Intel Rocket Lake-S Review: Core i9-11900K and i5-11600K Tested
Intel Reclaims The Single-Core Performance Crown
This review is, to say the least, anticlimactic. The net is bursting with Rocket Lake performance numbers well ahead of launch, with AnandTech even releasing their Core i7-11700K review back on March 5 – skirting embargo by buying a retail CPU offered for sale too early.
So what’s left to say about Rocket Lake-S? It is the last hurrah of 14nm on desktop from the company that has famously struggled of late in developing and releasing desktop products with more competitive process technology.
These 11th Gen parts are also the first desktop CPUs not based on Skylake in years. Sure, they may look just like last year’s Comet Lake processors on the surface, and they may be compatible with last year’s motherboards, but Rocket Lake-S actually brings new architecture and a chipset with some needed modernization.
The promise of significantly higher IPC is a welcome one, particularly as AMD has made such impressive strides in that area with Zen 3. And don’t knock the improved integrated graphics, as we are living through the worst GPU shortage we’ve ever experienced.
The new platform (we tested these CPUs on a shiny new ASUS ROG MAXIMUS XIII HERO Z590 motherboard) brings official PCI Express 4.0 support (and more CPU lanes) to Intel desktops. Finally Intel users can revel in those outrageous Gen4 straight line SSD speeds!
There is no getting around it, however: back-porting 10nm laptop architecture to 14nm, necessary to get the single-threaded performance Intel was after this generation, is hardly ideal – necessitating as it did a regression in the flagship CPU’s core count compared to 10th Gen desktop.
Yes, the Core i9 processor this generation is back to 8 cores, down from 10 cores with the i9 parts a year ago.
|Intel 10th and 11th Gen “K” Series Desktop Processors|
|Model||Cores / Threads||Base / Boost||TVB Frequency (Single/Multi)||Smart Cache||Memory Type||TDP||Launch Price||Launch Date|
|Core i9-10900K||10 / 20||3.70 GHz / 5.20 GHz||5.30 GHz / 4.90 GHz||20 MB||DDR4 2933||125W||$488-$499||Q2 2020|
|Core i9-11900K||8 / 16||3.50 GHz / 5.10 GHz||5.30 GHz / 4.80 GHz||16 MB||DDR4 3200||125W||$539||3/30/2021|
|Core i7-10700K||8 / 16||3.80 GHz / 5.10 GHz||–||16 MB||DDR4 2933||125W||$374-$387||Q2 2020|
|Core i7-11700K||8 / 16||3.60 GHz / 4.90 GHz||–||16 MB||DDR4 3200||125W||$399||3/30/2021|
|Core i5-10600K||6 / 12||4.1 GHz / 4.8 GHz||–||12 MB||DDR4 2666||125W||$262-$263||Q2 2020|
|Core i5-11600K||6 / 12||3.90 GHz / 4.90 GHz||–||12 MB||DDR4 3200||125W||$262||3/30/2021|
Was it worth a step back to 8-core territory to get the single-threaded performance crown back from AMD, if that is indeed the case? AMD’s 12-core Ryzen 9 5900X carries a compelling $549 list price – though Ryzen 9 availability has been pretty dismal since launch with only the 6-core and 8-core SKUs readily available today.
I won’t get into the horrors of 2020-2021 product availability too much, because it’s a crappy and depressing topic, but that does play a role when shopping for parts these days. Intel has been doing very well here.
Outwardly these processors look very much like their precessesors, as once again we are looking at LGA1200 parts. Under the heatspreader things are different, to be sure, and among other things these new processors (finally) natively support DDR4-3200; up from DDR4-2933 with 10th Gen Core.
