They must have hit some kind of performance issue (it is Windows) I can't believe they are taking this long to make it available to everyone.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Intel Looking to Mirror M1's Manufacturing Process for 'Meteor Lake' CPUs
- Thread starter MacRumors
- Start date
- Sort by reaction score
Sadly Apple has nixed the Intel versions of MacOS if not this year the next and it won’t matter how good the Intel chip becomes.
Exactly like when the PowerPC chips became better after Apple dropped it, you didn’t hear about someone running MacOS Leopard on Power6 & Power7 and beyond.
Eventually Intel hackentoshes will go away too.
Exactly like when the PowerPC chips became better after Apple dropped it, you didn’t hear about someone running MacOS Leopard on Power6 & Power7 and beyond.
Eventually Intel hackentoshes will go away too.
Good news for the worlds energy consumption, but I would not go a far as to claim that Apple is challenged, since Apple and Intel have different philosophies. Intel delivers a CPU, sometimes combined with a GPU, whereas able delivered a SoC complete with RAM, SSD, CPU, GPU, AI/ML cores and super wide and fast bus to connect it all. So when it comes to raw CPU performance, Intel still might regain the lead, but from an overall systems performance point of view, I'm sure Apple will still be ahead. Especially when Mac apps starts to utilise it all.
I wil not go into the modular/upgradable vs. everything on one chip discussion. Lets wait and see what Apple has planned for the Mac Pro.
Henrik
I wil not go into the modular/upgradable vs. everything on one chip discussion. Lets wait and see what Apple has planned for the Mac Pro.
Henrik
As long as intel needs to run the cores at 5+GHz to get decent performance, efficiency goes out the window. And most advantages that more advanced process nodes provide.
Hmm, one can always dream of a M1 chip (yes I know it would never fit and would suck up too much power) iPhone with some sort of DEX capability to run on a full screen.
While competition in the CPU space has indeed been great since AMD and Apple lit a fire under the industry's ass... I don't feel like "competition is great!" is the proper take away from the article.
If anything Intel moving Meteor Lake from Intel N7 to TSMC 5nm is another worrying sign that Intel STILL can't keep up with a process node TSMC first commercialized in 2020. Having all high end semiconductor manufacturing centered around one company is the opposite of healthy competition.
If we want those sweet YoY performance gains to keep coming / if we ever want to get out of supply chain hell we should all want Intel, Samsung etc to succeed with their advanced nodes and start competing with TSMC again.
(That said, I can't say I don't enjoy watching Intel struggle...)
This claim seems somewhat dubious. If the Intel 4 Compute tile and completed SoC is all the way through the validation process of booting Window,Linux.Chrome then it is relatively pretty skeptical that Intel go back and 'reflow' the Compute tile on a completely different fab process. So "all of the chip blocks" seems to be an overreach on connotation. There might be a specific subset of a Gen 14 ( Meteor Lake) SoC line up that leans on TSMC more heavily (Intel has a dozen, or more, different SoCs to do). But the whole product line? Probably not.
Gen 14 was going to have some TSMC tile in it regardless, as the iGPU tile was always on track to be TSMC. ( given the mainstream dGPU work is all on TSMC, that only makes it consistent. )
If Intel was going to "reflow" something it seems more likely they would reflow a GPU part that was either TSMC N6 or TMSC N3 down to / back up to N5. TMSC N3 appears to have a longer than previous options gestation process. If Intel wanted to use N3 for the GPU tiles but they aren't going to get the volume throughput they wanted they could do an "smaller than wanted" iGPU tile with N5 instead of N3. And use those on desktops that had a smaller iGPU anyway. [ If it is a 'small' iGPU why throw N3 at it if going to crank up costs and/or cause a supply chain problem? ]. Also if N3 is constrained and Intel plans to have dGPUs on N3 late 2023 ... getting small iGPUs off of some of those limited N3 wafers would be a huge help.
