So they are still so far behind they have to use desktop power to catch up?
Does not seem like a good chip if you need to use 57w in a "15w" chip I am guessing if they are forced to use lower power state, like most oems will have to do, their "gains" will not show any more.
All depends on laptops power limitations, Zen have a very crappy cooling system. Surprised you do not consider that in mobile it is all about cooling. Apple know well this.
I'm so sorry for your loss... but PC Laptops simply cannot compete with Apple .. So many reviewers have found that all PC laptops throttle severely when you unplug them - they have to because PC's a YEARS behind apple in power efficiency. In addition to performance while on battery, longer battery life, better screens, better trackpads, better speakers, and rugged aluminum chassis.
maybe where you are.. but here, apples products are usually at least twice the price as the equivalent competition, performance wise, i wasnt talking about build and display quality.
I've found that any PC that is in the same build quality as a Mac will generally cost MORE than a Mac. And of course, it still doesn't have Apple Silicon.
AND AMD laptops, every new generation top shelf, are not x2 to x3 the price of an Intel top shelf laptop as well? Instead of an Apple an AMD fashion statement? mb
not here... for the most part, intel and and amd based notebooks are comparable in price depending on specs.. but apples are always quite a but more expensive compared to what are are getting.. it isnt called the apple tax for no reason
Screens, trackpads, speakers, cases are just as good on PC as Mac, in a similar price bracket.
Power efficiency - yeah, mac has that won currently.
But for the case, i'd take a carbon/composite case over aluminium anyday. Far less likely to scratch stuff when you throw it in a bag, and a ton lighter.
I use a macbook m1 and an X1 nano G2 for work - the mac is 35% heavier than the thinkpad - that's huge.
That is a lie. You cant test every windows laptop out there so if you claim all windows laptops perform the same then you must be a liar or ignorant maybe both.
What in earth? I just checked some Mac Book Pro 2023 prices & specs, and realised how ridiculous the pricing scheme is. Basically, since DIMM or M.2 can't be upgraded by user, you're hostage to Apple's outrageous prices. Limited memory and storage sizes for those prices is major letdown. User replaceable RAM and M.2 drives are essential to me.
The two interesting things to look at are: - cost compared to "equivalent" earlier chips - power when doing something that crosses chiplet boundaries (eg game uses CPU+GPU)
For all the advantages of chiplets in some dimensions, it seems unavoidable that they will cost more, and that they will use more power in chip-to-chip links. How *much* more of either? Well, that's the question...
Yep. Totally. Ryzen still crushed Core in Perf/efficiency. a 30% performance advantage is not something to boast about if the power consumption is ~2X (I am speculating, but the peak for the 7840U should be around 30W).
TDP have no meaning anymore. AMD and Intel can set a low TDP but it'll turbo well pass it. And laptop makers will almost always boost even beyond the turbo. This is for both AMD and Intel.
The only laptop maker that actually makes good, fast, quiet, cool laptops is Apple and soon to be Qualcomm.
The TDP of laptops can vary significantly, with manufacturers making adjustments to accommodate the format, cooling, and specific model focus. Intel is the one that stretches Boost PL2 the most in relation to PL1.
Btw, Apple also doesn't stick to the TDP specified on paper, the M3 max (56w) easily exceeds 140w at full load. Their main advantage besides having ASICs to accelerate specific software is full control of both sides, software and hardware. But there's still the downside of being stuck with Apple's OS, which doesn't work well for everyone.
The real issue is where do you get that claim of M3 Max (56W) TDP... Apple does not, as far as I know, use TDP as any sort of metric, either internally or in advertising. They will tell you what sized power adaptor ships with the laptop, they will tell you how loud it gets, they will tell you how long the battery lasts for various tasks – but they will not tell you a TDP.
If you look at various review sites, they are just MAKING UP numbers. One site claims M3 Max TDP is 30W, another claims it's 78W, you claim it's 56W – sure sign that people are JUST MAKING THINGS UP...
So you're welcome to claim that M3 Max exceeds 140W at peak (though I doubt that doing so was "easily" achieved); you're not however entitled to be outraged by this as some sort of lie on Apple's part where they exceed some supposed designed or advertised spec.
Agreed so much. TDP has been a useless number for quite some time now pretty much across the board while, at the same time, processors have been using loads more power and generating a lot more heat. And people wonder why I just pair a bluetooth keyboard with my phone and generally ignore the desktop/laptop market. I'll pass on the noisy cooling fans thanks.
They just about caught up with AMD HS processors like 7840HS. So, they show the comparison to 7840U. I would have been fine with catching up, BUT why is it not better in battery life in video playback tests!! The whole point of TILES was to boost the battery life in these light tasks. Yet every single benchmark on YouTube shows video playback battery life only matching AMD now. What was the point of these TILES then? they took a hit of single thread performance for it.
Tiles were NEVER going to be lower power! The fact that so many people assumed this shows just how ignorant most supposedly tech people are about anything they don't hear in an Intel ad...
The supposed advantages of tiles/chiplets are - ability to mix and match to create different products. Maybe valuable to AMD and Intel, negative value to customer, especially once you get to Intel with crazy 1000+ SKU plans...
