Most upcoming Arrow Lake CPUs will leverage TSMC nodes for compute, Intel 20A only for Core i5 and lower: Leak

Arrow Lake
(Image credit: Tom's Hardware)

Long-time PC hardware leaker Golden Pig Upgrade has made a slew of predictions for Intel's upcoming Arrow Lake CPUs (via VideoCardz). Some of the claims are easy to believe, such as Arrow Lake lacking support for DDR4 and only having the weakest iGPU that's currently paired with current-generation Meteor Lake chips. However, Golden Pig Upgrade also says Arrow Lake won't have hyper-threading, and most models won't be made at Intel; instead, they rely on TSMC's processes.

The leaker's Bilibili post is written in Chinese, and since the standard Google-translated text actually leaves out the part about hyper-threading entirely, we'll be relying on VideoCardz's translation. Golden Pig Upgrade makes eight individual claims about Arrow Lake, ranging from unsurprising to almost contradictory to what we thought we knew.

Five of these claims are pretty easy to believe: Arrow Lake will be branded Series 2, it won't support DDR4, it doesn't have low-power E-cores, it only has four Xe cores, and the higher-end SKUs won't be PCH-free. We already knew that Intel's old Gen branding was done away with and that Meteor Lake doesn't support DDR4, so that's par for the course. Turning off the low-power E-cores also makes sense because they aren't that useful for high-power CPUs. Arrow Lake will also seemingly come with the four-core GPU tile that low-end Meteor Lake models use.

Two of the remaining three predictions raise eyebrows: The leaker says that Arrow Lake has no hyper-threading and that its NPU performs the same as Meteor Lake's. We've seen leaked Arrow Lake CPUs without hyper-threading before, but as these are engineering samples, hyper-threading may be turned off for stability or testing purposes. There have been rumors that Intel will retire hyper-threading, but typically from sources with poor track records. In contrast, Golden Pig Upgrade has an excellent track record. The claim that the NPU isn't just the same as Meteor Lake's but performs the same would seem contradictory. Intel says Arrow Lake triples AI performance, though perhaps that's only for specific CPUs or refers to GPU improvements.

The final and most interesting claim from Golden Pig Upgrade is that Intel's 20A node will only be used for midrange Arrow Lake compute tiles with six P-cores and eight E-cores. It's hard to tell if the leaker means this specifically for mobile CPUs or if this also applies to desktop models, but either way, it's a big assertion. Given how much Intel has touted 20A and its introduction of PowerVia backside power delivery and RibbonFETs, it's hard to imagine Intel won't utilize it for most of the Arrow Lake lineup. The leaker also didn't mention what TSMC node Intel would use, though perhaps we're supposed to infer it's 3nm.

On the other hand, it's rumored that Intel will rely on TSMC's 3nm process for Lunar Lake MX, though this is seemingly a variant of Lunar Lake. Additionally, slides from Intel indicate that there isn't that much capacity for 20A and 18A chips in 2024 or 2025, so if this rumor is true, then perhaps it's just because of supply issues. Still, it would be very surprising if the CPU that's supposed to return Intel to node leadership is largely powered by TSMC, even if that only applies to the lower-end mobile SKUs.

Matthew Connatser

Matthew Connatser is a freelancing writer for Tom's Hardware US. He writes articles about CPUs, GPUs, SSDs, and computers in general.

  • baboma
    Here's a translation using ChatGPT. It's much better than Google's machine translation.

    original (https://www.bilibili.com/opus/904913222308462613)
    ======
    也不懂为啥一个还剩半年多就上市的CPU,还能有这么多错得离谱的爆料…

    首先Arrow Lake不可能叫15代酷睿,大概率就是酷睿Ultra处理器(系列2)了

    没有超线程没有LP E-core,但是NPU还是在的,算力和笔记本Meteor Lake一样

    4Xe core核显,因为没有达到7个以上所以任务管理器里叫Intel Graphics不叫Arc Graphics,参考Ultra5 125U

    没有DDR4支持

    移动端涉及先进工艺的Tile全部来自TSMC,20A应该只会出现在ARL-S的6+8核心,即桌面端non-K U5及以下型号

    桌面S及HX仍然无法像AMD平台那样做无PCH方案
    ======

    translated:
    I don't understand why there are so many wildly inaccurate leaks about a CPU that's still more than half a year away from being released...

    Firstly, Arrow Lake can't possibly be called the 15th generation Core processor; it's most likely going to be the Core Ultra processor (Series 2).