There was talk of an "Arrow Lake" that was going to be on TSMC N3
Intel has indicted that Meteor Lake is using an Intel 4 [ chart in article is little wonky about 20A being in Meteor Lake. I suspect someone copy and pasted from the Arrow Lake colum. ]
Intel confirms 15th Gen Core "Arrow Lake", Xeon "Granite Rapids" and "Sierra Forest" 2024 series - VideoCardz.com
Intel confirms Arrow Lake and Lunar Lake During Intel’s Investor’s Meeting 2022, the company’s CEO Pat Gelsinger confirmed three product series planned for the 2024 launch. For the first time, Intel confirmed Arrow Lake will succeed Meteor Lake as a client product (Intel Core series). This is a...
At some point early in the Arrow Lake design process there was an N3 Compute tile. (leaked by AdortedTV much earlier).
Intel Arrow Lake-P with 320EU GPU confirmed by a leaked roadmap, targeting to compete with Apple 14" premium laptops - VideoCardz.com
Intel Arrow Lake-P GPU confirmed feature up to 320EUs Just a few days ago Intel officially confirmed it will launch Arrow Lake architecture by 2024. This successor to Meteor Lake is now confirmed to utilize a combination of three process nodes. It will also take advantage of Intel’s new die...
videocardz.com
On the leaked slide Arrow Lake's compute tile is 'N3' which presumable would be TSMC N3. (Although there is an Intel 3 also which likely why changed plan. )
It may not be that Intel 4 doesn't work, but that Intel EMIB/Foveros doesn't have the throughput that Intel wants/needs. Going to all TSMC for the 'tiles'/chiplets could let Intel offload some of the 2.5D/3D packaging off to TSMC for a subset of a SoC lineup.
For Apple the 'M1' , 'M1 Pro' , and 'M1 Max' are monolithic c dies. If Intel is going to try to compete with those mostly monolithic packages with complex 2.5/3D packages then there is manufacturing throughput they'll have to make up ground on. Does Intel even have enough packaging facilities to soak up load of switching a huge percentage of the product line up into complex packaging?
[ Somewhat sketipcal that TSMC could soak up all of Intel's complex packing volume either. The "overflow" ? Yes. All of it? "probably not if Intel disaggregates mainstream laptop, desktop , parts of dGPU, high end Xeon SP (and parts of W) and all of HPC gpGPUs ]
Doesn’t Microsoft use an arm version of windows on that tablet thing they got? You can’t get that on a Mac?
Well, as "bleeds in six colors" Mac user, Intel playing catch up by emulating Apple isn't the kind of competition that benefits me. Especially when you consider that I'll likely be using Apple hardware with 3nm chips by the time Intel (maybe) gets their chips to market in 2023. Sure, the incredible innovation of Apple Silicon has knocked Intel of its complacency. But true competition would have Intel and others shipping comparable products that offer their own, unique innovations right now. I find it difficult to give Intel any credit at this point, justbecause they've finally woken up and put their tremendous resources to good use. Our collective gums are still bleeding from the years of productivity loss we all suffered as a result of Intel acting like the indolent rich kid that just sat around not wanting to work too hard.
Soooo, let me get this straight. Intel is having a 3rd party manufacture this future line of chips, then they're going to turn around and sell those chips to OEMs. So TSMC will make its profit, then Intel will make its profit, then OEMs will make their profit. Unless their margins are all razor-thin, how much are these machines going to cost?
But then again it works for Qualcomm, and other chip design outfits, right?
But then again it works for Qualcomm, and other chip design outfits, right?
If on the original plans Intel's Gen 15 "Arrow Lake" was mostly a "shrink" mapping of Redwood Cove processor design from Intel 4 (old Intel 7hm) to TSMC N3 (with perhaps TSMC packaging if all th tiles were all TSMC) and Intel has already started down that road. But after early start and feedback on Intel 4 yields they swapped Arrow Lake over to Intel 4, then that could shorten the time.
Backporting from TSMC N3 to N5 would be an incrementally easier change if designed to be back portable in first place. Drop some of the core count and they could keep the tile size approximately the same. ( if dragging around the E core count bloat from Gen 13 (Raptor Lake of > 8 ) just dump a large block ).
Arriving late in the Gen 14 rollout timeline could be possible if there is a decent amount of re-use/re-purpose lying around. Even more so if they keep that TSMC Gen 15 compute tile group moving after "canceled" them as primary to the bulk of the product SoCs.