- ability to improve yield because chiplets are smaller. This is of dubious value given real yields, and yield loss in packaging. But people will not let this one die because it's part of an elaborate theology that insists yields for consumer sized chips are high, which in turn justifies the wide range of SKUs from CPU vendors.
- ability to place chiplet on "appropriate" technology node. You realize "appropriate" technology node means OLDER technology node, right?
- scalability. This is, IMHO, pretty much the ONLY place where chiplets actually make sense. They allow you to build larger products than would otherwise be reasonable. You'll also notice that scalability is the *main* way AMD uses chiplets, and the *only* way Apple uses them. I expect it will also be the only way nVidia uses them.
So look at the pattern. As usual, Intel is being driven by MARKETING, not technical considerations. Marketing decided that chiplets would solve a bunch of their problems (something new and shiny! a way to create even more SKU confusion that is already the case!) And don't get me started on Ponte Vecchio. I don't want to have to keep kicking Intel, but, honestly, I suspect tiles will turn out to be exceedingly disappointing for users – and for entirely predictable reasons.
It didn't have to be this way! Intel could be using tiles the way AMD does, in very small doses to create mid and highrange products; or like Apple does, to create only high-range products that can't be created any other way. This would work for Intel and for their customers. But it would be less exciting for Intel Marketing. So...
The reviews prove you wrong. In light workloads, Meteor Lake exhibits 2.5x longer battery life compared to Raptor Lake. By moving the CPU cores and the GPU on separate tiles they are able to fully power them down in workloads where the 2 ultra low power cores on the SOC tile are sufficient.
You hardly need chiplets to - power down high performance cores AND - include very low power cores
Rather than rushing to defend Intel READ WHAT I SAID! I said that I expect the problem to be in situations that involve substantial chiplet communication, like games. I did NOT say that idle life would be negatively affected by chiplets because that would be a stupid thing to say, obviously wrong.
The lower power for the idle case is NOT the result of the chiplet design. Is that so hard to understand?
People like me try to educate, but we assume a public that is willing to do its part, in this case understanding the difference (and reasons for) - lower IDLE power - lower HIGH-DEMAND power If every posting attempt that tries to educate people is turned into an adversarial contest of "who can misinterpret this post in the silliest way" the result will be that even fewer people who actually understand anything will post in public. Is that a result people want?
Tiles with uniform process wouldn't reduce power, but Intel selected LP nodes for its SOC and IO tile (to my knowledge), enabling lower power consumption.
Also -- power gating within a piece of silicon is excellent but can't compete with turning off an entire die.
So yes, chiplets aren't required for every strategy used to power down cores and include low power cores, but they do enable strategies that require competing fabrication technology selections.
"Marketing decided that chiplets would solve a bunch of their problems."
divorced from reality the fanatics remain. So, let's get this straight, after everyone and their mothers gave a host of reasons to go chiplet, its mostly just yields in the end (you claim), but Intel doing the same thing (going chiplet for many claimed reasons but yield is the only 'correct' one) is a sign they are off the farm and doing things for entirely marketing? How is that not also the case for the other companies? Anyway, you also did that "let's complain about 1009's of sku's even though any one customer wont care about most of them" crusty old fanboy-sourced talking point, so why I expected the rest of your post to be better, is beyond me
- Appropriate process nodes: You are forgetting e. g. (HBM) memory, which is produced on entirely different process nodes. As the size of memory cells does not shrink the same way as logic does, SRAM also need not be produced with the latest process node, all it'd do is make your product more expensive. Availability and price are also consumer concerns, too. ;-)
- Ability to mix-and-match Being able to get more products and for a company to react flexibly to market demands is a boon to customers. AMD being able to replace Zen 4 by Zen 4c tiles or AI accelerators to create new products is great for everyone as taping out monolithic designs could be prohibitively expensive.
- Power It depends on the implementation and whether e. g. you integrate memory and other things. Integration of more and more components that used to sit in different places on the motherboard onto a SiP does lower system power over time. However, it seems that Intel's implementation is not as successful and power-efficient as that of other companies.
- Using tiles like AMD and Apple This is a fundamental misunderstanding of the underlying technologies. E. g. Apple has been using advanced packaging techniques in many of its products for many years, including iPhones, the Apple Watch and now all Macs. It seems that you think only the Mx Ultra series of processors qualify. But that is not the case. All recent Apple SoCs integrate memory, for example.
Ditto for AMD: their Zen 4-based Dragon range and Phoenix laptop CPUs use chiplets throughout, for example.
- Chiplets are cheaper and once High-NA EUV comes online, absolutely necessary The biggest oversight is in your post: chiplets make products a lot cheaper (https://www.nextplatform.com/2021/06/09/amd-on-why... Bigger designs are closer to the reticle limit, the size of a single stepper exposure window. That window will be cut in half with High-NA EUV steppers, which means that many chips are simply no longer feasible. (Stitching two half exposure windows is possible, but expensive and slows down throughput.)
