Intel Core Ultra Explained

Intel's Meteor Lake Core Ultra mobile chips are here, and they're exciting! Nearly DOUBLE the iGPU performance but... at what cost? And what is an NPU? And why are they changing the name? And when will this video description be over?

Purchases made through some store links may provide some compensation to Linus Media Group.

FOLLOW US
---------------------------------------------------  

MUSIC CREDIT
---------------------------------------------------
Intro: Laszlo - Supernova

Outro: Approaching Nirvana - Sugar High

CHAPTERS
---------------------------------------------------
0:00 Intro
1:17 Gaming Performance
4:15 Productivity Performance
8:20 NPU
9:30 Tile Architecture
11:30 NPU Benchmark
12:11 New Naming Scheme
14:20 Conclusion
16:31 Outro

[LinusTechTips]

Clarification: While the Intel Core Ultra 7 155H can be configured up to a 65W TDP, it was running in the mid-30s during our testing on this device both in productivity and in games, for example reaching an average of 36W in performance mode during our Cyberpunk benchmark. We apologize for not including this information in the video.

[davidpomerantz]

While the old naming was confusing, at least it was for new pc enthusiasts. Now it's confusing for everyone 😂

[gorgorgonzales]

No word on batterylife? That was intels biggest problem in the mobile sector as far as I am concerned..

[error-null]

PLEASE put some icons in the graphs. Just an intel/AMD/Apple logo will go a long way.
I constantly have to look back and forth to see which number is which laptop. If a logo is in my peripheral vision while looking at the bars, that would help a huge amount.

[nextnomura]

Remember when Intel says about "glued together" Epyc CPUs?

[DigitalJedi]

I'm super proud of what we've been able to do with Meteor Lake. Just like with Alchemist, this is a 1st-gen attempt at something really new from the old ways at Intel. I would expect some of your gaming stutter and encoding to go away with the full TDP. MTL is meant to be performant in areas that last gen wasn't, such as the iGPU and NPU, while maintaining competition with Ryzen 8000 and M3 where last gen did well. It looked like you laptop was pulling some punches, as as you mentioned, it was capped around 36W while the 1370P can draw 64W quite comfortably. I would expect to match the 1370P at 55W. I am more than happy to take questions about anything.

For the NPU vs GPU question, that's entirely the expected and intended behavior. Think of the NPU like an E-core for the GPU when doing I/O.

The tiles are my point of pride though. I'm not an architect in any form. I don't make cores or GPUs or anything like that. I worked on the physical interconnects between the tiles, and let me tell you guys, there's interesting stuff in there I wish I could talk about more. Hopefully I'll get to put a paper out on this around the launch window of Lunar Lake. When you mentioned modularity, I would like to point out that the sizes of tiles are not set in stone.

[FredFredBurger]

Intel: "something something glued-together CPUs". My my, how the turns have tabled.

[purewater_ch]

no way ludwig's bidet is in LTT store now

[Randomness]

Just to reiterate for that bit at 2:13. CONSISTENT FRAMES > HIGHER FRAMES

I will always take 30fps over 60 that's constantly dipping to 5 or 10 every couple of seconds

[corylyon]

Why so kittle mention about the power modes. The 155H in perf. Vs. Std. Modes were basically different experiences.

[aidenworth]

Small rant here but I feel like part of the reason why that older video did so good was because it was easily searchable. I’m not a fan of how LTT has been labeling their videos in the last few years since it’s hard to tell what the video actually is. This one’s better but if your earlier video was titled “Luke’s new 2024 gaming PC” It would have 100% gotten a click from me but I totally skipped over it. Also I’m slightly scared I’m already shadowbanned for these mild takes so if someone could give a thumbs up or comment here that would be awesome.

[Bmoney]

Audio sound a little odd on this one. A little more bass heavy and maxed out

[FlameMage]

I'd advocate for doing a bit of coloring for the Y-axis on the charts. We're seeking out the Intel Core Ultra 7 155H (Std. and Perf. Mode) by reading 5 lines of white text some. The video is all about that though so why can't the processor's name be in yellow or something to quickly say "hey it's this one/it's these two look here" faster so we can shift our attention to the bars. Even the bars could have distinct coloring but that might be too loud looking idk.

[Deeppurp]

Hey LTT - you can tell Linus' LAV mic is being muffled by being under his hoodie.

Tiles - Glue but intel branded!

[magfal]

You have an M3, it has an NPU.
You can run linux on any recent-ish high end Android smartphone and you've got access to a capable NPU there too.

I've used my One Plus 7 Pro as a network attached 7 TOPS NPU machine to do low power batch image processing tasks.

A USB Coral TPU is also an option.

[Pavlogal]

No power draw / power efficiency comparisons? I think that's pretty important for laptop CPUs

[AndreiKucharavy]

A quick note - single precision FLOPs/Watt is the current battle in the ML space. The price of SoA models training is now dominated by the electricity draw, just as their cost in inference, so any saves there, even at a price of lower performance is what matters. So if you want to benchmark NPUs, showing power draw and inference speed on a set of standard models is what will be most interesting to people who need them or want to integrate into their software.

[HermioneTheCat]

did linus just say "closlier" at 2:13? LMAO

[tuannguyenminh]

Did they really compare a high performance 65W max TDP CPU against an efficiency oriented CPU 30W max TDP (according to AMD website)? Also no efficiency mentioned in a laptop CPU review/"explained" video?

[joetheaugustine]

For not having an NPU to compare against, Apple has had a "16-core Neural Engine" in their M-series products since it was originally released in 2020.