The Intel 12th Gen Core i9-12900K Review: Hybrid Performance Brings Hybrid Complexity
by Dr. Ian Cutress & Andrei Frumusanu on November 4, 2021 9:00 AM EST![](https://images.anandtech.com/doci/17047/21ADL_Chip_Angle_2_Color_BKG_3000pixels_678x452.jpg)
Today marks the official retail availability of Intel’s 12th Generation Core processors, starting with the overclockable versions this side of the New Year, and the rest in 2022. These new processors are the first widescale launch of a hybrid processor design for mainstream Windows-based desktops using the underlying x86 architecture: Intel has created two types of core, a performance core and an efficiency core, to work together and provide the best of performance and low power in a singular package. This hybrid design and new platform however has a number of rocks in the river to navigate: adapting Windows 10, Windows 11, and all sorts of software to work properly, but also introduction of DDR5 at a time when DDR5 is still not widely available. There are so many potential pitfalls for this product, and we’re testing the flagship Core i9-12900K in a few key areas to see how it tackles them.
Let’s Talk Processors
Since August, Intel has been talking about the design of its 12th Generation Core processor family, also known as Alder Lake. We’ve already detailed over 16000 words on the topic, covering the fundamentals of each new core, how Intel has worked with Microsoft to improve Windows performance with the new design, as features like DDR5, chipsets, and overclocking. We’ll briefly cover the highlights here, but these two articles are worth the read for those that want to know.
- Intel Architecture Day 2021: Alder Lake, Golden Cove, and Gracemont Detailed
- Intel 12th Gen Core Alder Lake for Desktops: CPUs, Chipsets, Power, DDR5, OCh
At the heart of Intel’s processors is a hybrid, or heterogeneous, core design. The desktop processor silicon will have eight performance cores (P-cores) and eight efficiency cores (E-cores), the latter in two groups of four. Each of the cores is designed differently to optimize for their targets, but supports the same software. The goal is that software that is not urgent runs on efficiency cores, but time-sensitive software runs on performance cores, and that has required a new management control between the processor and Windows has been developed to enable Alder Lake to work at its best. That control is fully enabled in Windows 11, and Windows 10 can get most of the way there but doesn’t have all the bells and whistles for finer details – Linux support is in development.
The use of this hybrid design makes some traditional performance measurements difficult to compare. Intel states that individually the performance cores are +19% over 11th Generation, and the efficiency cores are around 10th Generation performance levels at much lower power. At peak performance Intel has showcased in slides that four E-cores will outperform two 6th Generation cores in both performance and power, with the E-core being optimized also for performance per physical unit of silicon. Alternatively, Intel can use all P-cores and all E-cores on a singular task, up to 241W for the Core i9 processor.
On top of all this, Intel is bringing new technology into the mix with 12th Gen Core. These processors will have PCIe 5.0 support, but also DDR5-4800 and DDR4-3200 support on the memory. This means that Alder Lake motherboards, using the new LGA1700 socket and Z690 chipsets, will be either DDR4 or DDR5 compatible. No motherboard will have slots for both (they’re not interchangeable), but as we are quite early in the DDR5 lifecycle, getting a DDR4 motherboard might be the only way for users to get hold of an Alder Lake system using their current memory. We test both DDR4 and DDR5 later on in the review to see if there is a performance difference.
A small word on power (see this article for more info) – rather than giving a simple ‘TDP’ value as in previous generations, which only specified the power at a base frequency, Intel is expanding to providing both a Base power and a Turbo power this time around. On top of that, Intel is also making these processors have ‘infinite Turbo time’, meaning that with the right cooling, users should expect these processors to run up to the Turbo power indefinitely during heavy workloads. Intel giving both numbers is a welcome change, although some users have criticized the decreasing turbo power for Core i7 and Core i5.
As we reported last week, here are the processors shipping today:
Intel 12th Gen Core, Alder Lake | |||||||||
AnandTech | Cores P+E/T |
E-Core Base |
E-Core Turbo |
P-Core Base |
P-Core Turbo |
IGP | Base W |
Turbo W |
Price $1ku |
i9-12900K | 8+8/24 | 2400 | 3900 | 3200 | 5200 | 770 | 125 | 241 | $589 |
i9-12900KF | 8+8/24 | 2400 | 3900 | 3200 | 5200 | - | 125 | 241 | $564 |
i7-12700K | 8+4/20 | 2700 | 3800 | 3600 | 5000 | 770 | 125 | 190 | $409 |
i7-12700KF | 8+4/20 | 2700 | 3800 | 3600 | 5000 | - | 125 | 190 | $384 |
i5-12600K | 6+4/16 | 2800 | 3600 | 3700 | 4900 | 770 | 125 | 150 | $289 |
i5-12600KF | 6+4/16 | 2800 | 3600 | 3700 | 4900 | - | 125 | 150 | $264 |
Processors that have a K are overclockable, and those with an F do not have integrated graphics. The graphics on each of the non-F chips are Xe-LP graphics, the same as the previous generation.
