Thursday, December 14th 2023

Intel Core Ultra "Meteor Lake" Processor Lineup Overview

On December 14 Intel launched its first generation Core Ultra "Meteor Lake" line of mobile processors, and here is a a brief overview of the various processor models on offer at launch, thanks to a compilation by ComputerBase.de. "Meteor Lake" is Intel's first completely disaggregated processor, in which its numerous components are broken up into chiplets fabricated on different foundry nodes that strike the right performance/Watt suitable to the component, all held together by Intel's Foveros packaging technology (an evolution in multi-chip modules with a design focus on reducing inter-chiplet latencies to levels comparable to components on a monolithic chip). "Meteor Lake" also introduces a 3-tiered heterogeneous CPU architecture, with the introduction of the low-power island CPU cores.

Intel's mobile processor lineup is broadly categorized into the U-segment, targeting thin-and-light and ultraportable devices; and the H-segment, targeting notebooks of conventional thickness. At launch, the Core Ultra H-segment, and U-segment processors will coexist with P-segment processor models from the 13th Gen Core "Raptor Lake" series; as well as the upcoming 14th Gen Core "Raptor Lake Refresh" HX-segment. The P-segment is positioned between the U- and H-segments, targeting a class of devices that either what to be thin-and-light mainstream notebooks, or higher performance ultraportables. The HX-segment caters to high performance gaming notebooks and mobile workstations.
In the first table above, you get a brief overview of how the Core Ultra U-segment SKUs compare to 13th Gen Core "Raptor Lake" U-segment SKUs that are currently in the market, and will gradually phase out over 2024. The U-segment SKUs are based on the "Meteor Lake" UP4 (type-4) package, which is designed for a compact PCB footprint, and lower Z-height. This chip uses a smaller Compute tile with a 2P+8E CPU core configuration, 12 MB of shared L3 cache; and a smaller Graphics tile that uses a compacted iGPU that only has 4 Xe cores worth 64 EU (512 unified shaders). The SoC tile is unchanged from the H-segment (UP3/type-3) package, however, the I/O tile is compacted for fewer PCIe lanes.

At the very top, is the Core Ultra 7 165U, which maxes out the UP4 package, enabling all 12 cores—that's 2 "Redwood Cove" P-cores, 8 "Crestmont" E-cores in the Compute tile; and 2 "Crestmont" low-power island cores located in the SoC tile. Its P-cores boost up to 4.90 GHz, and E-cores up to 3.80 GHz. The iGPU boosts up to 2.00 GHz. The processor's base power is set to 15 W, with 57 W maximum turbo power. Intel is preparing to launch a much more lower power variant of this chip, the Core Ultra 7 164U, which retains the core configuration, a slightly lower 4.80 GHz P-core maximum boost, and lower 1.80 GHz iGPU boost; but a staggering 9 W base power, with 30 W maximum turbo power. This isn't just due to lower clocks, but because of a more aggressive power-management profile.

Located a notch below is the Core Ultra 7 155U, which has the same core-configuration as the 165U and upcoming 164U, but lower clock speeds; with the P-cores boosting up to 4.80 GHz, E-cores up to 3.80 GHz, and iGPU up to 1.95 GHz. This SKU gets the same 15 W base and 57 W turbo power values as the 165U. And then there's the Core Ultra 7 134U, which retains the maxed out core-configuration for the UP4 package, but further lowers clock speeds to 4.40 GHz maximum P-core boost, 3.60 GHz maximum E-core boost, and 1.75 GHz maximum iGPU boost; but with the same aggressive power figures as the 164U, of 9 W base and 30 W turbo. Both the 164U and 134U are expected to be available some time in Q1-2024 unlike the others that are available starting now. These two chips lack a conventional DDR5 memory interface, and are restricted to LPDDR5X-6400 memory type.

Located a notch below is what we expect to be the bestselling models due to their middle-of-the-market pricing. The Core Ultra 5 135U has an identical core-configuration to the Core 7 models, with 2P+8E+2LP, but lower clock speeds of 4.40 GHz P-core boost, 3.60 GHz E-core boost, and 1.90 GHz iGPU boost; and with 15 W base and 57 W turbo power values. A notch below, is the Core Ultra 5 125U, clocked 100 MHz lower still.

We now move onto the H-segment, and here we see Intel switch over to the larger UP3 (type-3) package, which packs a larger Compute tile that contains a maximum of 6 "Redwood Cove" P-cores, a maximum of 8 "Crestmont" E-cores, and 24 MB of shared L3 cache. The package also gets a larger Graphics tile that has 8 Xe cores worth 128 EU, or 1,024 unified shaders. The SoC tile remains constant, but the I/O tile is larger, with more SerDes interfaces that are configurable as more PCIe lanes.

