Intel Raptor Lake CPU Falls to Core i9-12900K, Ryzen 9 5950X in Early Leaked Benchmark

Core
(Image credit: Shutterstock)

Hardware detective Komachi_Ensaka (opens in new tab) has dug up what appears to be the first benchmark for Intel's Raptor Lake processors. Although Alder Lake just came out recently, Intel is already preparing the way for its successor.

BAPCo's Crossmark software labeled the 13th Gen Raptor Lake chip as a "Genuine Intel 0000" processor, so its exact model is unknown for the moment. Therefore, we should take the specifications with a pinch of salt since it's normal for software to misreport unreleased hardware.

According to the report, the Raptor Lake processor reportedly wields 24 cores and 32 logical cores. For reference, the Core i9-12900K comes with eight Golden Cove cores and eight Gracemont cores, for a total of 16 cores and 24 logical cores. It's possible the core count is from a reporting error on the software's part, but more likely is that Intel has upped Raptor Lake's E-core count. There have been rumors that Intel might add more Gracement cores to Raptor Lake, and the report on the mystery Raptor Lake processor indicates an 8P+16E design. 

Intel Raptor Lake Processor (Image credit: Business Applications Performance Corporation)

The person or company that tested the Raptor Lake processor has erased the submission (opens in new tab) from Crossmark's database, which isn't too surprising. In fact, Intel has a history of close ties with Business Applications Performance Corporation (BAPCo), the company that created the Crossmark benchmark. We don't use this benchmark in our own testing, so take that into consideration — especially when looking at any AMD test results.

Given the time frame, it's also important to highlight that the Raptor Lake processor is an early engineering sample. This means that clock speeds aren't finalized, and software likely isn't optimized for it, similar to what we saw with Alder Lake. It's very likely that this early sample is running at reduced fixed clock speeds.

Since BapCo's software isn't part of our benchmarking suite, we'll turn to HotHardware's results (opens in new tab) for comparison. The publication paired the Core i9-12900K with DDR5-4800 memory, which is the same data rate used in the Crossmark submission. Therefore, the results should be comparable.

Intel Raptor Lake CPU Benchmarks

Swipe to scroll horizontally
ProcessorOverallProductivityCreativityResponsiveness
Core i9-12900K2,3762,2552,4862,430
Ryzen 9 5950X1,6941,6721,7461,609
Raptor Lake1,5911,4511,8041,442

Overall, the Core i9-12900K delivered up to 49.3% higher performance than the Raptor Lake processor. Even the Ryzen 9 5950X outperformed the Raptor Lake part by 6.5%, which is a bit embarrassing for Intel. We'll have to wait for more leaked benchmarks to see how Raptor Lake stacks up, though, as the early nature of the Raptor Lake silicon obviously plays a part here.

Raptor Lake is expected to land in 2022 on the same LGA1700 socket as Alder Lake. A recent Intel patent about the Digital Linear Voltage Regulator (D-LVR) suggests that Raptor Lake could boast a 25% lower power consumption. Other rumors claim that the 13th Gen processors will allegedly arrive with an improved hybrid design for better performance and an enhanced CPU cache for gaming.

Zhiye Liu
RAM Reviewer and News Editor

Zhiye Liu is a Freelance News Writer at Tom’s Hardware US. Although he loves everything that’s hardware, he has a soft spot for CPUs, GPUs, and RAM.

  • -Fran-
    Well, this is at best an ES, so nothing much to say here.

    I'd just look at the horizontal scaling of the scores and try to draw a line instead by normalizing:
    12900K: ~5%+ overall to productivity, ~5%- overall to creativity and ~2%+ overall to responsiveness.
    5950X: ~1%+ overall to productivity, ~3%- overall to creativity and ~5%- overall to responsiveness.
    ES: ~9%+ overall to productivity, ~13%- overall to creativity and ~9%- overall to responsiveness.

    I've bolded the one I find interesting, as I think that implies the IPC gap between the CPUs across the BAPCo test suit (whatever they use). And then the responsiveness could mean how fast is the CPU at moving things around between the cores, which makes sense for Raptor Lake to have bigger latency overall, since it's adding even more slow cores.

    There are some interesting things you can still extrapolate from this info and, following the lines of Alder Lake, RL may be another CPU with tradeoffs and heavy scheduler dependency (much like AMD's chiplet, but more).

    Regards,
    Reply
  • wifiburger
    early bios & early eng sample
    most likely that thing was locked to base cock & memory controller was slow down

    if it pulled that score with locked freq / early bios most likely it will outperform zen3, 12th gen easily when it's out
    Reply
  • DavidMV
    Raptor Lake has the same Gracemont E-core design and improved Golden Cove cores (Maybe +15% IPC)... so it will be as good or better than Alder Lake in every way. This was obviously running at low clock speeds. The product lineup is also already well known for Raptor Lake... it does double the E-Core count across much of the lineup. Other big changes include a more efficient voltage regulator that significantly reduces power consumption.

    i9-13900 = 8P+16E
    i7-13700 = 8P+8E
    i5-13600k = 6P+8E
    i5-13400 = 6P+4E
    i3-13100 = 4P+0E
    Pentium = 2P+0E

    I'm not going to upgrade to Alder Lake, but good chance I'll get an i7 Raptor Lake late next year.

    Both i5s are also looking very attractive for Raptor Lake! 10 and 14 cores in the midrange.
    Reply
  • btmedic04
    my guess is that this was a baseline functionality test of a very early production engineering sample. still neat to see it performing in the ballpark of the year old 5950x this early when it is still a year out from launch
    Reply
  • Alvar "Miles" Udell
    While I place no validity in the benchmark results given it is likely a very early sample running at low frequencies, it is nice to see that thanks to AMD's pressure Intel is actually having to innovate again, increasing core counts, IPC, and PPW. If Intel keeps aggressively pricing them, us end users may actually be treated to a proper war again.

    Imagine if in a couple of years this same thing will be able to be said about Intel GPUs...
    Reply
  • DavidMV
    Alvar Miles Udell said:
    Imagine if in a couple of years this same thing will be able to be said about Intel GPUs...

    I think ultimately, maybe by the end of the decade, CPUs and high performance GPUs are going to merge into one device again. They are definitely converging and not diverging in architecture at this point. I think this hybrid core design is one step.
    Reply
  • escksu
    DavidMV said:
    I think ultimately, maybe by the end of the decade, CPUs and high performance GPUs are going to merge into one device again. They are definitely converging and not diverging in architecture at this point. I think this hybrid core design is one step.

    No, i have to say this will not happen. The main reason is power consumption. If you merge a 200w cpu with a 300w gpu, you have 500w..you will need very expensive cooling solutions. And then, it will also make the silicon way too big and too exoensive. No doubt you ca. Have interpose and multiple dies but its also extremely big and expensive.

    I would say multi-die cpu (like sapphire rapids) will likely become common. Future amd and nvidia might use multiple smaller dies to create a bigger one. This means you do not need to create different designs for different models.

    Eg. A future amd 9700xt has 1 rdna5 die. Then a higher end 9800 has 2 (9800xt 3). 9900xt has 4...
    Reply
  • DavidMV
    escksu said:
    No, i have to say this will not happen. The main reason is power consumption. If you merge a 200w cpu with a 300w gpu, you have 500w..you will need very expensive cooling solutions. And then, it will also make the silicon way too big and too exoensive. No doubt you ca. Have interpose and multiple dies but its also extremely big and expensive.

    Most tasks do not peg both at 100% power at the same time. Most tasks are very heavy one sided.

    I think if you look at some leaks of intel's roadmap even in the shorter term (Meteor Lake and beyond), they are considering putting a substantially larger number of iGPU execution units on the chip to take away the need for a dedicated GPU in many cases. Apple is going the same way. Phones went that way a long time ago. My prediction is even beyond that the architectures are converging and that at some point separate iGPU execution units won't even make sense. There will just be a sea of small CPU cores.

    Memory architectures will take some work. Right now GPU memory is bandwidth optimized and CPU memory is latency optimized. But I still think it is merging reasonably fast.
    Reply
  • TerryLaze
    DavidMV said:
    Most tasks do not peg both at 100% power at the same time. Most tasks are very heavy one sided.

    I think if you look at some leaks of intel's roadmap even in the shorter term (Meteor Lake and beyond), they are considering putting a substantially larger number of iGPU execution units on the chip to take away the need for a dedicated GPU in many cases. Apple is going the same way. Phones went that way a long time ago. My prediction is even beyond that the architectures are converging and that at some point separate iGPU execution units won't even make sense. There will just be a sea of small CPU cores.

    Memory architectures will take some work. Right now GPU memory is bandwidth optimized and CPU memory is latency optimized. But I still think it is merging reasonably fast.
    iGPUs will become stronger, no question, but they will still be a far cry from being "high performance" on the other side the iGPUs of today are already enough for a lot of people.
    Also intel already tried knights landing and cpu cores as gpu, and it was terribad for consumer level tasks, it was alright for very special things but not for gaming.
    Reply
  • Friesiansam
    Given that there is no way of verifying exactly what CPU is being compared here, only a "genuine Intel 0000", what is the point of trying to infer anything from the published figures?
    Reply