At the top of the stack is the Core i9-12900K, with eight P-cores and eight E-cores, running at a maximum 241 W. Moving down to i7 gives eight P-cores and four E-cores at 190 W, and the Core i5 gives six P-cores and four E-cores at 150 W. We understand that future processors may have six P-core and zero E-core designs.
Compare at $550+ | |||||||
AnandTech | Cores P+E/T |
P-Core Base |
P-Core Turbo |
IGP | Base W |
Turbo W |
Price |
R9 5950X | 16/32 | 3400 | 4900 | - | 105 | 142 | $799 |
i9-12900K | 8+8/24 | 3200 | 5200 | 770 | 125 | 241 | $589* |
R9 5900X | 12/24 | 3700 | 4800 | - | 105 | 142 | $549 |
* AMD Quotes RRP, Intel quotes 'tray' as 1000-unit sales. Retail is ~$650 |
The Core i9-12900K, the focus of this review today, is listed at a tray price of $589. Intel always lists tray pricing, which means ‘price if you buy 1000 units as an OEM’. The retail packaging is often another +5-10% or so, which means actual retail pricing will be nearer $650, plus tax. At that pricing it really sits between two competitive processors: the 16-core Ryzen 9 5950X ($749) and the 12-core Ryzen 9 5900X ($549).
Let’s Talk Operating Systems
Suffice to say, from the perspective of a hardware reviewer, this launch is a difficult one to cover. Normally with a new processor we would run A vs B, and that’s most of the data we need aside from some specific edge cases. For this launch, there are other factors to consider:
- P-core vs E-core
- DDR5 vs DDR4
- Windows 11 vs Windows 10
Every new degree of freedom to test is arguably a doubling of testing, so in this case 23 means 8x more testing than a normal review. Fun times. But the point to drill down to here is the last one.
Windows 11 is really new. So new in fact that performance issues on various platforms are still being fixed: recently a patch was put out to correct an issue with AMD L3 cache sizes, for example. Even when Intel presented data against AMD last week, it had to admit that they didn’t have the patch yet. Other reviewers have showcased a number of performance consistency issues with the OS when simply changing CPUs in the same system. The interplay of a new operating system that may improve performance, combined with a new heterogeneous core design, combined with new memory, and limited testing time (guess who’s CPUs were held in customs for a week), means that for the next few weeks, or months, we’re going to be seeing new performance numbers and comparisons crop up.
From Intel’s perspective, Windows 11 brings the full use of its Thread Director technology online. Normally the easiest way to run software on a CPU is to assume all the cores are the same - the advent of hyperthreading, favoured core, and other similar features meant that add-ins were applied to the operating system to help it work as intended at the hardware level. Hybrid designs add much more complexity, and so Intel built a new technology called Thread Director to handle it. At the base level, TD understands the CPU in terms of performance per core but also efficiency per core, and it can tell P-core from E-core from favoured P-core from a hyperthread. It gathers all this information, and tells the operating system what it knows – which threads need performance, what threads it thinks needs efficiency, and what are the best candidates to move up or down that stack. The operating system is still king, and can choose to ignore what TD suggests, but Windows 11 can take all that data and make decisions depending on what the user is currently focused on, the priority level of those tasks, and additional software hooks from developers regarding priority and latency.
The idea is that with Windows 11, it all works. With Windows 10, it almost all works. The main difference Intel told us is although Windows 10 can separate cores apart, and hyperthreads, it doesn’t really understand efficiency that well. So its decisions are made more in regards to performance requirements, rather than performance vs efficiency. At the end of the day, all this should mean to the user is that Windows 10 tries to minimizes the run-to-run variation, but Windows 11 does it better. Ultimate best-case performance shouldn’t change in any serious way: a single thread on a P-core, or across several P-cores for example, should perform the same.
Let’s Talk Testing
This review is going to focus on these specific comparisons:
- Core i9-12900K on DDR5 vs the Competition
- Core i9-12900K on DDR5 vs Core i9-12900K on DDR4
- Power and Performance of the P-Core vs E-Core
- Core i9-12900K Windows 11 vs Windows 10
Normally when a new version of Windows is launched, I stay as far away from it as possible. On a personal level, I enjoy consistency and stability in my workflow, but also when it comes to reviewing hardware – being able to be confident in having a consistent platform is the only true way to draw meaningful conclusions over a sustained period. Nonetheless, when a new operating system is launched, there is always the call to bulk wholesale move testing to a new platform. Windows 11 is Windows 10 with a new dress and some details moved around and improved, so it should be easier than most, however I’m still going to wait until the bulk of those initial early adopter issues, especially those that might affect performance are solved, before performing a flat refresh of our testing ecosystem. Expect that to come in Q2 next year, where we will also be updating to NVMe testing, and soliciting updates for benchmarks and new tests to explore.