But yeah... to swap out Compute Tiles for most (or all) of the Gen 14 SoC product line up because of some hiccup discovered in 2021 and still hit 2023-ish deadlines ? Probably not.
You can. Sort of. I’m running a windows 11 Arm preview in Parallels 17 on my M1 air. Only obstacle is tha Microsoft’s license terms prohibits Windows 11 Arm to run on the M1 hardware, since they, for now, has made an exclusive deal with Qualcomm
Not necessarily Apple.
An TMSC tiles from Intel and MediaTek ( N5 Intel compute, N6 or N5 Intel Memory/IO , N3/N4 Intel GPU , N4 mediatek modem) all on TSMC INFO-Lsi package Q1-Q2 2024 to compete against Quaclomm's 2024 era offering for "always connected" Windows laptops would make sense.
That would be in no way shape or form the whole Gen 14 (Meteor Lake) SoC line up, but there are some synergies in letting TSMC package all of that up since all the tiles are coming from TSMC production lines.
Yes indirectly competing with Apple. But AMD and Qualcomm are more direct threats. AMD isn't likely to have a 'modem on package' solution and Qualcomm won't have "native x86" so a competitive advantage for Intel there on missing gaps in the Windows space.
And it would offload some 2.5/3D packaging out of Intel's facilities. Which when Intel is trying to disaggregate a relatively large number of SoC product categories would help also.
So Microsoft wants to copy Apple's manufacturing process, can't wait for the ads to come out about how they're different and better...followed shortly by abysmal benchmark and thermal charts.
To be fair, MS just put out native Teams a week or so ago. They likely do not want to take chances with a janky Windows on ARM release because they know they will have a hit if they do it right.
Intel Gen 14 (Meteor Lake) SoCs were previously planned ot have TSMC N3 iGPU tiles/chiplets on them.
Leak from November 2021 but planning probably preceded this by years.
Intel Meteor Lake GPU tile rumored to utilize TSMC 3nm process technology - VideoCardz.com
Intel allegedly orders TSMC to build Meteor Lake GPU tiles According to Commercial Times, who are citing their own industry sources, Intel 14th Gen Core CPU series codenamed “Meteor Lake” is to use TSMC 3nm process technology for the graphics tile. Intel Meteor Lake will be a consumer processor...
videocardz.com
GPU wise Intel is moving when Apple is moving... been that way for several years since Intel decided to get into the mainstream discrete GPU business. ( Intel never had enough EUV fab machine buys lined up to do CPUs and legacy business in addition to a new substantively large GPU business also. If the new GPU line didn't work for some reason then wouldn't have fab couldn't use in house. So it was always going to be outsourced. If the business gets long term traction then perhaps they fold it back to in-house fabs but it made no sense to do the initial bootstrap internally. Makes less sense given the fab problems there were having but even if were not... it was dubious. ).
The only thing supposedly "new" with the Digitimes report is that Intel would be moving the Compute (CPU cores) tile over to something else. Given the M1 Pro and Max are mostly GPU cores in terms of die space consumed ... Intel wasn't going to be "behind" for the bulk of the die space used inside the completed package.
P.S. As stated in another reply it would be not much of a radical change at all for this "new" N5 order to actually be an adjustment for a subset of the iGPU tile orders so that some are switched to N5 to save cost and N3 wafer consumption for other Intel GPU products. It could have little to do with the CPU tile at all. [ e.g. put N5 iGPU tiles on desktop SoCs that are likely paired with discrete GPUs (from AMD/Intel/Nvidia ) and leave the N3 iGPU tiles to laptops and later Intel dGPUs. ]
Possibly could be the CPU core Compute tile/chiplet, but that tile is only likely less than 1/3 of the die space used on the overall SoC. Kind of missing the forest for a 'branch on a tree' here with the "New" news here. It isn't to "copy Apple" , a N5 compute tile is more likely to hit a better price point while having higher manufacturing availability. ( Even if the Intel 4 fab process is "clean" and high yield , Intel will have limited ability to make it but they don't have enough EUV machines to use. Nobody does.... there is effectively a shortage. TSMC has to most so Intel ooutsourcing load there makes sense. Nothing to do with Cupertino kool-aid . ).
Last edited: