Performance

I’m not a big one on posting first-party benchmark results, but the high-level overview from Intel was this:

  • At 3.3 GHz, 12900K is +19% better in Single Thread Performance over the 11900K
  • Over the 11900K, the 12900K is +19% better at 1080p High with RTX 3090
  • Over the 11900K, the 12900K gets +84% better fps when concurrently streaming
  • Over the 11900K, the 12900K is +22-100% better in content creation (Adobe)
  • Over the 11900K, the 12900K is +50% faster in BlenderMT at 241W (vs 250W)
  • Over the 11900K, the 12900K performs the same in BlenderMT at only 65W (vs 250W)

All of Intel’s tests were using Windows 11, with DDR5-4400 vs DDR4-3200. Intel did have a small one slide of comparisons against AMD in gaming with an RTX 3090, however they stated they were done without the latest L3 patch fix, and admitted that they would have preferred to show us full results. By the time this article goes live, we may have seen those results at Intel’s event.

This is a reasonable set of data, very focused on the Core i9, but when the reviews come out we’ll be able to see where it sits compared to the other parts, as well as the competition. The only thing that concerns me right now leading up to the launch is the behavior of demoting workloads to E-cores when not in focus when on the Balanced Power Plan (mentioned on the Thread Director page). It won’t be until I get hands-on with the hardware as to whether I see it as an issue or not.

Another factor to mention is DRM. Intel has made statements to this, but there is an issue with Denuvo as it uses part of the CPU configuration to identify systems to stop piracy. Due to the hybrid nature, Denuvo might register starting on a different core (P vs E) as a new system, and eventually lock you out of the game either temporarily or permanently. Out of the top 200 games, around 20 are affected and Intel says it still has a couple more to fix. It’s working with Denuvo for a high-level fix from their side, and with developers to fix from their end as well. Intel says it’s a bit harder with older titles, especially when there’s no development going on, or the IP is far away from its original source. A solution to this would be to only launch those games on specific cores, but look out for more updates as time marches on.

Conclusions

Well, it’s almost here. It looks like Intel will take the ST crown, although MT is a bit of a different story, and might rely explicitly on the software being used or if the difference in performance is worth the price. The use of the hybrid architecture might be an early pain point, and it will be interesting to see if Thread Director remains resilient to the issues. The bump up to Windows 11 is also another potential rock in the stream, and we’re seeing some teething issues from users, although right now users who are looking to early adopt a new CPU are likely more than ready to adopt a new version of Windows at the same time.

The discourse on DDR4 vs DDR5 is one I’ve had for almost a year now. Memory vendors seem ready to start seeding kits to retailers, however the expense over DDR4 is somewhat eyewatering. The general expectation is that DDR5 won’t offer much performance uplift over a good kit of DDR4, or might even be worse. The benefit of DDR5 then at this point is more to start on that DDR5 ladder, where the only way to go is up. This will be Intel’s last DDR4 platform on desktop it seems.

On the processors themselves, the Core i5 and Core i7 parts look very competitive and in line with respective popular AMD processors. Both the Core i5 and Core i7 have extra E-cores, so we’ll see if that comes in handy for extra performance, or they’ll just end up burning power and performance per watt needs re-examining. The Core i9 challenge is probably sided on Intel for single thread, but all the questions will be over proper multi-threaded performance.

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+4/20 2800 3600 3700 4900 770 125 150 $289
i5-12600KF 6+4/20 2800 3600 3700 4900 - 125 150 $264

After not much CPU news for a while, it’s time to get in gear and find out what Intel has been cooking. Come back on November 4th for our review.

Package Improvements and Overclocking
Comments Locked

395 Comments

View All Comments

  • Bik - Saturday, October 30, 2021 - link

    Let's check your argument. If 4 small cores > 1 big core while using same die area, how about just one big core for single thread performance and give us an addition of 24 more small cores? Clearly Intel (and Apple) not doing that, rather they reduce number of small cores (m1 pro, core i5 alderlake) instead. There clearly a catch, we'll know when review is out.
  • Bik - Saturday, October 30, 2021 - link

    7*4=28 cores. my math sucks.
  • Wrs - Saturday, October 30, 2021 - link

    For 1P 28E they'd probably run into fabric/coherency limits. If you have 30 nonlocking threads then that could be a rather efficient design. Let's see if we have that in a modern computer... *stares at the 10.5k CUDA cores on an RTX card*
  • nandnandnand - Saturday, October 30, 2021 - link

    Simple. You can have more than one task running at a time that can't be parallelized and will only respond to greater single-thread performance, and one or more applications running that can use multiple threads but not scale to benefit from dozens or hundreds of cores.

    Games will typically be using up to 8 cores in the upcoming generation because the PS5 and XSX have 8 cores and will be optimized to get the most out of 8. So Intel puts 8 big cores in the flagship but stops there. If the game can also use some of the small cores, great.

    Also, we already know that next year's Raptor Lake will have up to 8 big cores and 16 small cores, and it looks like Intel will go to 8 big cores and 32 small cores after that. They can tweak to improve IPC across all the cores, but the small cores will be getting a greater proportion of the die area going foward. Apple has a different strategy, if their leaks are correct.
  • ThinkingReviewer200 - Friday, October 29, 2021 - link

    Thanks for the article :)

    For the second page of the review, the sentence, "For performance, Intel lists a single P-core as ~19% better than a core in Rocket Lake 11th Gen" may have an inaccuracy. This percentage should perhaps be ~14% for according to the graph image 128% - 112% is 16% then 16%/112% is ~14% depending on the perspective, in this case perhaps the 11th gen. However, check this logic and math for these are percents inside percents calculations.
  • mode_13h - Saturday, October 30, 2021 - link

    They're not additive. It's easier to see, if you drop the percentages and just call a Rocket Lake core 1.12x as fast as a Comet Lake core. Then, Alder Lake-P is 1.28x as fast as comet lake. So, Alder Lake-P should be 1.28/1.12 as fast as a Rocket Lake core which is 1.143 or about 14%.
  • mode_13h - Saturday, October 30, 2021 - link

    Ah, but if you look closely at the chart on page 2, it says "fixed-frequency". So, that's talking about essentially "instructions per clock". However, the stats on the last page are including clock speed differences.

    So, I guess that's how they get up to 19%? It would take only a clock speed increase of about 4.3% to makeup the difference. It's still not easy to see where they find that, but if both cores throttle back and Alder Lake's P core manages to hold a higher clock speed, then it's not hard to see how we could reach a 4.3% difference.
  • tech4fun - Friday, October 29, 2021 - link

    I just canceled my Netflix account. Fanboi battle royale!
  • nandnandnand - Friday, October 29, 2021 - link

    What took you so long?
  • Kendog52404 - Sunday, October 31, 2021 - link

    Does Alder Lake support DDR4-3000 memory? Or, does it need to be 3200? I haven't bought memory for about 5 years, and it's DDR4-3000 that I had been planning to bring over from the old board. If I have to buy new memory, regardless, that changes things, and I'll need to take a serious look at DDR5.

Log in

Don't have an account? Sign up now