    There's no hyper-threading and no LP E-core, but the NPU is still present, with compute power similar to the Meteor Lake laptop.

    It features a 4Xe core integrated graphics, and since it doesn't reach 7 cores or more, in the task manager it's referred to as Intel Graphics rather than Arc Graphics, similar to the Ultra5 125U.

    There's no DDR4 support.

    For mobile platforms, all the advanced process tiles come from TSMC, and 20A should only appear in the 6+8 core configuration of ARL-S, which means non-K U5 and lower models for desktops.

    For desktop S and HX, it's still not possible to implement a PCH-less solution like on the AMD platform.
    Reply
  • Metal Messiah.
    The final and most interesting claim from Golden Pig Upgrade is that Intel's 20A node will only be used for midrange Arrow Lake compute tiles with six P-cores and eight E-cores. It's hard to tell if the leaker means this specifically for mobile CPUs or if this also applies to desktop models, but either way, it's a big assertion.

    Why is it hard to tell and so confusing ? The leaker has clearly mentioned this applies to the ARL-S series, aka the Desktop models.

    For desktops, it will only be for the 6+8 core config SKU that will feature the Intel 20A process node for the compute tile, while all of the remaining chips are said to be from TSMC (Desktop & Mobile).

    The 6+8 configuration will be featured on the Core Ultra 5 2** series desktop parts which will utilize a non-K design.
    Reply
  • George³
    Let's clear things up. The new Intel node cannot provide particularly high frequencies, at least for now. Therefore, it is used in processors with low base and not particularly high turbo frequencies.
    Reply
  • TerryLaze
    George³ said:
    Let's clear things up. The new Intel node cannot provide particularly high frequencies, at least for now. Therefore, it is used in processors with low base and not particularly high turbo frequencies.
    But TSMC has even lower clocks...
    Reply
  • Argolith
    TerryLaze said:
    But TSMC has even lower clocks...
    How could you possibly know? Are you an engineer at Intel? Did you see the samples?
    Reply
  • TerryLaze
    Argolith said:
    How could you possibly know? Are you an engineer at Intel? Did you see the samples?
    Ryzen is on the market for years, if TSMC could have better clocks then ryzen could have better clocks.
    Reply
  • dalek1234
    baboma said:
    Here's a translation using ChatGPT. It's much better than Google's machine translation.
    ...
    Somebody asked me few month ago to translate something to another language that I know. The text contained medical terminology which I wasn't confident I would translate accurately, so I thought I'd ask ChatGPT to translate. I read the translation and I was amazed at how accurate it was. I actually think it did a better job than most bilingual humans would.
    Reply
  • edzieba
    The "no LP E-core" more likely refers to the extra core on the chipset die, rather than having no E-cores on the CPU die(s), as the same leak literally mentions a 6+8 (6 P-core 8 E-core) die.

    As per baboma's translation
    baboma said:
    For mobile platforms, all the advanced process tiles come from TSMC, and 20A should only appear in the 6+8 core configuration of ARL-S, which means non-K U5 and lower models for desktops.
    It sounds like there will be multiple configurations using both TSMC sourced chiplets and 20A chiplets, with there being at least one common chiplet between the desktop -S line and mobile lines that is fabbed on Intel 20A. For the mobile line 'advanced' dies (so presumably SKUs above 6+8) have chiplets from TSMC, with 'lower' SKUs being undefined (could be all Intel, all TSMC, or a mix). Similarly, at least one desktop -S SKU is Intel sourced, with the rest undefined.
    Reply
  • The Historical Fidelity
    TerryLaze said:
    Ryzen is on the market for years, if TSMC could have better clocks then ryzen could have better clocks.
    So 5.7 GHz for Ryzen is not fast enough? TSMC’s N3 process is more mature than Intel 2. Changing to ribbon fet is a drastic redesign of how transistors are made and could come with a significant clock speed regression until engineers have time to understand and enhance the ribbon fets.
    Reply
  • The Historical Fidelity
    dalek1234 said:
    Somebody asked me few month ago to translate something to another language that I know. The text contained medical terminology which I wasn't confident I would translate accurately, so I thought I'd ask ChatGPT to translate. I read the translation and I was amazed at how accurate it was. I actually think it did a better job than most bilingual humans would.
    That’s my problem with AI, we are finally at the precipice where humans can start to rely on machines to think for us. It’s only downhill from here if we get to a future point where we rely on AI to think and make every decision for us, then a thousand year solar flare takes AI away.
    Reply