At launch, the Core Ultra 7 165H is the fastest processor model from the series, maxing out all components present on the chip. It features a 6P+8E+2LP configuration, with a maximum P-core boost frequency of 5.00 GHz, maximum E-core boost of 3.80 GHz, the full 24 MB of L3 cache, all 8 Xe cores, and an iGPU boost frequency of 2.30 GHz. The chip comes with a healthy 28 W processor base power, and maximum turbo power configurable between 64 W and 115 W (the latter is an OEM design choice). Intel is planning to launch an even faster model in Q1-2024, the Core Ultra 9 185H. This one is a beast, with 5.10 GHz maximum P-core boost, 3.80 GHz E-core boost, 2.35 GHz iGPU boost, and a maximum turbo power of 115 W, with a 45 W base power. To use this chip, notebook designers will need a 115 W-capable power design.

The Core Ultra 7 155H is positioned a notch below the 165H, and is currently available like it. The P-cores boost up to 4.80 GHz (that's 200 MHz lower than the 165H); while the E-core boost frequency is untouched at 3.80 GHz. The iGPU boosts up to 2.25 GHz. Much like the 165H, its base power is set to 28 W, and turbo power configurable between 64 W or 115 W. On both the 165H and 155H, the optional 115 W turbo power mode serves to enable better boost frequency residency, but requires OEMs to use a more capable VRM and cooling solution.

Things get interesting with the Core Ultra 5 H-series. The Compute tile in this chip is configured with 4 P-cores, 8 E-cores, and 18 MB of shared L3 cache (by disabling two P-cores and L3 cache blocks, it's not a physically smaller tile than the one on the 165H). The Graphics tile sees 7 out of 8 Xe cores being enabled, amounting to 112 EU or 896 unified shaders. The Core Ultra 5 135H features a P-core boost frequency of 4.60 GHz, and an 3.60 GHz E-core boost. The iGPU boosts up to 2.20 GHz. The Core Ultra 5 125H, on the other hand, features a 100 MHz lower P-core boost, while retaining all other specs of the 135H. Both these chips feature a 28 W base power, with configurable maximum turbo power values that are either 64 W or 115 W, depending on the OEM's choice.
On all processor models, regardless of their segment, Intel is enabling its Gen 3 NPU, with 2 NCEs (neural compute engines). Intel didn't put out the TOPS (teraOPS) figure for the NPU, but the whole-package AI performance for the top processor model, which it lists as 34 TOPS. If you recall from our recent coverage of the AMD Ryzen 8000 "Hawk Point" processor launch, AMD bumped up the throughput of its XDNA NPU from 10 TOPS on "Phoenix" to 16 TOPS on "Hawk Point," and mentioned a whole-SoC throughput of 39 TOPS for the top processor model.
Source: ComputerBase.de
Add your own comment

25 Comments on Intel Core Ultra "Meteor Lake" Processor Lineup Overview

#1
tabascosauz
If nothing else, it's pretty interesting to see what kind of idle power Intel can pull off here. Foveros is supposed to be a little bit ahead of both EMIB and Infinity/Fanout Link (whose efficiency potential seemed sabotaged by other unrelated design aspects) in terms of energy efficiency, and Intel is pulling out all the stops with the island E-cores.

Tempted to make a glue joke, but this is unironically some cutting edge glue :D

If this is the level of disaggregation that Intel has to resort to to maintain their usual levels of good availability, then APUs won't be staying monolithic for long (well, if AMD is as unsatisfied as everyone else with the state of Phoenix).
Posted on Reply
#2
Selaya
is the yield for 4nm that bad?
the die's about the size of a GA106, but cpu yields are different from gpu ig ... ?
Posted on Reply
#3
Solaris17
Super Dainty Moderator
Selayais the yield for 4nm that bad
its 7nm iirc. 4n (lack of m) is what intel calls there approximation of 4nm in comparison.

I will probably be picking up a 155H. I have been thinking of replacing my personal laptop and I am curious what kind of advancements have been made irl over the years.
Posted on Reply
#4
Selaya
oh i was referring to amd's phoenix since thats got really bad availability, im aware intel 4's 7nm
Posted on Reply
#5
Solaris17
Super Dainty Moderator
Selayaoh i was referring to amd's phoenix since thats got really bad availability, im aware intel 4's 7nm
ohh I see. I hope both can ramp up availability next year. I know there is a lot of AI muttering; but I am excited about both camps. The core counts keep going up and the power usage is low all things considered. I like the advancements in the mobile space. Personally.
Posted on Reply
#6
tabascosauz
Selayaoh i was referring to amd's phoenix since thats got really bad availability, im aware intel 4's 7nm
N4's yield is fine. If N4 was the problem then SD 8G2, SD 7G2, SD 8+G1, A16, and Dimensity 9000 would all be vaporware. Something else related to AMD's supply was clearly a huge problem, but it's hard to point the finger anywhere else since nothing else in their portfolio shared N4 or had any supply problems.

Phoenix had a little bit of time in actually meaningful thin laptop designs (ie. not HS in a gaming laptop, not Z1 in a handheld), but pretty much immediately after those designs actually hit the market (available, not announced) the entire family was abruptly shelved and now it's replaced by 8040 Hawk Point - just Phoenix in new clothes with a new NPU, to capitalize on "AI".

Meteor Lake announcement was today and Thinkpad announcement also today - X1 Carbon Gen12 is already available to order online. Phoenix Thinkpads spent about 6+ months in limbo after announcement before becoming available, at basically the same time as every other not-gaming Phoenix laptop, also announced and AWOL for most of 2023. So hopefully, all the issues have been solved with Hawk Point and Strix Point, but Phoenix essentially spent its entire life on the market not existing (with those aforementioned exceptions).

The other tiles in MTL-U are N5 and N6 products, so no problems there, and clearly Foveros has matured to a point where it is not a problem either.
Posted on Reply
#7
Crackong
Wait,
So yesterday you tell me your Ultra 7 part beats AMD by 11% in the 'Same 28W part'
Now you are telling me that 'Same 28W part' is actually has a PL2 of 115W ??

Posted on Reply
#8
R0H1T
Wait till you see the RPL refresh mobile parts o_O
Posted on Reply
#9
Onasi
I still maintain that Core Ultra is an atrociously dumb name and the shortening of the model numbers that came with it actually does nothing to make their mobile lineup more readable at a glance. Especially since new naming scheme co-exists with old. Just look at that table, Christ, what is this. Not like AMD is much better, but still.
Posted on Reply
#10
fancucker
No mention of the battery life leaks? Its up to 2x more efficient than the 7840U, A very promising start too, when the CPU Tile is replaced with the 30-40%+ higher IPC ARL its over for Apple and AMD.
Posted on Reply
#11
FoulOnWhite
I'm guessing the Core Ultra 9 185H will be more likely for gaming laptops.
CrackongWait,
So yesterday you tell me your Ultra 7 part beats AMD by 11% in the 'Same 28W part'
Now you are telling me that 'Same 28W part' is actually has a PL2 of 115W ??

Did you miss this bit- 28 W base power, with configurable >maximum turbo< power values that are either 64 W or 115 W?
Posted on Reply
#12
AnotherReader
Judging by the frequencies of the P cores, Intel 4 seems to be shaping up nicely. This might be a welcome return to form after the doldrums of the 10 nm, i.e. Intel 7, generation. However, the lack of independent reviews is concerning.
Posted on Reply
#13
RandallFlagg
Solaris17its 7nm iirc. 4n (lack of m) is what intel calls there approximation of 4nm in comparison.

I will probably be picking up a 155H. I have been thinking of replacing my personal laptop and I am curious what kind of advancements have been made irl over the years.
Ya, because TSMC N5 / N4 is 7nm using the old metrics too.
Posted on Reply
#14
Squared
AnotherReaderJudging by the frequencies of the P cores, Intel 4 seems to be shaping up nicely. This might be a welcome return to form after the doldrums of the 10 nm, i.e. Intel 7, generation. However, the lack of independent reviews is concerning.
There are a bunch of reviews on YouTube, but not from any sources I usually use. I think one of them reviewed the Apple M3 Pro and said that its regression in gaming performance was probably just an optimization thing. He had no idea that the move from 256 bit memory on the M2 Pro to 192 bit on the M3 Pro could tank graphics performance. That's par for competence among the Meteor Lake reviews I've seen.
Posted on Reply
#15
Aquinus
Resident Wat-man
I really wish Intel would stop doing stupid things with their naming schemes. 7 Ultra vs i7 makes no sense. At least with i3, i5, i7, and i9 you knew that 3 5 is bigger than 5 3 and so on. Intel really should just think of something that'll scale for the next 20 years and stop messing with it. All I want is a model number to tell me is the generation of a product and the SKU based on math where number X being larger than number Y means that one is a higher end SKU than the other. This whole "Ultra" thing just muddies the water. It's beyond dumb. If you really need to annotate a SKU, just add a letter to the end like every other freaking SKU.
Posted on Reply
#16
Squared
Looking at the charts, comparing to Raptor Lake, both the U-series and H-series chips see lower P-core boost clocks but similar E-core clocks. The H-series chips see around 30% more EUs and much higher graphics clocks accross the lineup. The U-series chips see fewer EUs but much higher graphics clocks.

