Skip to main content

Core i9-13900K Early Review Shows Big Gains Over Core i9-12900K

Raptor Lake wafer
Raptor Lake wafer (Image credit: Future)

Thanks to an early Intel Core i9-13900K retail review on Bilibili, we get an early look at Raptor Lake's architectural enhancements and performance improvements over Intel's 12th Gen Alder Lake architecture on finalized CPU silicon. Be warned that this review is super early, released far earlier than the October 20 official launch. So take all this info with a grain of salt.

The review shows us that Intel's design philosophy going into Raptor Lake was to improve both its frequency headroom and its multi-threaded performance - by doubling the E cores and improving the architecture's L2 and L3 cache performance. 

But starting with core frequencies first, the Core i9-13900K sees a massive clock speed improvement over its predecessor, the Core i9-12900K, featuring a 5.8 GHz peak boost on cores 1 and 2, while cores 3-8 see a peak turbo boost clock of 5.5 GHz. Likewise, the E cores saw a stellar improvement, going from 3.7 GHz on the Core i9-12900K to 4.3 GHz on the Core i9-13900K - despite core counts doubling from 8 to 16 simultaneously.

According to the reviewer, Intel has also optimized the ring bus to help improve access delays between the cores. In Intel's Alder Lake design, the ring bus would drop 3600 MHz when the E cores were primarily active over the P cores. However, with the Core i9-13900K, the ring bus now operates at more than 4600 MHz. This small but essential change vastly improves core-to-core communication latency to around 30-33 ns across all 24 cores, excluding exceptions. Alder Lake's core latency results show roughly 30-33 ns for eight cores and 35-40ns for the rest.

Raptor Lake Cache Improvements

Overall, the Core i9-13900K sees a 5-11% latency improvement in the L1, L2, and L3 caches compared to the Core i9-12900K on the P cores. In contrast, the E cores see a much more significant 16-18% improvement in benchmarks tested.

On top of this, thanks to Raptor Lake's bigger L2 and L3 cache sizes, latency also improves for a longer duration, as each test can stay in the L2 or L3 cache for longer since cache sizes have increased. Raptor Lake accomplishes this boost in latency with two methods, the first is due to Raptor Lake's frequency improvements to the cores, and second, the cache performance largely remains the same when taking frequency out of the equation.

It is good because higher cache capacity usually directly impacts cache latency. But with Raptor Lake, we are not seeing this behavior as suitable for performance. The only exception to the latency improvements is with the L3 cache, where we know a bit of extra latency at the edges of the L3 test. However, the E cores are fair, just as good, and even better where the L3 cache is concerned, and they see a latency reduction compared to Alder Lake.

Bandwidth has also improved a lot, but it depends on the workload. For example, read performance increases with the L1 cache by 12.5% in single-threaded testing on the P cores. Everywhere else, however, performance is the same between both architectures - including the E cores. But, in multi-threaded workloads, cache bandwidth is vastly improved from 11% to 44%. According to the reviewer, it is due to monstrous pure L3 cache bandwidth improvements, from higher association up to 12 way vs. 10 way.

Raptor Lake Performance

In Cinebench R15, R20, and R23, the Core i9-13900K gained a 12.5% average performance improvement over the Core i9-12900KF with the P cores only in the single-threaded test. However, in testing with the E-cores, the Core i9-13900K sees a 16% single-threaded improvement in performance over the Core i9-12900K's E cores, but the same 12% in the other Cinebench versions.

Multi-threaded results showed even more outstanding results, with a 48% average performance improvement for the Raptor Lake part in all three Cinebench versions. Additionally, in all other tests the review conducted, including 7z decompression, compression, cryptography, 3DMark, and more, the Core i9-13900K was, on average, 41% faster than the Core i9-12900KF.

The reviewer also tested pure IPC results with a locked 3.6 GHz frequency. For the P cores, they netted a 12% IPC improvement for the Core i9-13900K over the Core i9-12900KF and a 6% improvement for the E cores.

Gaming tests were also conducted in several games, including Ashes of the Singularity, CSGO, and more. Overall, the Core i9-13900K netted a 10%+ improvement over the Core i9-12900KF.

If this reviewer's data is accurate, Raptor Lake is a significant upgrade over Alder Lake in almost every way. It provides a generational leap in performance in most areas without moving to a newer Intel node. These improvements can be attributed to the L2 and L3 cache density, latency, bandwidth improvements, and the added frequency headroom. Not to mention adding 2x more E cores compared to Alder Lake.

However, we have to stress that these are very, very early results for Raptor Lake, they may very well be accurate, but we have to take them with a grain of salt for now until we see more 3rd party reviews come out around Raptor Lake's launch date of October 20.

Aaron Klotz
Freelance News Writer

Aaron Klotz is a freelance writer for Tom’s Hardware US, covering news topics related to computer hardware such as CPUs, and graphics cards.

  • jkflipflop98
    Clever girl. . .
    Reply
  • Tom Sunday
    All will come out in the wash and how Raptor Lake will finally fare against the already much touted Ryzen 7000. This is what the masses really care about and want to know! Hopefully all of the independent (hopefully much unbiased hands-on) testing should be done in the months ahead and hundreds of these new CPU puppies should also be in the hands of the first consumers. Will AMD have an edge right out of the gate as the upper 7000 series demands a new MB and DDR5? Hopefully I will be able in cutting through all the upcoming hype and storefront talk of the tech-channels to see where my money should be going. Of course real cash in my already very shallow pockets is basically zero and for especially 'non-essentials' in times like these! But at least Interesting tech-times ahead!
    Reply
  • btmedic04
    Performance hasnt been this close since the Athlon XP vs Pentium 4 'Northwood' days. Im excited to see it a second time in my lifetime between these two companies
    Reply
  • gg83
    Intel sure knows how to squeeze every drop out of their nodes. I bet they can get 3 more gens out of 10nm intel7.
    Reply
  • Rdslw
    btmedic04 said:
    Performance hasn't been this close since the Athlon XP vs Pentium 4 'Northwood' days. I'm excited to see it a second time in my lifetime between these two companies
    I like that both sides have omph, but completely different. It means that we will see prices that are very nice for us, and we will be able to specialize each side
    I'm only worried that on low power, AMD will absolutely destroy intel machines, so on AMD 35W and under prices will be high and availability will be low.
    Still it seems like AMD will give us around 60% on similar 5xxx/6xxx laptop chips, so I am very happy about it, I just wish intel laptops would keep up, so prices won't skyrocket.
    Reply
  • watzupken
    Results of next gen CPU will be out in no time. If Intel can barely beat AMD, it just shows that they have lost any competitive advantage, be it from an architecture standpoint and also from a fab standpoint. The little core spamming is basically what is saving Intel for now. But it's not a trick that they can use forever before AMD pull the same trick. While I am an Alder Lake user myself, I am skeptical that Intel can use the same 10nm while drastically increasing every aspect of Raptor Lake over Alder Lake.
    Reply
  • TerryLaze
    Rdslw said:
    I like that both sides have omph, but completely different. It means that we will see prices that are very nice for us, and we will be able to specialize each side
    I'm only worried that on low power, AMD will absolutely destroy intel machines, so on AMD 35W and under prices will be high and availability will be low.
    Still it seems like AMD will give us around 60% on similar 5xxx/6xxx laptop chips, so I am very happy about it, I just wish intel laptops would keep up, so prices won't skyrocket.
    On a 35W system CPU compute becomes much less useful, how many people need cinebench or other heavily multithreaded apps on a 35W laptop?
    Intel with quicksync gives the users a lot of hours of video playback at extremely low power or photo editing thanks to the GPU that has AI, things people actually use on a 35W laptop.
    If AMD could keep up with that, then maybe we would get better prices.
    watzupken said:
    Results of next gen CPU will be out in no time. If Intel can barely beat AMD, it just shows that they have lost any competitive advantage
    ??Because you believe that companies operate on e-peen??
    The only thing intel wants is to make sales and the least amount of money they spend on doing these sales the more money they make.
    Barely beating the competition is the best way to maximize profit.
    Reply
  • RodroX
    Yay more leaks! (but now no 6 GHz at stock....)

    Exciting times!!!
    Reply
  • alceryes
    If this is true, if Intel really did mitigate the E-core performance deficit in games, and now the E-cores actually benefit gaming and other workloads, across the board, then this next gen is shaping up to be a spectacular performer!
    Reply
  • TerryLaze
    alceryes said:
    If this is true, if Intel really did mitigate the E-core performance deficit in games, and now the E-cores actually benefit gaming and other workloads, across the board, then this next gen is shaping up to be a spectacular performer!
    They are not going to benefit anything that's not capable of using that many cores, for gaming especially there barely is any difference from a current quad and upwards other than cache maybe.
    This is just going to stop hurting performance so maybe now some people are going to shut up about e-cores hurting performance in some cases, although that is doubtful as well.

    The thread director will still be the same so there will still be occasions where games will be placed on the e-cores by mistake, it's only the lag that occurs on the p-cores when e-cores are loaded that is fixed.
    Reply