Good points; mix and match, appropriate process node, improves silicon yield, well, that raises the good package question it's a two-sided proposition, scalability I will point out a difference between AMD MCM and Intel tiles, AMD is a multiprocessing system in package especially when you move to Epyc otherwise Ryzen and Meteor are SIPs (system in package) unless its a SOC (APUs), and yes physical interconnect eats power and effects latency. One aspect AMD and Apple both share are lower volume economies of production scale in relation Intel that the production chain makes up to secure their margin objective on lower volume by increasing AMD and Apple laptop pricing. mb
Typos right off the bat: the U-series chips offer 12 cores total (2P+8E+2LP-E). Another thing to mention right away: ALL of the series offers the same 8E+2LP-E setup. We already know the 2LP-E cores cannot be disabled/changed in any way from their setup on the SoC Tile, but it is curious Intel didn't offer any 4E setups like before (maybe the E-core island on the Compute Tile is 8-cores only now? I dunno).
Without speculating, it's entirely plausible, but in the same breath, this is Intel's first lineup and we won't know entirely until we see more SKUs or how well the current SKUs are adopted within the market.
Another thing I don't see mentioned here: the reason for the differing number change in the Ultra 7 164U and and Ultra 5 134U isn't just the TDP being set to 9-Watts: these ones also ONLY have support for soldered-on LPDDR5 memory, and cannot be used with traditional SO-DIMM DDR5 modules. My guess is that is one of the ways Intel got that overall TDP reduced.
I'm assuming Meteor Lake is too soon for CAMM2, being that it just got standardized by JEDEC. Probably in the cards for the next 2H 2024 launches from AMD and Intel though.
I'm hoping Samsung's Galaxy Book4 Ultra has CAMM, seeing as Samsung has decided to implement their own design. They're both supposed to launch in 2024.
DDR5 is awfully inefficient (5~6x pJ/b compared to LPDDR5) to couple with such low TDP variants. Two DIMMs of DDR5-5600 will consume around 20W at peak BW.
No Socket Packaged variants of Meteor Lake and so no InWin Chopin or ASRock Desk Mini Builds based on Meteor lake. And all the Rumors of Ryzen 8000G are little solace because AMD's not going to be motivated to Release any Ryzen 5000G replacement for DIY if there's no Meteor Lake S competition to help encourage AMD to stop the Milking of the Ryzen 5000G series for DIY Inwin Chopin and ASRock Desk Mini very small form factor builds.
The biggest looser here is DIY and the lack of any Socket Packaged APUs or SOCs with more powerful integrated graphics for very small builds where the form factor is too small to accommodate any dGPU!
Interesting that multiple youtube influencers were able to get early access to Asus/MSI laptops and were able to get their quick reviews in. But intel did not bother to send them to anandtech. Does this mean this site is no longer go to site for product review and audience seem to go to youtube for those early reactions. I think Anandtech need to leverage their youtube channel which seem to be inactive.
Sadly, the writing has been on the wall for that probably since Ian Cutress left the publication. However, Ryan Smith and co still write good, in-depth, articles, and I enjoy the discussion, so I'll keep coming here first before going to other sites. And, also, while this site has annoying ads, it's nothing compared to TomsHardware, which literally crashes half the devices' web browsers I view it on!
Yes, we're here to support AT all the way. But it could be a good idea to start churning out some videos and blowing those others on YT out of the water.
No, they're the same number. x8 PCIe 5.0 x12 PCIe 4.0 (x4/x4/x4 bifurcation) x8 "PCIe 4.0 for general I/O". So that means it's DMI/chipset lanes, but worded to sound like they are PCIe and not DMI.
Where as 13th gen desktop has x16 PCIe 5.0 (x8/x8 bifurcation) x4 PCIe 4.0 x8 DMI 4.0 (going directly to the chipset) Z690 chipset supports a further x20 PCIe 4.0 lanes x8 PCIe 3.0 lanes
"So that means it's DMI/chipset lanes, but worded to sound like they are PCIe and not DMI."
There's no DMI (or equivalent) with MTL. There are PCIe lanes coming from both the SoC and I/O tiles, but the backhaul between those tiles is a high-bandwidth D2D interconnect. The I/O tile is mostly where they put the TB4 controllers, which are sizable and are placed in a separate tile to help spread out the signal routing.
The general I/O lanes on MTL are just that. They're lanes not dedicated to another purpose (such as storage).
The statement about not having actual review units available at launch might not be as universal as written here. A couple of European sites got at least a few laptops sampled (including some from Acer). So, some laptops are out there already. Not sure why Anandtech didn't get one for a first look?
Intel's advantages have always been its ecosystem and advanced manufacturing processes. Now, AMD has TSMC's processes, and Apple has its macOS. With the future of ASIC design relying on machine learning, Intel must find new strengths. After all, a company is always just 18 months away from collapse.
if intel "fake" 4 is on par to tsmc n4, intel won't use tsmc for the soc chiplet.
on graphic side, 7840u will still win againts the u7u. u7u only has 4 xe cores which equivalent to around 2,1 tflops which is much less than 12 cores rdna3
What are you talking about? The SoC chiplet is 6nm LP from TSMC. Intel literally doesn't aim as low as TSMC LP nodes typically, even when they do put out a full node, so TSMC was cost-optimal for the requirements.
Be wary of these numbers. The new Intel systems provided for review were equipped with LPDDR5-7200 memory while the AMD system was equipped with DDR5-6400. In addition, the systems going out for retail availability are using DDR5-6400 as well, NOT the LPDDR5-7200 the review systems used, giving them much lower performance.
Intel is at TSMC to learn by benchmarking their TSMC 6 and 3 tapers against Intel in-house. Intel is also at TSMC for a very long time likely as much as 100 M units annually of components that Intel cannot effectively produce because for those specific components their cost : price / margin is < Intel variable cost approximately $39 to $55 where range high touches a marginal cost value. mb
raptorlake 1370p 28w cpu base clock faster P core 3ghz,e core 4ghz than meteorlake 165H 28w base clock,but 165H gpu clock faster 9ghz with 8 core 2.3ghz,than 1370p 6core 1.5ghz
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
89 Comments
Back to Article
Marlin1975 - Thursday, December 14, 2023 - link
So they are still so far behind they have to use desktop power to catch up?Does not seem like a good chip if you need to use 57w in a "15w" chip
I am guessing if they are forced to use lower power state, like most oems will have to do, their "gains" will not show any more.
TEAMSWITCHER - Thursday, December 14, 2023 - link
Yea.. Apple pretty much has everyone beaten with Mobile computers.zamroni - Friday, December 15, 2023 - link
amd 7840u beats m2 pro in many aspectshttps://www.notebookcheck.net/R7-7840U-vs-M2-Pro-v...
sharath.naik - Friday, December 15, 2023 - link
AMD 7840u beats Meteor Lake too. Looks like intel was playing naughty by running Meteor Lake at higher power in their comparison slides.https://www.notebookcheck.net/Intel-Meteor-Lake-An...
Gondalf - Monday, December 18, 2023 - link
All depends on laptops power limitations, Zen have a very crappy cooling system. Surprised you do not consider that in mobile it is all about cooling. Apple know well this.TEAMSWITCHER - Friday, December 15, 2023 - link
I'm so sorry for your loss... but PC Laptops simply cannot compete with Apple .. So many reviewers have found that all PC laptops throttle severely when you unplug them - they have to because PC's a YEARS behind apple in power efficiency. In addition to performance while on battery, longer battery life, better screens, better trackpads, better speakers, and rugged aluminum chassis.It's a total butt whooping....
Qasar - Friday, December 15, 2023 - link
yep... and at 2 to 3x the price too...max - Saturday, December 16, 2023 - link
Nope. That's lie. Same build and display quality range PC laptop costs fairly the same as Apple.Qasar - Saturday, December 16, 2023 - link
maybe where you are.. but here, apples products are usually at least twice the price as the equivalent competition, performance wise, i wasnt talking about build and display quality.nestle96 - Sunday, December 17, 2023 - link
Apple was selling crappy 5400 rpm hdd in 2019 imacs..stockholm syndrome.lemurbutton - Saturday, December 16, 2023 - link
I've found that any PC that is in the same build quality as a Mac will generally cost MORE than a Mac. And of course, it still doesn't have Apple Silicon.Qasar - Sunday, December 17, 2023 - link
" it still doesn't have Apple Silicon."like thats a positive selling point to most people :-)
zamroni - Monday, December 18, 2023 - link
many business grade PC laptops with mil-std durability certification are much less expensive than comparable macbookBruzzone - Tuesday, December 19, 2023 - link
AND AMD laptops, every new generation top shelf, are not x2 to x3 the price of an Intel top shelf laptop as well? Instead of an Apple an AMD fashion statement? mbQasar - Wednesday, December 20, 2023 - link
not here... for the most part, intel and and amd based notebooks are comparable in price depending on specs.. but apples are always quite a but more expensive compared to what are are getting.. it isnt called the apple tax for no reasondontlistentome - Friday, December 15, 2023 - link
Screens, trackpads, speakers, cases are just as good on PC as Mac, in a similar price bracket.Power efficiency - yeah, mac has that won currently.
But for the case, i'd take a carbon/composite case over aluminium anyday. Far less likely to scratch stuff when you throw it in a bag, and a ton lighter.
I use a macbook m1 and an X1 nano G2 for work - the mac is 35% heavier than the thinkpad - that's huge.
lemurbutton - Saturday, December 16, 2023 - link
Macs have far superior screens, trackpads, and speakers than Windows laptops. Not even close.lemurbutton - Saturday, December 16, 2023 - link
The reason Macs are heavier is because it uses an all metal enclosure vs plastic.Qasar - Sunday, December 17, 2023 - link
so says the apple fan boynestle96 - Sunday, December 17, 2023 - link
That is a lie. You cant test every windows laptop out there so if you claim all windows laptops perform the same then you must be a liar or ignorant maybe both.NorthRocks - Monday, December 18, 2023 - link
What in earth? I just checked some Mac Book Pro 2023 prices & specs, and realised how ridiculous the pricing scheme is. Basically, since DIMM or M.2 can't be upgraded by user, you're hostage to Apple's outrageous prices. Limited memory and storage sizes for those prices is major letdown. User replaceable RAM and M.2 drives are essential to me.name99 - Thursday, December 14, 2023 - link
The two interesting things to look at are:- cost compared to "equivalent" earlier chips
- power when doing something that crosses chiplet boundaries (eg game uses CPU+GPU)
For all the advantages of chiplets in some dimensions, it seems unavoidable that they will cost more, and that they will use more power in chip-to-chip links.
