increased the performance by a further 7%, making this configuration almost 20% faster than the 5950X.
Factorio is a new addition to our benchmark battery and this simulation game was not included in the rest of the games as we don't measure frames per second, but updates per second. This automated benchmark calculates the time it takes to run 1000 updates. This is a single-threaded test and it appears to be heavily dependent on cache capacity.
As you can see, the 12900K does exceptionally well when compared to the 5950X and especially its predecessor, the 11900K. Upgrading to DDR5 memory only increased performance by 3%, but the 12900K was 22% faster than the 5950X and 30% faster than the 11900K.
When it comes to code compilation performance, the 12900K isn't an issue and delivers performance similar to 5950X, at least when using overpriced DDR5 memory. When using DDR4 memory, the performance of 12900K is more comparable to that of the 3950X, as the faster DDR5 memory increases the performance by 10%. Overall, a good result, as the 12900K is positioned between the 5900X and 5950X again.
The last application we'll look at is Blender, which is where the 12900K sits between the 5900X and 5950X. That means it was over 50% faster than Intel's flagship previous generation, the 11900K.
As great as the application performance looks, it comes at the cost of huge power consumption. While the 5950X achieved a total system consumption of 221 watts with slightly faster performance, the 12900K reached a peak value of over 350 watts. That's a 60% increase in total system power consumption, which is an astronomical difference.
The power consumption with DDR4 or DDR5 memory was the same.
For those wondering why the 5950X uses less power than the 5900X, this is not a new thing but to be expected and is due to the higher silicon quality of the 16-core part.
For those wondering about clock speeds, here is a look at the 12900K under all-core load in Cinebench R23. As you can see, the P cores clock at around 4.9 GHz at 4888 MHz and the E cores behave at 3.7 GHz.
Then the 12900K clocks at 5.1 GHz for the P-cores and 3.9 GHz for the E-cores in this workload with a single core load.
As for operating temperatures, the 12900K is a power hungry CPU for this type of all-core workload and as such, I couldn't avoid intermittent throttling when the Corsair iCUE H115i Elite Capellix was installed as the CPU was regularly at or very close to 100 ° C ran. These coolers have mounting pressure issues with the 12900K, but I've done my best to fix this while using copious amounts of thermal paste, but ultimately this type of setup isn't ideal for a longer core load.
Switching to the MSI CoreLiquid S360 helped and the mounting pressure was more consistent across the IHS. This cooler avoided throttling of any kind, but the core temperature still hit 96 ° C, although temperatures mostly fluctuated between the mid to low 80s.
While we only had time for limited testing, it seems like 240mm AIO liquid coolers are out of the question as 12900K owners need larger 360mm versions. That means you'll pay over $ 100 for a decent cooler, with most Corsair and MSI models trading at or over $ 200. The Ryzen 9 5950X can meanwhile be kept cool with a reasonably decent air cooler.
Enough of the serious business, now is the time to play.
Starting with F1 2021, we come across some strange data that should probably be sorted by the 1% low results. Sorted by the average frame rate, as we can see here, the Ryzen 9 5900X and 5950X are the faster gaming CPUs and break the 400 fps limit.
The 12900K with DDR4 memory was comparable to the 10900K, at least if you look at the average frame rate. If we look at the 1% low, we see that the 12900K is actually 28% faster and 15% faster than the Ryzen 9 processors. Interestingly, DDR5 increased the average frame rate slightly, but also reduced the 1 percent low. So the strong 1% low result cannot be attributed to the use of DDR5 memory, as DDR4 was faster.
Next up we have Tom Clancy's Rainbow Six Siege, where the 12900K mixes with the 5950X and 5900X. When combined with DDR4 memory, it was a bit slower, although we're talking about a slight 2.5% difference from the 5900X. Compared to the 11900K, the 12900K was 22% faster on average and a massive 43% faster at the 1% low.
Despite reduced quality settings, we can see that Horizon Zero Dawn remains largely GPU-limited. As a result, the 5900X, 5950X and 12900K all delivered comparable performance averaging ~ 190 fps. Needless to say, DDR5 with GPU limited performance has no chance of providing additional performance.
The situation is similar when testing Borderlands 3. These high-end CPUs are more than powerful enough to get the best out of the 6900 XT with the medium quality settings. The 12900K was a bit better when it came to 1% low performance and again this is not due to the introduction of DDR5 memory as we see pretty much the same when using DDR4.
Moving on to Watch Dogs: Legion, we have some noticeable numbers. This is the first example where DDR5 offers a notable increase in performance, increasing the average frame rate from 142 fps to 164 fps, which is a significant 15% increase in performance.
The 12900K was comparable to the 5900X and 5950X with DDR4 memory, but 15% faster with DDR5 when only DDR5 was only 15% more expensive.
We're back to the GPU-limited results in the new Marvel & # 39; s Guardians of the Galaxy. For those wondering why I should include such a title in our gaming benchmark suite, it's because it accurately mirrors how most games work. That said, they're completely GPU-limited when using high-end CPUs, even though we turned down the quality settings with a 6900 XT at 1080p.
Shadow of the Tomb Raider can be very CPU-intensive and here we don't use the integrated benchmark, but the village section. The 12900K did very well, taking first place in our graphics, but strangely enough, this made use of DDR4 memory.
When combined with DDR5, the 12900K was comparable to the 5950X in terms of average frame rate, while the 1% low was 4% faster. But the point here is that we have an example where the 12900K was faster with DDR4 memory and with a significant margin of 9%.
Moving on to Hitman 3, we find the 12900K back at the top of our graphic, this time with DDR4 or DDR5 memory. We're looking at the Ryzen 9 5950X or 5900X's performance, which is very good, but AMD wasn't blown out of the water in this case.
In Age of Empires 4, we find that the 12900K offers a significant increase in performance, especially when it comes to 1% low performance. Interestingly, DDR4 had the best performance on the title, increasing the average frame rate by 6%.
The 12900K was 20% faster than the 11900K and 25% faster than AMD's 5950X and 5900X. So some really impressive results for the new 12th generation processor.
Despite being a very CPU-intensive title, Cyberpunk 2077 remains GPU-limited in our tests, even with slightly reduced visual quality settings. Not much can be learned from these results, although it was interesting to see DDR5 back up a few frames slightly versus DDR4.
Power consumption while playing
When we reviewed Blender beforehand, we saw how extreme the power requirements of the Core i9-12900K are compared to parts like the Ryzen 9 5950X. While this data is relevant and accurate for high-core workloads like Blender, it is not accurate when viewing games.
Most games don't take full advantage of 8-core processors, let alone 16-core processors, and while some like Cyberpunk 2077 can spread the load fairly evenly across a wide range of cores, they never get fully or even heavily loaded on their own today's most demanding games.
In this case, the power consumption isn't what you'd expect, and as you can see here, the 10900K, 11900K, 12900K, and Ryzen 9 5950X are all comparable. Even the 5900X, which uses less power when gaming, is not noticeably better in this regard and only lowers the overall system utilization by 7% compared to the 12900K with DDR5. So, for gamers, the whole power consumption argument is a bit pointless, at least until games are putting those CPUs under full load, and at that point you'll want something faster anyway.
10 games average
Intel seems rightly to claim the world's best gaming performance again, even if it averages just 2.5% if we use the data collected so far. If we focus on the 1% low data, the Core i9-12900K with DDR5 memory was 7% faster than the 5950X, which is a decent increase in performance.
Speaking of DDR5 memory, we're seeing a modest 2% increase in DDR4 performance when we look at the 1% lows, and we're using extreme DDR5-6000 CL36 memory, which is likely to cost more than most Z690 motherboards.
To be honest, there is no difference in gaming performance between most of these high-end CPUs as you are almost a GPU limited in games today, even with an RTX 3090 or 6900 XT at 1080p with the quality settings down.
Windows 10 performance
Before concluding this test, we wanted to take a quick look at Windows 10's performance to see if there was any difference to using Windows 11 with the Core i9-12900K and DDR5-6000 memory. Starting with the Cinebench R23 multi-core test, we see that performance on Windows 10 drops by 5%, not a huge drop, but a remarkable one.
As you can see from the single-core data, the performance degradation in the multi-core test is a result of the Windows scheduler, which did not also prioritize the P-cores. For the single-core test, we only use the P-cores, so that the performance remains almost the same.
Interestingly, the 7-Zip compression and decompression performance remains unchanged when using Windows 10. We expected a slight decrease, but it didn't.
The DDR5 configuration was slower than DDR4 when tested with Windows 11 in Shadow of the Tomb Raider, and that's interesting because we're seeing better performance here on Windows 10 and this result is comparable to Windows 11 with DDR4. There is also a major difference, the 1% low is 10% higher in Windows 10, for example.
This made no sense so we went back to Windows 11 to retest, but the data was all correct. For some reason, the 12900K does better in this CPU-demanding title on Windows 10.
It seems that when comparing Alder Lake on Windows 11 vs. Windows 10, the performance is a little over the board – this could deserve a detailed 30-40 game benchmark comparing the two in the near future. But right now we also have some Rainbow Six Siege benchmarks that indicate stronger performance on Windows 11, which Intel recommends. In this test, Windows 11 increased performance by 6% at the average frame rate and by 9% at the 1% low.
Is Intel back? What we learned
That's all of the Core i9-12900K tests we've done so far, and we can draw some pretty solid conclusions based on them. Of course, there are many more tests we'd love to do and no doubt will in the future, but we didn't want this test with things like memory scaling, IPC testing, tinkering with cores, etc. We also skipped overclocking for now, since it is extremely disappointing and this CPU is as power hungry and hot enough as it is.
Expect to cover this and more topics soon, but let's first focus on the data we have available to determine if the Core i9-12900K is worth buying, and if so, who should buy it?
DDR4 vs. DDR5
First things first, DDR4 vs DDR5. This will help simplify the discussion. DDR5 offers very little extra performance when compared to high quality DDR4-3200 memory, and let's recall that we used DDR5-6000, which is about as good as it gets right now, and expensive too.
For gaming, you'll see a few percent performance increase with the best example from Watch Dogs, which got up to 20% improvement with DDR5. However, this was an outlier in our tests, although it's probably a good indication of how they're going to compare in the future. How relevant this is in the short term is difficult to say, but probably not very.
But even if we pretend that DDR5 memory regularly delivers up to 20% more gaming performance in CPU-limited scenarios, it's still not worth the extra cost. Right now, you can buy a high quality DDR4-3200 memory kit like the one we used for $ 190, with more reasonable CL16 versions costing just $ 100. DDR5, on the other hand, appears to start at $ 280 for a 5200 CL40 kit, which is much slower than what we used to test.
Meanwhile, Corsair's Dominator Platinum DDR5-5200 CL38 memory sells for $ 330, and that's still slower than what we used to be. But if we compare that to DDR4-3200 CL16 memory, it means you're paying more than three times more for 20% more performance at best. The reality of the situation, however, averages a ~ 2% increase in performance.
And even that could be seen as an exaggeration, as we are testing with a 6900 XT at 1080p with reduced quality settings. So if you're using this extreme graphics card at 1440p and the same quality settings, there is almost no situation where DDR5 offers a performance boost, let alone 20%.
Frankly, the only people who should consider DDR5 are those who buy 6900 XTs or RTX 3090s and stick them on over-the-top $ 700 motherboards. For everyone else, DDR4 makes a lot more sense. For this reason, we recommend all potential Alder Lake buyers to opt for DDR4 for the time being and only consider DDR5 when it comes very close to DDR4 in terms of price.
Core i9 vs. Ryzen 5000
The Core i9-12900K costs $ 650, which is 13% cheaper than the Ryzen 9 5950X and 25% more than the 5900X. When considering the price of a motherboard, we find that Intel Z690 boards start at $ 200 for products like the Gigabyte Z690 UD, followed by the MSI Z690-A for $ 220. These are entry-level boards. You should spend at least $ 300 on something like the Gigabyte Z690 Aero G for $ 280 or the Asus Prime Z690-A for $ 300 to pair with your flagship Core i9 CPU.
On the AMD side, we know that the $ 160 Asus TUF Gaming X570-Plus is an excellent motherboard and for those looking for a high-end option, the $ 260 MSI X570 Tomahawk WiFi is great. There are cheaper B550 options available, but we believe those who invest in a 5900X or 5950X would be better off buying the X570.
Based on this information, it looks like AMD will only save you $ 20-40 on the motherboard, which isn't a big deal in the high-end space.
Well if you're just playing the 12900K is great, but it's not amazing as it costs 25% more than the 5900X and we've seen an average performance increase of 4% in a suite of mostly CPU-intensive games. So if you're looking for high-end gaming performance without going past the point of diminishing returns, the 5900X might be a better deal, or even the 5800X or one of the smaller Alder Lakes (which we'll be reviewing shortly). But if you are looking for the ultimate gaming performance then the Core i9-12900K is for you.
Then for CPU-intensive workstation-type applications, as our original rating of DDR5 memory says, you're better off with DDR4 in most cases. But if DDR5 is your workload and time is money, then opting for the newer, faster memory is a breeze. For everyone else, DDR4 makes more sense.
All that power
The elephant in the room is power consumption, and this is especially important with core-heavy workloads. Productivity performance was pretty similar between the Ryzen 9 5950X and the Core i9-12900K, but the AMD chip uses significantly less power for these workloads. That's a tradeoff, and depending on your particular use case, AMD could be faster anyway. We think the 5950X's 16 power cores, which are consuming power relative to what we see from the 12900K, are the better choices, but it will depend on the workload. The Ryzen 9 5950X is also a lot easier to keep cool as it sucks about 130 watts less.
Still, the Core i9-12900K is a mighty impressive CPU and it signals Intel's back and competes once again with AMD in the high-end desktop PC segment.
Deal or no deal?
So would we buy the Core i9-12900K? If we were about to put together a high-end gaming PC, we think we could. It's not the obvious choice, but if motherboard prices are competitive in your area, it is without question a viable option.
If waiting is an option, we would probably find it necessary to wait and see what AMD's V-Cache brings in early next year. AMD's updated platform later next year, which will also support DDR5, but that's probably still a year away at this point.
For now, we could easily go either way, the Ryzen 9 5900X or the more modern Core i9-12900K, we believe you can't go wrong with either option. This kind of close competition is great and if a brand wants to guarantee sales it needs to be more aggressive with pricing.
That's all we have to say about the Core i9-12900K for now. Tomorrow we're testing the Core i7-12700KF and then the Core i5-12600K, followed by an avalanche of Alder Lake benchmark content. Stay tuned.