So going off only these numbers, CPU performance is down, threaded CPU performance is unchanged, U-series GPU performance is unchanged, and H-series GPU performance is way better. But I think the E-cores are a new microarchitecture and there's more memory bandwidth, which should result in better multicore performance and better U-series GPU performance.

I've seen game benchmarks where the H-series outperforms the 12 CU Phoenix, and Phoenix 2 only has 4 CUs, so even the U-series chips might outperform AMD's offerings. So I think Intel has retaken the iGPU performance lead for the moment.
Posted on Reply
#17
Onasi
AquinusAt least with i3, i5, i7, and i9 you knew that 3 is bigger than 5 and so on.
I know this is just a typo, but it seems that the jokes about confusing Intels naming schemes are slowly seeping into reality and are starting to warp it.
Posted on Reply
#18
Aquinus
Resident Wat-man
OnasiI know this is just a typo, but it seems that the jokes about confusing Intels naming schemes are slowly seeping into reality and are starting to warp it.
Definitely a typo, but if we're not careful, Intel might actually do it. :laugh:
Posted on Reply
#19
GoldenX
Great to see the PL2 meme has reached laptops too.
Posted on Reply
#20
Wirko
The "i" that was recently sacrificed is exactly the "i" printed on the chip in this picture. As if Intel were embarrassed of their glorious past.

Posted on Reply
#21
kondamin
OnasiI still maintain that Core Ultra is an atrociously dumb name and the shortening of the model numbers that came with it actually does nothing to make their mobile lineup more readable at a glance. Especially since new naming scheme co-exists with old. Just look at that table, Christ, what is this. Not like AMD is much better, but still.
I won't be using the full name when discussing one.
the number part and letter/s at the end will do

I do wonder when the first official reviews will start popping up
Posted on Reply
#22
Prima.Vera
W. T. F. is with this new naming convention??
Why for the name of Jeezus Christ they had to make it so complicated and without any logical sense??
Posted on Reply
#23
tabascosauz
GoldenXGreat to see the PL2 meme has reached laptops too.
What do you mean "reached"? Mobile is where it started :laugh: when the entire stack got its doubled core count in 8th gen, all hell broke loose.

Unfortunately, until Intel can show that their general P-core design can pull out decent performance at half to 2/3 their existing PL2, efficiency outside of pure idle will remain abysmal for -U parts. Gonna reserve judgment for when more reviews come out, but Redwood Cove does not look so good on that front, even if they are doing pretty innovative things for idle and SOC tile power.

Rembrandt/Phoenix -U are generally delivering full load perf at 20-30W PPT in these relevant laptop designs, with either a graceful power/perf dropoff with time, or none at all.
Posted on Reply
#24
oooiii222
intel's so-called low-power island is like a joke in the face of the "H" suffix of the standard pressure processor, in the evaluation, this part of the two ultra-small core has almost no chance of any call. In addition, in the current windows environment, I really don't see the need to use npu for ai acceleration in low-power scenarios. I think Intel needs to completely change the scheduling of processors in low-power situations, and can be radical and directly shut down the CPU and GPU parts and only use low-power islands. In addition, as a surface user, I am looking forward to Intel's ability to separate the low-power island and io control unit into ultra-low voltage processors
Posted on Reply
#25
watzupken
So Intel has gone from selling high end cores to just E-cores... I wonder how much will Intel charge for 6 P-cores. I understand that the E-cores should be better in Meteor Lake, but still, is it still a much cheaper core.

The results are actually not good. Barring the iGPU which I believe only applies to the top tier H class Meteor Lake, the rest are getting cut down iGPU configurations. So in the ultra low power settings, it will be interesting to see how the 4 core iGPU will hold up against the iGPU on the Ryzen 7 7840U. Power efficiency have improved, but I don't think is a significant step.
Posted on Reply
Add your own comment
Dec 21st, 2024 22:41 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts