Intel’s brand new flagship, the Core i9-11900K releases today and we’ve been given one to test and review. In this article, we’ll put it through its paces against its closest competitor in specification: The AMD Ryzen 7 5800X, and we’ve got an Intel Core i9-10850K for comparison as well because it’s the current high-performance value champion.
Intel has been lagging behind in the CPU wars for six months now. They’ve lacked a CPU that can challenge AMD’s Zen 3 line up for raw performance and were missing features notably PCIe 4.0 support. The new 11th generation Rocket Lake CPUs seek to address that Intel is making some bold performance claims.
This i9-11900K CPU boasts 5.3GHz peak boost speeds using Thermal Velocity Boost, 8 cores, and 16 threads. It uses the new Xe architecture integrated GPU. The road to Rocket lake hasn’t been smooth though: it’s suffered a convoluted development, originally scheduled to be released on a 10nm production process, then backported to 14nm when that failed. This the end of the line for this architecture, this process node, and this socket as far as Intel are concerned. This should represent the pinnacle of their current capability so we’re eager to find out what it can do.
We’ve taken great care to ensure this test is fair. To do that we’ve controlled every variable that we can. All the synthetic and gaming results you’ll see are obtained with the same RAM settings across the 3 CPUs under test. We’ve tested using an up to date BIOS (0605), released just 6 days before this release. We’ve used exactly the same motherboard for both Intel CPUs, and a B550 motherboard for the Ryzen 5800X, the MSI Mortar.
For all the gaming and synthetic tests, we kept to Intel’s specifications for multi-core enhancements, power limits, and Thermal Velocity Boost. We did this because to our mind this is comparable to how we’ve tested the 5800X using PBO. Both CPUs were allowed to perform as they do with minimal setup, according to the manufacturer’s intentions, but with the automatic optimisations in place. It’s also the default behaviour of this motherboard.
We verified this behaviour with A-B testing in a number of metrics and with both our RAM settings and motherboard settings the results represent this CPU performing at its best, outside of more involved manual tuning or overclocking. RAM was set to 3600MHz CL16-16-16-32 in all tests except the specific memory tests. There’s also the issue of ‘Gear 1 and Gear 2’ memory controller settings analogous to Ryzen’s Infinity Fabric and memory controller ratio settings – these tests are run in Gear 1 with the memory controller speed matched to memory speed. We’ve also got a separate article digging deeper into the impacts of Memory speed on performance on this CPU.
The Test System
We ran both Intel CPUs in the Asus Z590 ROG Maximus XIII Hero. With 14 phase 90Amp VRMs this high-end Z590 motherboard is an overclockers dream and we found it very flexible in terms of memory settings. We ran the tests with the 0605 BIOS from ASUS including Intel’s latest Microcode updates. We used a Fractal Design Celcius S28+ AIO Cooler and an Ion+ 860W Platinum Power supply.
For RAM, we used our 16GB Samsung B-Die 4400MHz CL16 kit, but run it at 3600MHz CL16 in order to match as closely as possible the settings in our Ryzen testing.
For the GPU we used the EVGA RTX 3080 XC3 Ultra but run our test settings in order to expose the CPU performance as much as possible, this powerful and consistent GPU helped us do that.
The Ryzen comparison system is identical with the exception of an MSI Mortar B550 Motherboard.
|CPU||Intel Core i9-11900K||Intel Core i9-10850K||AMD Ryzen 5800X|
|Architecture||Rocket Lake-S||Comet Lake-S||Zen 3|
|Boost Technologies||TVB, Turbo Boost Max 3.0, Adaptive Boost Technology||TVB, Turbo Boost Max 3.0||PBO|
|On Board Graphics||UHD 750||UHD 630||None|
|Power Draw||125W (TDP) ~250W Max||125W (TDP) ~250W Max||105W (TDP)|
Intel Core i9-11900K Performance Analysis
Cinebench R20 is a test of single and multicore performance whilst rendering a scene. It is almost entirely independent of memory speed which allows us to isolate raw CPU performance.
We conducted three runs and averaged to obtain these results. In the multi-core tests, we see that the 10850K and 5800X are neck and neck, both ahead 100 points at 5990 to the 11900K’s 5860. The 4 point difference between the 10850K and 5800X is imperceptible but the last-gen Intel CPU has 10 cores, not 8 to obtain this result.
Looking at the single-core performance again averaged over 3 runs we can see the difference: Both the 11900K and 5800X score an identical 624 points average, whilst the 10850K lags 100 points behind with a score of 516.
This result is close enough that cooling set up or silicon quality on the chip could influence it but on the raw numbers, the Ryzen 7 CPU performs best in Cinebench R20 overall by matching the 10850K’s multi-core score, and the 11900K’s single-core score.
Using Blender to render a couple of scenes, again we get a sense of the rendering performance of these CPUs.
Note that shorter bars are better indicating less time taken: Different scenes favour different aspects of a CPUs performance, and in this test, we can see that for the ‘Classroom’ render, the 11900K and 10850K are neck and neck at 435 seconds, but the 5800X finishes first about 20 seconds quicker.
In BMW27 the i9-10850K takes the lead at 135 seconds, the 11900K is 10 seconds slower, and the 5800X finishes last about 20 seconds behind. There’s no clear winner here, and I feel obliged to point out that we’re using this as a test of the CPUs, and if you’re actually looking to accelerate 3D rendering a GPU will complete the task in a fraction of the time of these CPUs.
Moving on to gaming-oriented benchmarks, 3D Mark. Focussing in on the CPU component of the Fire Strike and Time Spy benchmarks, these tests do bring memory performance into play somewhat and also heavily favour higher core counts as it’s a parallel test that uses all cores.
The i9-11900K places last in Fire Strike, 500 points behind, and splits the difference between the other two CPUs under test in Time Spy.
So rounding out our synthetic benchmarks, we see a picture of the i9-11900K having a high single-core speed, on a par with the 5800X and able to match the 10850K in some workloads despite having 2 fewer cores. But it’s not faster and struggles to make a mark in these tests.
We ran our gaming benchmarks at 1080p to isolate CPU performance as much as possible, but retained settings that are relevant in the real world. The RTX 3080 helps us see differences in underlying performance.
Call of Duty: Warzone
Call of Duty: Warzone is our first test, and we ran a 5-minute battle Royale against bots to try and give an overview of performance, not a snapshot. This game is a mix of CPU and GPU performance and you need both to really achieve high frame rates even at 1080p.
We can see that the 10850K and 11900K perform almost identically here, within a couple of FPS on Average scoring just over 200FPS, min and maximum metrics. The 5800X is the clear winner though with stellar performance and 240FPS average. It’s disappointing that we’re not realising a generational performance lift in this test.
Rainbow 6 Siege
Rainbow 6 Siege has an inbuilt benchmark which we’ve found very consistent.
Here the i9-11900K falls about 20 FPS behind the 10850K on average but is 60 FPS behind the 5800X. Obviously, all three CPUs develop very high performance but it’s a shock to see Intel’s latest flagship unable to outperform either their last generation of AMD’s current equivalent in this highly CPU dependent game.
Doom Eternal is also very well optimised and capable of high frame rates and we logged two minutes of play to give us these results:
The 11900K and 10850K perform nearly identically here again, with the 5800X clearly in the lead demonstrating that even with higher settings we’re not GPU-limited in these tests thanks to the power of the RTX 3080.
Shadow of the Tomb Raider
Moving on to more demanding titles, Shadow of the Tomb Raider’s inbuilt benchmark has exceptional consistency and gives us a breakdown of CPU performance, it’s those numbers we’re looking at here to completely isolate it from GPU performance.
This test is a close-run thing, the i9-10850K is marginally behind, the 5800X marginally in front on average. In reality, it’ll be your GPU that dictates performance in this game, but we’re seeing a trend in performance emerge now between these three CPUs.
Red Dead Redemption 2
Red Dead Redemption 2 hands another win to the Ryzen 5800X.
Again it’s surprising to see the newest CPU bringing up the rear here, 15 FPS on average behind the 5800X and slightly behind the 10850K.
Flight Simulator 2020
And finally, the game that places the biggest demand on CPU power here, Flight Simulator 2020. This benchmark comprises a three-minute flight from La Guardia over Manhattan and delivers a stern test of the CPU. GPU utilisation stays under 70% here and performance is ultimately dependent on CPU speed.
Here the i9-11900K outperforms the 10850K across the board, delivering 61 FPS on average. That’s not a bad score by any means, but the 5800X beats it once again at 63 FPS average, although performance is slightly less consistent with lower lows, 1% lows and 0.1% lows. Intel made bold claims in their launch presentation about the 11900K’s performance, stating that it was capable of beating the 5900X by 11% – it’s possible that that is the case in other tests or different circumstances, but in this benchmark, it’s not the case falling slightly behind on average.
Gaming performance conclusions
Rounding up the game testing sees the Intel Core i9-11900K in an interesting position: We’re used to seeing the latest component develop a commanding lead. In these tests Intel’s new flagship, the i9-11900K, not only failing to beat a six-month-old part from AMD, but on occasion struggling to match the last generation part from Intel themselves, and one that’s not even their top-flight product.
Memory Speed Scaling
There’s been some discussion online about memory ratios – ‘Gear 1 and Gear 2’ modes in relation to the i7-11700K and i9-11900K. It also helps explain how we arrived at our memory settings for these benchmarks. We’ll touch on this now to cover key points but if it interests you please see our companion article which digs deeper into the effects of memory latency on performance for this CPU and the i9-10850K.
Gear 1 and Gear 2 are simply the full speed or half speed memory controller ratios for the CPU to control RAM. Much like Ryzen’s ‘uclock’ setting this controller to half speed induces latency, and that latency induces a performance penalty.
Let’s look at a couple of A-B tests in our most consistent benchmarks to demonstrate this effect:
You can see that ‘Gear One’ offers a slight performance bump, a few FPS, but it’s not a marked difference.
RAM speed also has its own impact on latency. To demonstrate here’s a series of runs of Shadow of the Tomb Raiders benchmarks at different RAM frequencies, but timings retained at CL 16-16-16-32 up to 3600Mhz, and CL17 at 4000MHz for stability. We’re running Gear 2 throughout here because Gear 1 wasn’t stable at 4000MHz: Remember this CPU is only officially rated up to 3200MHz or a 1600MHz Memory Clock speed because the actual RAM clock speed is half the transfer speed.
You can see how the performance gain is significant, but peaks at around 3600MHz and tails off at 4000MHz because we have to loosen timings to maintain stability. The detriment of running 2400MHz RAM is serious, and this data challenges the notion that ram speed is unimportant to Intel CPUs or less important than Ryzen. It clearly makes a big difference to potential performance. This is why we felt it was vitally important to give this CPU the same advantage as the 5800X, and as it happens that occurs around the same RAM settings, 3600MHZ CL16 and gear 1. Overall RAM latency clearly has a big impact on this CPU’s performance.
If you’d like to see a more in-depth analysis of this including data from the 10850K, please read our linked article focussing on the topic here.
Power and thermals
Power draw and the consequent heat output has long since been the cost of high performance on Intel’s 14nm CPUs. We ran tests to explore this on the i9-11900K.
The most illuminating result was using the all-core load in Cinebench, and toggling Thermal Velocity Boost to ascertain its effects on both CPU temperature and power draw. These numbers are reported by HWinfo64, total package power and temperature, and in both cases with the 280MM AIO running at full speed.
The first run to the left shows behaviour with the thermal velocity boost enabled – you can see that stock power limits are enforced and the CPU regulates power to 250W. The ASUS motherboard allows this behaviour in its default configuration. All cores sit at about 4.7GHz and the CPU does a good job of holding temperatures at 70°C. In the second run to the right, disabling Thermal Velocity boost actually allows the CPU to disobey power limits to achieve and maintain as high clock speeds as possible and it goes pretty wild, drawing up to 330W and hitting its new target of 90C before backing off the power and clocks to prevent overheating. Before that, a few cores are hitting 5.1GHz with most at 5Ghz. As a result of over-riding the power and thermal constraints, it scores 6042 points vs around 5900 points in the first run where the lower power limit is enforced.