For our testing, we’re leveraging the following platforms:
Alder Lake Test Systems | ||
AnandTech | DDR5 | DDR4 |
CPU | Core i9-12900K 8+8 Cores, 24 Threads 125W Base, 241W Turbo |
|
Motherboard | MSI Z690 Unify | MSI Z690 Carbon Wi-Fi |
Memory | SK Hynix 2x32 GB DDR5-4800 CL40 |
ADATA 2x32 GB DDR4-3200 CL22 |
Cooling | MSI Coreliquid 360mm AIO |
Corsair H150i Elite 360mm AIO |
Storage | Crucial MX500 2TB | |
Power Supply | Corsair AX860i | |
GPUs | Sapphire RX460 2GB (Non-Gaming Tests) NVIDIA RTX 2080 Ti (Gaming Tests), Driver 496.49 |
|
Operating Systems | Windows 10 21H1 Windows 11 Up to Date Ubuntu 21.10 (for SPEC Power) |
All other chips for comparison were ran as tests listed in our benchmark database, Bench, on Windows 10.
Highlights of this review
- The new P-core is faster than a Zen 3 core, and uses 55-65 W in ST
- The new E-core is faster than Skylake, and uses 11-15 W in ST
- Maximum all-core power recorded was 272 W, but usually below 241 W (even in AVX-512)
- Despite Intel saying otherwise, Alder Lake does have AVX-512 support (if you want it)!
- Overall Performance of i9-12900K is well above i9-11900K
- Performance against AMD overall is a mixed bag: win on ST, MT varies
- Performance per Watt of the P-cores still lags Zen3
- There are some fundamental Windows 10 issues (that can be solved)
- Don’t trust thermal software just yet, it says 100C but it’s not
- Linux idle power is lower than Windows idle power
- DDR5 gains shine through in specific MT tests, otherwise neutral to DDR4
474 Comments
View All Comments
mode_13h - Saturday, November 6, 2021 - link
> the only way I think they can remedy this is by designing a new core from scratchI'm not sure I buy this narrative. In the interview with AMD's Mike Clark, he said AMD takes a fresh view of each new generation of Zen and then only reuses what old parts still fit. As Intel is much bigger and better-resourced, I don't see why their approach would fundamentally differ.
> or scaling Gracemont to target Zen 4 or 5.
I don't understand this. The E-cores are efficiency-oriented (and also minimize area, I'd expect). If you tried to optimize them for performance, they'd just end up looking & behaving like the P-cores.
GeoffreyA - Sunday, November 7, 2021 - link
I stand by my view that designing a CPU from scratch will bring benefit, while setting them back temporarily. Of course, am no expert, but it's reasonable to guess that, no matter how much they change things, they're still being restricted by choices made in the Pentium Pro era. In the large, sweeping points of the design, it's similar, and that is exerting an effect. Start from scratch, and when you reach Golden Cove IPC, it'll be at lower power I think. Had AMD gone on with K10, I do not doubt it would never have achieved Zen's perf/watt. Sometimes it's best to demolish the edifice and raise it again, not going to the opposite extreme of a radical departure.As for the E-cores, if I'm not mistaken, they're at greater perf/watt than Skylake, reaching the same IPC more frugally. If that's the case, why not scale it up a bit more, and by the time it reaches GC/Zen 3 IPC, it may well end up doing so with less power. Remember the Pentium M.
What I'm trying to say is, you've got a destination: IPC. These three architectures are taking different routes of power and area to get there. GC has taken a road with heavy toll fees. Zen 3, much cheaper. Gracemont appears to be on an even more economical road. The toll, even on this path, will go up but it'll still be lower than GC's. Zen, in general, is proof of that, surpassing Intel's IPC at a lower point of power.
GeoffreyA - Sunday, November 7, 2021 - link
Anyhow, this is just a generic comment by a layman who's got a passion for these things, and doesn't mean to talk as if he knows better than the engineers who built it.Wrs - Sunday, November 7, 2021 - link
It's not trivial to design a core from scratch without defining an instruction set from scratch, i.e., breaking all backward compatibility. x86 has a tremendous amount of legacy. ARM has quite a bit as well, and growing each year.Can they redo Golden Cove or Gracemont for more efficiency at same perf/more perf at same efficiency? Absolutely, nothing is perfect and there's no defined tradeoff between performance and efficiency that constitutes perfect. But simply enlarging Gracemont to near Golden Cove IPC (a la Pentium M to Conroe) is not it. By doing so you gradually sacrifice the efficiency advantage in Gracemont, and might get something worse than Golden Cove if not optimized well.