So many LGA pads, so little time
Note: The following benchmarks for both the Core i9-11900K and Core i5-11600K were performed using the most current available microcode at the time, using BIOS 0605 for the ROG MAXIMUS XIII HERO motherboard used for testing. As I write this up a new beta BIOS 0703 is available, which promises further performance improvements and adds Adaptive Boost Technology support for the i9-11900K. In the coming days we may have to revisit this review – if indeed this new BIOS (or subsequent releases) greatly affect the numbers presented below. Suddenly I’m having AMD AGESA-level anxiety about benchmarking Intel CPUs. GREAT.
|PC Perspective CPU Test Platforms|
|Motherboards||ASUS ROG MAXIMUS XIII HERO (BIOS 0605)
MSI MEG X570 GODLIKE (BIOS 7C34v1C, AGESA ComboAM4PIV2 188.8.131.52)
ASUS ROG MAXIMUS XII EXTREME (BIOS 0403)
Gigabyte Z390 AORUS PRO (BIOS F12d)
Gigabyte Z390 I AORUS PRO WIFI (BIOS F8c)
ASUS Strix Z370-H Gaming (BIOS 2203)
Gigabyte H270 Gaming 3 (BIOS F9d)
MSI MEG X570 ACE (BIOS 7C35v192, AGESA 184.108.40.206)
|Memory||G.Skill Flare X 16GB (2x8GB) @ DDR4-3200 14-14-14-34 1.35V
Crucial Ballistix LT 16GB (2x8GB) @ DDR4-2933 16-18-18-36, 1.35V
Crucial Ballistix LT 16GB (2x8GB) @ DDR4-2666 16-18-18-36, 1.35V
Corsair Dominator Platinum 16GB (2x8GB) @ DDR4-2400, 16-18-18-36, 1.35V
Crucial Vengeance LPX 16GB (2x8GB) @ DDR4-2400, 16-18-18-36, 1.35V
|GPU||NVIDIA GeForce RTX 2080 FE|
|Storage||Corsair Neutron Series XTi 480GB|
|Power Supply||CORSAIR RM1000x, SilverStone ST1000-PTS|
|Operating System||Windows 10 64-bit Version 1909, November 2019 Update (Build 18363.592)|
|GPU Drivers||NVIDIA GeForce Game Ready 445.87, 461.72|
This first CPU workload is going to be the most demanding for sustained multi-core performance, as the Classroom test takes several minutes to complete even with a very fast CPU. I mention this as Intel’s “K” series desktop CPUs carry 125W TDPs once again this generation, though power limits on virtually any retail board will be off by default.
To help sort through the results with the power limit metric in mind, I performed CPU workloads with the Core i9-11900K at both default (no limits) and manual (enforce all limits) settings. The Core i5-11600K was just tested in its default “out-of-box” state, but it’s a less power hungry CPU so I can’t see enthusiasts intentionally limiting power draw. Plus I’d already had to re-test once when BIOS 0605 was released, and didn’t have time to double-up on all testing.
This isn’t a power limit expose or anything, but it’s certainly worth pointing out that the Core i9-11900K is nearly a minute slower (56 seconds) on average in the Classroom test with all limits enforced. The sustained clocks from all eight cores just can’t be as high when the processor is forced down to its 125W rating after the initial high-power boost period expires. For most people, if you run “auto” settings with any enthusiast board, you’re only going to be thermally – not power – limited.
Everyone’s favorite benchmark! Well, not so much. But it’s an easy way to check single vs. multi-threaded performance among CPUs, and then make charts like this:
Look at that! Intel has taken back single-threaded performance in Cinebench, with an 8-point single-core victory over the Ryzen 9 5900X (we don’t have a 5950X to test, if that makes a difference), moving up to 12 points at different power settings. The Core i9-11900K produces an average score (all test results are the average of three separate runs) of 647 – and actually moves up to 651 with limits enforced, somehow.
This single-core showing is especially impressive considering the 10th Gen Core i9-10900K only managed 538 points in the same test, a deficit of more than 100 points. As to multi-threaded perf, well, just look at the chart.