This second run is very much a ‘gloves off no limits’ approach, with normal behaviour overridden just to demonstrate the kind of power draw you may encounter if you’re looking to overclock this CPU. The first run is much more indicative of ‘normal’ behaviour and power draw, although in most cases after the higher power time limit, Tau expires, the package power will drop to 125W for extended full core loads.
Another result of note is that simply changing the CPU cooler settings from automatic behaviour where it scales speed with CPU temperature to full speed all the time yields a 100 point increase in Cinebench R20 – cooling the CPU more aggressively and holding lower temperatures allows it to achieve higher performance.
The power draw of this CPU can be pretty insane, and you do need both a very solid motherboard power delivery set up and a high-end cooling solution to get the best of it, particularly if you intend on overclocking it.
So, where does this information leave us?
This CPU is a disappointment. We’ve got Intel’s flagship product here, and yet we see it fail to consistently outperform their last-generation chip, and fail to beat the primary competitor from AMD.
Let’s not pretend Intel haven’t tried: They’re used to the top dog position and if they could beat AMD they would. The Zen 3 CPUs were released six months ago so there was a clear target to aim for, and in the synthetics, we can see that they’ve matched it, like for like. But in the gaming tests, it can’t compete.
Ultimately what we’re seeing here is the consequences of the limitations of that ageing 14nm process. At 10nm perhaps this CPU would have run cooler, more efficiently, and at a higher clock speed. Perhaps it would have less cache latency helping gaming performance. But that’s not the case. Intel has laid it all on the table and this is it.
Then we come onto the real issue, which is price. This is a $539 product. You have to ask yourself what justifies that price? The raw performance doesn’t, and to cap it all you need to invest at least $250 more in a motherboard plus a top tier cooling solution to support it. Not only is the Ryzen 7 5800X $100 cheaper, but it also delivered the results here running on a $150 motherboard. For the cost difference, you could have a 5900X and 12 core performance that blows the 11900K into the weeds for any application that can make use of them.
This 11th CPU generation needs to be viewed as what it is – a stopgap that brings Intel up to the specification of Zen 3 chips, with native PCIe 4.0 support but cannot compete on raw performance. It’s also the end of the line for this process, a representation of what many years of refinement and tweaking can do but also what it can’t do. It can’t beat the competition.
An area we haven’t assessed is the performance of the new integrated GPU – it has some features that may make a significant difference if you do a lot of video encoding or transcoding and again Intel make some bold claims in their productivity slides – so if you’re considering the 11th Generation for a PC focussed on those tasks it will pay to dig out more specific benchmarks.
Finally, Asus released yet another BIOS just 5 days before this release, giving us insufficient time to re-test and revalidate all our results. It claims to enable ‘Adaptive Boost Technology’ for this specific CPU, the only one in the product stack to use it. That may give a small bump in multi-core workloads in a correctly configured system, but given that it’s a Beta, and this CPU has actually existed for some time prior to launch, we don’t see it making a step-change in performance. It’s something we’ll review later.
Ultimately, if you need a PCIe 4.0 platform for content creation or high-performance computing you’ll be looking at AMD anyway, the Ryzen 9 5900X and 5950X are seriously performant parts when available.
If you want a very powerful CPU on a budget then Intel caters to that at the moment with the i9-10850K which has been as low as $320, the i7-10700K or if you do want Rocket lake then I cannot see there is a huge gap in performance between this i9-11900K and the i7-11700K beneath it – it’s still an 8-core, 16-thread parts with very good gaming performance and more than enough versatility. And of course, if it’s just gaming you’re interested in, then the Zen 3 Ryzen 5 5600X and 7 5800X which are now more readily available offer the same or better performance at just $300 and $450 respectively, and with a lower platform cost, whilst the i5-10600K is discounted, doesn’t need as expensive of a motherboard, and offers excellent gaming performance as well.
Nice try Intel, but sadly this CPU just isn’t good enough to justify its price tag. The box is really lovely though.