How *much* more of either? Well, that's the question...
Calabros - Thursday, December 14, 2023 - link
A U cpu with 57W TDP. What a time to be alive.lemurbutton - Thursday, December 14, 2023 - link
I guarantee you that some laptop manufacturers will even boost beyond 57w.Gothmoth - Thursday, December 14, 2023 - link
intel... selling snake oil harder than last time...lolHulk - Thursday, December 14, 2023 - link
That comment is worthy of recognition. I actually "laughed out loud" a little. Thanks for the mid-day break.Terry_Craig - Thursday, December 14, 2023 - link
It's funny how they have a U line, with a base TDP and lower turbo, but they compare the H models with AMD's lower TDP competitors.Thorburn - Thursday, December 14, 2023 - link
The AMD Ryzen 7840U being compared to has a 28W Default TDP.yankeeDDL - Friday, December 15, 2023 - link
Yep. Totally. Ryzen still crushed Core in Perf/efficiency. a 30% performance advantage is not something to boast about if the power consumption is ~2X (I am speculating, but the peak for the 7840U should be around 30W).lemurbutton - Thursday, December 14, 2023 - link
TDP have no meaning anymore. AMD and Intel can set a low TDP but it'll turbo well pass it. And laptop makers will almost always boost even beyond the turbo. This is for both AMD and Intel.The only laptop maker that actually makes good, fast, quiet, cool laptops is Apple and soon to be Qualcomm.
Terry_Craig - Thursday, December 14, 2023 - link
The TDP of laptops can vary significantly, with manufacturers making adjustments to accommodate the format, cooling, and specific model focus. Intel is the one that stretches Boost PL2 the most in relation to PL1.Btw, Apple also doesn't stick to the TDP specified on paper, the M3 max (56w) easily exceeds 140w at full load. Their main advantage besides having ASICs to accelerate specific software is full control of both sides, software and hardware. But there's still the downside of being stuck with Apple's OS, which doesn't work well for everyone.
Apple Worshipper - Thursday, December 14, 2023 - link
Lol. Easily exceeds ? Where did you get that fact ?Dante Verizon - Thursday, December 14, 2023 - link
Look for the product review on specialized websites, Apple guyname99 - Thursday, December 14, 2023 - link
The real issue is where do you get that claim of M3 Max (56W) TDP...Apple does not, as far as I know, use TDP as any sort of metric, either internally or in advertising.
They will tell you what sized power adaptor ships with the laptop, they will tell you how loud it gets, they will tell you how long the battery lasts for various tasks – but they will not tell you a TDP.
If you look at various review sites, they are just MAKING UP numbers. One site claims M3 Max TDP is 30W, another claims it's 78W, you claim it's 56W – sure sign that people are JUST MAKING THINGS UP...
So you're welcome to claim that M3 Max exceeds 140W at peak (though I doubt that doing so was "easily" achieved);
you're not however entitled to be outraged by this as some sort of lie on Apple's part where they exceed some supposed designed or advertised spec.
Terry_Craig - Thursday, December 14, 2023 - link
"70W USB-C Power Adapter (M3 or M3 Pro with 11-core CPU) or 96W USB-C Power Adapter (M3 Pro with 12-core CPU or M3 Max)"They probably used the power adapter as a reference or asked Apple for information.
zamroni - Friday, December 15, 2023 - link
16" with m3 max uses 140w charger.that means at least the max tdp is 80% of it, i.e. 112 watt
redondo - Friday, December 15, 2023 - link
notebookcheck m3 max review clocks the max tdp at 145 watts.web2dot0 - Sunday, December 17, 2023 - link
https://www.notebookcheck.net/m3-7Y30-vs-M3-vs-App...78W buddy 😂😂😂
PeachNCream - Friday, December 15, 2023 - link
Agreed so much. TDP has been a useless number for quite some time now pretty much across the board while, at the same time, processors have been using loads more power and generating a lot more heat. And people wonder why I just pair a bluetooth keyboard with my phone and generally ignore the desktop/laptop market. I'll pass on the noisy cooling fans thanks.sharath.naik - Thursday, December 14, 2023 - link
They just about caught up with AMD HS processors like 7840HS. So, they show the comparison to 7840U. I would have been fine with catching up, BUT why is it not better in battery life in video playback tests!! The whole point of TILES was to boost the battery life in these light tasks. Yet every single benchmark on YouTube shows video playback battery life only matching AMD now. What was the point of these TILES then? they took a hit of single thread performance for it.name99 - Thursday, December 14, 2023 - link
Tiles were NEVER going to be lower power!The fact that so many people assumed this shows just how ignorant most supposedly tech people are about anything they don't hear in an Intel ad...
The supposed advantages of tiles/chiplets are
- ability to mix and match to create different products.
Maybe valuable to AMD and Intel, negative value to customer, especially once you get to Intel with crazy 1000+ SKU plans...
- ability to improve yield because chiplets are smaller.
This is of dubious value given real yields, and yield loss in packaging. But people will not let this one die because it's part of an elaborate theology that insists yields for consumer sized chips are high, which in turn justifies the wide range of SKUs from CPU vendors.
- ability to place chiplet on "appropriate" technology node.
You realize "appropriate" technology node means OLDER technology node, right?
- scalability.
This is, IMHO, pretty much the ONLY place where chiplets actually make sense. They allow you to build larger products than would otherwise be reasonable.
You'll also notice that scalability is the *main* way AMD uses chiplets, and the *only* way Apple uses them. I expect it will also be the only way nVidia uses them.
So look at the pattern. As usual, Intel is being driven by MARKETING, not technical considerations. Marketing decided that chiplets would solve a bunch of their problems (something new and shiny! a way to create even more SKU confusion that is already the case!)
And don't get me started on Ponte Vecchio.
I don't want to have to keep kicking Intel, but, honestly, I suspect tiles will turn out to be exceedingly disappointing for users – and for entirely predictable reasons.
It didn't have to be this way! Intel could be using tiles the way AMD does, in very small doses to create mid and highrange products; or like Apple does, to create only high-range products that can't be created any other way. This would work for Intel and for their customers. But it would be less exciting for Intel Marketing. So...
meacupla - Thursday, December 14, 2023 - link
IMO, best use of chiplets is CPU+GPU(powerful)PCIe 5.0 power consumption is no joke when stretched out on a mobo.
Now, if only Intel/AMD would have released one already. Instead we get to wait until 2025 for Strix Halo.
ricebunny - Thursday, December 14, 2023 - link
The reviews prove you wrong. In light workloads, Meteor Lake exhibits 2.5x longer battery life compared to Raptor Lake. By moving the CPU cores and the GPU on separate tiles they are able to fully power them down in workloads where the 2 ultra low power cores on the SOC tile are sufficient.name99 - Thursday, December 14, 2023 - link
You hardly need chiplets to- power down high performance cores AND
- include very low power cores
Rather than rushing to defend Intel READ WHAT I SAID!
I said that I expect the problem to be in situations that involve substantial chiplet communication, like games. I did NOT say that idle life would be negatively affected by chiplets because that would be a stupid thing to say, obviously wrong.
Farfolomew - Saturday, December 16, 2023 - link
I like your posts, name99, but the very first thing you said was "Tiles were NEVER going to be lower power!"name99 - Saturday, December 16, 2023 - link
The lower power for the idle case is NOT the result of the chiplet design. Is that so hard to understand?People like me try to educate, but we assume a public that is willing to do its part, in this case understanding the difference (and reasons for)
- lower IDLE power
- lower HIGH-DEMAND power
If every posting attempt that tries to educate people is turned into an adversarial contest of "who can misinterpret this post in the silliest way" the result will be that even fewer people who actually understand anything will post in public. Is that a result people want?
lmcd - Monday, December 18, 2023 - link
Tiles with uniform process wouldn't reduce power, but Intel selected LP nodes for its SOC and IO tile (to my knowledge), enabling lower power consumption.Also -- power gating within a piece of silicon is excellent but can't compete with turning off an entire die.
So yes, chiplets aren't required for every strategy used to power down cores and include low power cores, but they do enable strategies that require competing fabrication technology selections.
BINARYGOD - Thursday, December 14, 2023 - link
"Marketing decided that chiplets would solve a bunch of their problems."divorced from reality the fanatics remain. So, let's get this straight, after everyone and their mothers gave a host of reasons to go chiplet, its mostly just yields in the end (you claim), but Intel doing the same thing (going chiplet for many claimed reasons but yield is the only 'correct' one) is a sign they are off the farm and doing things for entirely marketing? How is that not also the case for the other companies? Anyway, you also did that "let's complain about 1009's of sku's even though any one customer wont care about most of them" crusty old fanboy-sourced talking point, so why I expected the rest of your post to be better, is beyond me
OreoCookie - Sunday, December 17, 2023 - link
Quite a few misconceptions here:- Appropriate process nodes:
You are forgetting e. g. (HBM) memory, which is produced on entirely different process nodes. As the size of memory cells does not shrink the same way as logic does, SRAM also need not be produced with the latest process node, all it'd do is make your product more expensive. Availability and price are also consumer concerns, too. ;-)
- Ability to mix-and-match
Being able to get more products and for a company to react flexibly to market demands is a boon to customers. AMD being able to replace Zen 4 by Zen 4c tiles or AI accelerators to create new products is great for everyone as taping out monolithic designs could be prohibitively expensive.
- Power
It depends on the implementation and whether e. g. you integrate memory and other things. Integration of more and more components that used to sit in different places on the motherboard onto a SiP does lower system power over time. However, it seems that Intel's implementation is not as successful and power-efficient as that of other companies.
- Using tiles like AMD and Apple
This is a fundamental misunderstanding of the underlying technologies. E. g. Apple has been using advanced packaging techniques in many of its products for many years, including iPhones, the Apple Watch and now all Macs. It seems that you think only the Mx Ultra series of processors qualify. But that is not the case. All recent Apple SoCs integrate memory, for example.
Ditto for AMD: their Zen 4-based Dragon range and Phoenix laptop CPUs use chiplets throughout, for example.
- Chiplets are cheaper and once High-NA EUV comes online, absolutely necessary
The biggest oversight is in your post: chiplets make products a lot cheaper (https://www.nextplatform.com/2021/06/09/amd-on-why... Bigger designs are closer to the reticle limit, the size of a single stepper exposure window. That window will be cut in half with High-NA EUV steppers, which means that many chips are simply no longer feasible. (Stitching two half exposure windows is possible, but expensive and slows down throughput.)
Bruzzone - Tuesday, December 19, 2023 - link
Good points; mix and match, appropriate process node, improves silicon yield, well, that raises the good package question it's a two-sided proposition, scalability I will point out a difference between AMD MCM and Intel tiles, AMD is a multiprocessing system in package especially when you move to Epyc otherwise Ryzen and Meteor are SIPs (system in package) unless its a SOC (APUs), and yes physical interconnect eats power and effects latency. One aspect AMD and Apple both share are lower volume economies of production scale in relation Intel that the production chain makes up to secure their margin objective on lower volume by increasing AMD and Apple laptop pricing. mbQasar - Wednesday, December 20, 2023 - link
source for this ? oh wait.. never mind.. you still dont post sources....NextGen_Gamer - Thursday, December 14, 2023 - link
Typos right off the bat: the U-series chips offer 12 cores total (2P+8E+2LP-E). Another thing to mention right away: ALL of the series offers the same 8E+2LP-E setup. We already know the 2LP-E cores cannot be disabled/changed in any way from their setup on the SoC Tile, but it is curious Intel didn't offer any 4E setups like before (maybe the E-core island on the Compute Tile is 8-cores only now? I dunno).nandnandnand - Thursday, December 14, 2023 - link
I'll be shocked if they don't release a 2P+4E Ultra 3 115U eventually, or silently.Gavin Bonshor - Thursday, December 14, 2023 - link
Without speculating, it's entirely plausible, but in the same breath, this is Intel's first lineup and we won't know entirely until we see more SKUs or how well the current SKUs are adopted within the market.Gavin Bonshor - Thursday, December 14, 2023 - link
Thanks for highlighting that!meacupla - Thursday, December 14, 2023 - link
I can't wait to use these in mini-PCs that have better thermal overhead.NextGen_Gamer - Thursday, December 14, 2023 - link
Another thing I don't see mentioned here: the reason for the differing number change in the Ultra 7 164U and and Ultra 5 134U isn't just the TDP being set to 9-Watts: these ones also ONLY have support for soldered-on LPDDR5 memory, and cannot be used with traditional SO-DIMM DDR5 modules. My guess is that is one of the ways Intel got that overall TDP reduced.meacupla - Thursday, December 14, 2023 - link
SODIMM DDR5 sucks and CAMM works with LPDDR5.So maybe some ultra thins will get user replaceable/upgradeable LPDDR5 RAM.
NextGen_Gamer - Thursday, December 14, 2023 - link
I'm assuming Meteor Lake is too soon for CAMM2, being that it just got standardized by JEDEC. Probably in the cards for the next 2H 2024 launches from AMD and Intel though.meacupla - Friday, December 15, 2023 - link
I'm hoping Samsung's Galaxy Book4 Ultra has CAMM, seeing as Samsung has decided to implement their own design.They're both supposed to launch in 2024.
quarph - Thursday, December 14, 2023 - link
DDR5 is awfully inefficient (5~6x pJ/b compared to LPDDR5) to couple with such low TDP variants.Two DIMMs of DDR5-5600 will consume around 20W at peak BW.
FWhitTrampoline - Thursday, December 14, 2023 - link
No Socket Packaged variants of Meteor Lake and so no InWin Chopin or ASRock Desk Mini Builds based on Meteor lake. And all the Rumors of Ryzen 8000G are little solace because AMD's not going to be motivated to Release any Ryzen 5000G replacement for DIY if there's no Meteor Lake S competition to help encourage AMD to stop the Milking of the Ryzen 5000G series for DIY Inwin Chopin and ASRock Desk Mini very small form factor builds.The biggest looser here is DIY and the lack of any Socket Packaged APUs or SOCs with more powerful integrated graphics for very small builds where the form factor is too small to accommodate any dGPU!
trivik12 - Thursday, December 14, 2023 - link
Interesting that multiple youtube influencers were able to get early access to Asus/MSI laptops and were able to get their quick reviews in. But intel did not bother to send them to anandtech. Does this mean this site is no longer go to site for product review and audience seem to go to youtube for those early reactions. I think Anandtech need to leverage their youtube channel which seem to be inactive.Farfolomew - Saturday, December 16, 2023 - link
Sadly, the writing has been on the wall for that probably since Ian Cutress left the publication. However, Ryan Smith and co still write good, in-depth, articles, and I enjoy the discussion, so I'll keep coming here first before going to other sites. And, also, while this site has annoying ads, it's nothing compared to TomsHardware, which literally crashes half the devices' web browsers I view it on!GeoffreyA - Saturday, December 16, 2023 - link
Yes, we're here to support AT all the way. But it could be a good idea to start churning out some videos and blowing those others on YT out of the water.wr3zzz - Thursday, December 14, 2023 - link
Finally a 9W chip. There hasn't been a new fanless notebook since I got my Spectre Folio 5 years ago.Dante Verizon - Saturday, December 16, 2023 - link
This thing gets super weak bellow 25wballsystemlord - Thursday, December 14, 2023 - link
So these chips have a total of 28 lanes of PCIe? That's more than the desktop chips.meacupla - Thursday, December 14, 2023 - link
No, they're the same number.x8 PCIe 5.0
x12 PCIe 4.0 (x4/x4/x4 bifurcation)
x8 "PCIe 4.0 for general I/O". So that means it's DMI/chipset lanes, but worded to sound like they are PCIe and not DMI.
Where as 13th gen desktop has
x16 PCIe 5.0 (x8/x8 bifurcation)
x4 PCIe 4.0
x8 DMI 4.0 (going directly to the chipset)
Z690 chipset supports a further
x20 PCIe 4.0 lanes
x8 PCIe 3.0 lanes
Ryan Smith - Friday, December 15, 2023 - link
"So that means it's DMI/chipset lanes, but worded to sound like they are PCIe and not DMI."There's no DMI (or equivalent) with MTL. There are PCIe lanes coming from both the SoC and I/O tiles, but the backhaul between those tiles is a high-bandwidth D2D interconnect. The I/O tile is mostly where they put the TB4 controllers, which are sizable and are placed in a separate tile to help spread out the signal routing.
The general I/O lanes on MTL are just that. They're lanes not dedicated to another purpose (such as storage).
eastcoast_pete - Thursday, December 14, 2023 - link
The statement about not having actual review units available at launch might not be as universal as written here. A couple of European sites got at least a few laptops sampled (including some from Acer). So, some laptops are out there already. Not sure why Anandtech didn't get one for a first look?thomasjkenney - Thursday, December 14, 2023 - link
Typo: "...low insensitive workloads..." maybe should be "...low intensity workloads..."?Ryan Smith - Friday, December 15, 2023 - link
Thanks!tonsui - Friday, December 15, 2023 - link
Intel's advantages have always been its ecosystem and advanced manufacturing processes. Now, AMD has TSMC's processes, and Apple has its macOS. With the future of ASIC design relying on machine learning, Intel must find new strengths. After all, a company is always just 18 months away from collapse.zamroni - Friday, December 15, 2023 - link
if intel "fake" 4 is on par to tsmc n4, intel won't use tsmc for the soc chiplet.on graphic side, 7840u will still win againts the u7u.
u7u only has 4 xe cores which equivalent to around 2,1 tflops which is much less than 12 cores rdna3
lmcd - Monday, December 18, 2023 - link
What are you talking about? The SoC chiplet is 6nm LP from TSMC. Intel literally doesn't aim as low as TSMC LP nodes typically, even when they do put out a full node, so TSMC was cost-optimal for the requirements.tipoo - Wednesday, December 20, 2023 - link
TSMC N7 has 36nm fin pitches, fab naming is all fake if you were awake for the last decademeneguzzo68 - Friday, December 15, 2023 - link
Does these Meteor Lake CPU include Wi-Fi 6E without Bluetooth ? What a shame. Is it so ?lmcd - Monday, December 18, 2023 - link
lol no, it includes both with CNViBlastdoor - Friday, December 15, 2023 - link
Meanwhile, apple/tsmc already hard launched 3nm M3-based laptops, so intel has been totally lapped in laptops.I
minde - Friday, December 15, 2023 - link
raptorlake u 1365 with 96eu gpu 6core 1.3ghz= 8ghz ,meteorlake 165u 4core 2ghz 62eu=8ghzminde - Friday, December 15, 2023 - link
raptorlake H 6core 96eu gpu 1.5ghz=9ghz,meteorlake H 8core 128eu 2.3ghz=18ghz.meteorlake H core ultra 7/9 gpu 2 times faster
dgingeri - Friday, December 15, 2023 - link
Be wary of these numbers. The new Intel systems provided for review were equipped with LPDDR5-7200 memory while the AMD system was equipped with DDR5-6400. In addition, the systems going out for retail availability are using DDR5-6400 as well, NOT the LPDDR5-7200 the review systems used, giving them much lower performance.shabby - Saturday, December 16, 2023 - link
If tsmc's 5/6nm fab catches on fire and they can't supply intel with parts whats going to happen?shabby - Saturday, December 16, 2023 - link
Just seems strange for them to rely on a competitor for a part in one of their chips.lmcd - Monday, December 18, 2023 - link
Intel can fab the SoC and IO tile on Intel 7 if they have to, with significant legwork and reduced performance.The GPU tile would end up needing to be gimped to make it work.
Bruzzone - Tuesday, December 19, 2023 - link
Intel is at TSMC to learn by benchmarking their TSMC 6 and 3 tapers against Intel in-house. Intel is also at TSMC for a very long time likely as much as 100 M units annually of components that Intel cannot effectively produce because for those specific components their cost : price / margin is < Intel variable cost approximately $39 to $55 where range high touches a marginal cost value. mbminde - Saturday, December 23, 2023 - link
raptorlake 1370p 28w cpu base clock faster P core 3ghz,e core 4ghz than meteorlake 165H 28w base clock,but 165H gpu clock faster 9ghz with 8 core 2.3ghz,than 1370p 6core 1.5ghzgue2212 - Friday, December 29, 2023 - link
Triggered by Apple's M3 CPU/GPU Shared Memory (my M3 Max has 36GB) I'm wondering how much memory the Meteor Lake's ARC GPU can address.