7-Zip 19.00 64-Bit
What could be more exciting than running 7-Zip’s built-in benchmark from the command line, and then viewing the text file it generates? Nothing, that’s what. Observe these large numbers:
AMD is still the 7-Zip king, with the Ryzen 9 5900X and Ryzen 7 5800X flanking the Ryzen 9 3900X at the top of the chart. The Ryzen 7 5800X in particular continues to look very, very good in these tests. And you can actually buy that one.
x264 HD Benchmark 5.0.1 64-Bit
While 2-pass 1080p encodes are démodé, I still get a thrill whenever I type arcane commands that result in text output – especially when it’s a test that runs four consecutive times and I get to average the results for you like this.
Basically a repeat of the 7-Zip results here. AMD wins multi-threaded workloads like this one, pure and simple. Also, this test was another odd instance of higher performance with power limits enforced – at least in the lighter pass 1 – with the i9-11900K. I blame immature BIOS revisions. Or my own incompetence. Or both.
Let’s look at some good ol’ fashion integer and floating point performance now, shall we? We have, thus far, only seen single-core perf via Cinebench, and we will begin with single-core here:
While AMD has made major strides in single-core performance, Intel has done very well with this new architecture and regains the lead. Multi-threaded results are next, and AMD’s core count advantage is too much:
3DMark Time Spy
Let’s a take a break from synthetic benchmarks that don’t really represent Real-World ™ usage patterns for a moment, and look instead at a (synthetic) graphics benchmark!
Revel in the density of a chart that tries to incorporate the overall, graphics, and CPU scores, taken from an average of three runs on each platform using an NVIDIA GeForce RTX 2080 FE graphics card:
Finishing in the top five is not the goal here, and the Core i9-10900K remaining on top isn’t much of a victory for Intel here. But this chart is arranged in order of “overall” score, which includes the CPU score. In terms of graphics scoring, the current Ryzen 5000 parts are slightly ahead of the pack.
DOTA 2 – High FPS Gaming
As with our 10th Gen Core review, DOTA 2 represents our lone ‘competitive gamer’ test, with a consistent benchmark being performed via this handy tool which uses the teamfight from match 3787212636. To make sure the CPU would be the bottleneck at 1920×1080, the game was configured to use DX11 and set to the “fastest” preset settings.
AMD’s Ryzen 7 5800X and Ryzen 9 5900X are so far ahead here it suggests that something is wrong. But the results are the average of multiple runs, and the settings were identical. The 11th Gen gains over the 10th Gen parts are impressive, but it’s obviously not enough.
Metro Exodus at 1280×720
It’s frustrating to benchmark CPUs with games in general, as the results don’t always scale the way you expect and questions about resolution and potential GPU-bound tests keep things extra annoying. To this end I set out to benchmark something at the most obnoxiously low settings I could find, but was thwarted in my attempts to capture the DOS-like magic of 640×480 res (or even 800×600) in modern titles. Yes, some games can go that low, but Metro Exodus wasn’t one of them. I content myself with 1280×720, but know that I could have gone a bit lower.
Oh, and you’ll notice that this chart only features the two Intel CPUs under review, as well as the two Ryzen 5000 Series CPUs on offer. This is due to the fact that I just had the idea of testing at such a low res, and benchmarking four CPUs was enough for now. Trust me and my lack of sleep: it’s enough.
AMD’s Ryzen 7 5800X continues to impress, and manages to coax quite a few extra FPS out of our RTX 2080. And there isn’t a driver disparity – these tests were run on identical versions of the game and drivers. Maybe we do need that shiny new microcode for these 11th Gen parts (once again, my testing was performed on what is now already an obsolete BIOS revision).
Power and Frequency
Quickly looking at (max) system power draw with the tested CPUs, we find that this new 8-core i9-11900K actually draws more power than the previous 10-core i9-10900K by default, though enforcing power limits obviously changes the story.
Speaking of manually enforcing power limits, here’s what multi-core frequencies look like in a benchmark that exceeds the high-power boost duration (at least with this BIOS version):
Take the above with a grain of salt. This boost behavior is subject to change. Anyhow, if you have a cooler that can handle it (I used a 280 mm AiO for all testing) there is no reason not to run without power limits if you have the power headroom in your system. Multi-threaded workloads go a lot quicker when the CPU isn’t clocking down after a pre-defined 28 or 56 second limit that doesn’t take thermals into consideration.
I think it’s safe to say that Rocket Lake is not everything that Intel would have wished for out of their 11th Gen desktop parts, but after the Ryzen 5000 Series launch they had to do something. Sure, we have enjoyed lower pricing on Intel’s retail parts since then (and actual availability!), but Intel wants performance leadership, not just price/performance competitiveness.
Intel has a lot on their plate right now, with some very serious competition from AMD (and Apple on the notebook side – though not if you listen to Justin) that could not have come at a worse time in the company’s history. Intel used to enjoy both process and performance leadership, and they have been behind with both. Under the new CEO there is a plan in place to get back on top, but we are a couple of years away from the realization of current plans.
Right now is quite literally the worst time in history to be buying PC components. Graphics cards are effectively extinct, and various other parts are either sold out, or marked up (certain desirable processors, high wattage PSUs, etc.). Intel has managed to keep their processors available, and if that trend continues then we will see adoption of Rocket Lake-S in the DIY sphere. Skeptics like me look at the new 8-core limit and eye deals on the 10-core i9-10900K, but single-core performance is so greatly improved with these 11th Gen processors that the 2-core deficit with the i9-11900K is minimal.
Ok, we know performance is good, but AMD’s CPU numbers continue to shine across many benchmarks. What is the impetus for an Intel build for enthusiasts right now – beyond recent price drops making Intel a very tempting alternative to AMD (and widespread availability of Intel’s entire processor lineup)? The total platform must be considered, and here we need to dive into Z590 to getting a clearer overall picture.
The fact that a motherboard upgrade is not required if you were on Z490 is quite welcome, but PCI Express 4.0 speeds, though connected directly to these new Rocket Lake CPUs, will still require motherboard support. Some existing boards (such as the Z490 TAICHI we looked at) were designed for PCIe 4.0 all along, and it will be interesting to see storage numbers with Z490 / Z590 with these new processors.
The rapid release of BIOS updates containing meaningful microcode updates has made early reviews like this one a little cloudy, and further testing certainly needs to be done. Intel needed to get their single-core performance up, and sacrificed core count to do it on 14nm by back-porting notebook architecture. As I said before, this is hardly ideal, but it worked.
Certainly, it’s a strange time to observe this industry. AMD’s existing 8-core Ryzen 7 5800X at $449 provides this new Intel launch some major competition out of the gate, and right now Intel has no answer to AMD’s 12-core and 16-core desktop parts.
It probably sounds weird, but I was most interested in PCIe Gen4 storage performance the greatly improved integrated graphics from these CPUs. For this latter point, consider: you can’t buy any sort of decent GPU anywhere near list price right now, and outside of the APUs AMD doesn’t offer iGPUs with Ryzen desktop. Something on this front is coming soon.
My initial conclusion with this launch: inconclusive. More testing needs to be done, and I can only hope that the latest BIOS update is the last one affecting performance for a while. Chasing microcode updates makes attempting to present accurate info in static CPU reviews a herculean task, and I definitely came up short this time.
Another point worthy of followup: the Core i5-11600K. Performance is up significantly over the previous-gen Core i5-10600K, while pricing (and core/thread count) remains the same. This could be the sleeper pick out of the 11th Gen for enthusiasts – though the 8c/16t Core i7-11700K is pretty compelling at $399 with only slightly lower clocks compared to the $539 8c/16t Core i9-11900K.
This is what we consider the responsible disclosure of our review policies and procedures.
How Product Was Obtained
The CPUs are on loan from Intel for the purpose of this review.
What Happens To Product After Review
The CPUs remain the property of Intel but are on extended loan for future testing and product comparisons.
Intel had no control over the content of the review and was not consulted prior to publication.
PC Perspective Compensation
Neither PC Perspective nor any of its staff were paid or compensated in any way by Intel for this review.
Intel has not purchased advertising at PC Perspective during the past twelve months.
If this article contains affiliate links to online retailers, PC Perspective may receive compensation for purchases through those links.