The big.LITTLE concept has proven advantages in mobile and definitely has merit with tweaks/support on desktop/server. The misconception you may have is that Golden Cove isn't an inherently inefficient core like Prescott (P4) or Bulldozer. It's just sometimes driven at high turbo/high power, making it look inefficient when that's really more a process capability than a liability.
GeoffreyA - Monday, November 8, 2021 - link
Putting together a new core doesn't necessarily mean a new ISA. It could still be x86.Certainly, Golden Cove isn't of Prescott's or Bulldozer's nature and the deplorable efficiency that results from that; but I think it's pretty clear that it's below Zen 3's perf/watt. Now, Gracemont is seemingly of Zen's calibre but at an earlier point of its history. So, if they were to scale this up slowly, while scrupously maintaining its Atom philosophy, it would reach Zen 3 at similar or less power. (If that statement seems laughable, remember that Skylake > Zen 1, and Gracemont is roughly equal to Skylake.) Zen 3 is right on Golden Cove's tail. So why couldn't Gracemont's descendant reach this class using less power? Its design is sufficiently different from Core to suggest this isn't entirely fantasy.
And the fashionable big/little does have advantages; but question is, do those outweigh the added complexity? I would venture to say, no.
mode_13h - Monday, November 8, 2021 - link
> they're still being restricted by choices made in the Pentium Pro era.No way. There's no possible way they're still beholden to any decisions made that far back. For one thing, their toolchain has probably changed at least a couple times, since then. But there's also no way they're going to carry baggage that's either not pulling its weight or is otherwise a bottleneck for *that* long. Anything that's an impediment is going to get dropped, sooner or later.
> As for the E-cores, if I'm not mistaken, they're at greater perf/watt than Skylake
Gracemont is made on a different node than Skylake. If you backported it to the original 14 nm node that was Skylake's design target, they wouldn't be as fast or efficient.
> why not scale it up a bit more, and by the time it reaches GC/Zen 3 IPC,
> it may well end up doing so with less power.
Okay, so even if you make everything bigger and it can even reach Golden Cove's IPC without requiring major parts being redesigned, it's not going to clock as high. Plus, you're going to lose some efficiency, because things like OoO structures scale nonlinearly in perf/W. And once you pipeline it and do the other things needed for it to reach Golden Cove's clock speeds, it's going to lose yet more efficiency, probably converging on what Golden Cove's perf/W.
There are ways you design for power-efficiency that are fundamentally different from designing for outright performance. You don't get a high-performance core by just scaling up an efficiency-optimized core.
GeoffreyA - Monday, November 8, 2021 - link
Well, you've stumped me on most points. Nonetheless, old choices can survive pretty long. I've got two examples. Can't find any more at present. The instruction fetch bandwidth of 16 bytes, finally doubled in Golden, goes all the way back to Pentium Pro. That could've more related to the limitations of x86 decoding, though. Then, register reads were limited to two or three per clock cycle, going back to Pentium Pro, and only fixed in Sandy Bridge. Those are small ones but it goes to show.I would say, Gracemont is different enough for it to diverge from Golden Cove in terms of perf/watt. One basic difference is that it's using a distributed scheduler design (following in the footsteps of the Athlon, Zen, and I believe the Pentium 4), compared to Pentium Pro-Golden Cove's unified scheduler. Then, it's got 17 execution ports, more than Zen 3's 14 and GC's 12. It's ROB is 256 entries, equal to Zen 3. Instruction boundaries are being marked, etc., etc. It's clock speed is lower? Well, that's all right if its IPC is higher than frequency-obsessed peers. I think descendants of this core could baffle both their elder brothers and the AMD competition.
GeoffreyA - Monday, November 8, 2021 - link
Sorry for all the it's! Curse that SwiftKey!mode_13h - Tuesday, November 9, 2021 - link
> it's got 17 execution portsThat's for simplicity, not by necessity. Most CPUs map multiple different sorts of operations per port, but Gracemont is probably designed in some way that made it cheaper for them just to have dedicated ports for each. I believe its issue bandwidth is 5 ops/cycle.
> It's clock speed is lower? Well, that's all right if its IPC is higher than frequency-obsessed peers.
It would have to be waaay higher, in order to compensate. It's not clear if that's feasible or the most efficient route to deliver that level of performance.
> I think descendants of this core could baffle both their elder brothers and the AMD competition.
In server CPUs? Quite possibly. Performance per Watt and per mm^2 (which directly correlates with perf/$) could be extremely competitive. Just don't expect it to outperform anyone's P-cores.
GeoffreyA - Wednesday, November 10, 2021 - link
I'm out of answers. I suppose we'll have to wait and see how the battle goes. In any case, what is needed is some new paradigm that changes how CPUs operate. Clearly, they're reaching the end of the road. Perhaps the answer will come from new physics. But I wouldn't be surprised there's some fundamental limit to computation. That's a thought.