Architecture All Access: Meteor Lake - Architecture Overview

[@jenniegafoor6643]

I think this is first time I understand processor Architecture very easily. ❤❤

[@__aceofspades]

Meteor Lake is a revolutionary change. New Intel 4 node, tiles, new P-core and E-core architectures, new low power e-cores, 2X IGP performance, AV1 encoding acceleration, etc. Meteor Lake has more changes than any other CPU architecture in over a decade. It basically takes the best of what Apple and AMD are doing, and adds even more on top of it and Intels leadership in performance.

[@marcobassini3576]

A thing where Intel consistently sucks is performance per $/€. I am sure that the upcoming Meteor Lake processors will continue this long tradition, placing themselves in the market as the most expensive processors for a given performance level. When I buy some food at the supermarket I always check the price as €/Kg to make an informed decision. I do the same when I buy a processor (€/performance).

[@yancgc5098]

Meteor Lake’s P-core is not new, it’s the same one that 13th gen Raptor Lake has but with more cache

[@IonorRea]

Meteor Lake being revolutionary is a nice hype term to raise the value of Intel stock shares but what will change in practice??? Previous premium "ultrabook" laptops with Intel's processors like HP Elite Dragonfly already have more than enough battery life for office tasks, so Meteor Lake efficiency improvements are more likely to make Intel look less inferior compared to jack-all-trades master of none Apple processors or Qualcom's smartphone derivates CPU modified for laptop needs than actually change user experience significantly when graphics performance will be still inferior to Apple's laptop well software optimized integrated graphics while battery life will be lagging behind efficiency oriented processors derived from smartphones used in Qualcom's Windows Arm64 laptops. Most of the performance per watt improvements for power users will be due to newer manufacturing processes instead of the addition of scalable processing cores design that for pluged-in performance tasks just taking away space that could be otherwise utilized by the more performance cores. Somehow I see the larger degree of on-demand processing power scalability as far more useful for servers where it can produce significant money savings on power consumption than actually making a difference in laptops other than allowing laptop manufacturers to make poor optimization and still getting away with that in a similar fashion as modern dedicated graphics cards made playable even VERY poorly optimized videogames because we already know that laptops with Intel processors are sufficiently efficient for office work related mobile use for over a half-decade if laptop manufacturers spend enough time with their product, not to mention further generations prior to Meteor Lake. I got over a 5-year-old 1kg Panasonic Toughbook (business line not rugged) 12,5" laptop that still gets around 10 hours of battery when the display is set over 200 nits running Win 11 with Intel's 8 gen 15W CPU despite Intel lagging in deployment of newest manufacturing tech in their processors, so how much will be latest improvements making difference outside servers in the real world other than making you feel good while looking at benchmarks?

[@lbgstzockt8493]

I think many people are willing to pay that price if it gives them good battery life. Look at Apple, their laptops are very expensive but the combination of battery life and performance is, as of yet, unrivaled. If Intel can change that they may have a very lucrative product on their hand and really change up the laptop market.@@marcobassini3576

[@conradwiebe7919]

I love the focus on energy efficiency. I think for home media servers it will be awesome to have the compute tile powered down during the long periods of time between use.

[@j340_official]

Congratulations on bringing up Meteor Lake. Looking forward to when you bring this new technology to Desktop, hopefully with thunderbolt 5!

[@HDRPC]

5 in 2024

[@NicolasChanCSY]

After watching a few videos on Meteor Lake, I have the following questions. Why 2 low-power E-cores instead of the normal 4? Is it because of area saving or power saving? (Edit: to be clear, I know there are 8 E-cores in the top config of MTL, but why just 2 in the SoC tile?) Network-on-Chip is mentioned in the technology deep-dives slides shared by presses, how is it different from previous ring bus?

[@tipoomaster]

There's more e-cores on the compute tile, the idea with the low power island is to be able to turn off the compute tile when demand is very low

[@NicolasChanCSY]

@@tipoomaster Yeah. But why 2, but not 4, which is the same number of cores in a normal E-core? It is interesting to know how they make this decision to modify their E-core. Because of die area restriction? Because data says that low demand tasks do not need more than 2 cores anyway? Because the low power design is shallower and narrower and has less cache?

[@tipoomaster]

@@NicolasChanCSY They're the same cores on issue width etc, just on different voltage rails afaik. It probably just fit in the die area of the IO tile, and if something calls for more performance than the LP-E cores you want the compute tile awake I guess. Future families may see more LP-E cores of course.

[@crispysilicon]

​@@NicolasChanCSYE clusters come in both 2 and 4 core now.

[@EnochGitongaKimathi]

Good questions. The NOC Fabric is something I'm also interested to learn more about. The SOC is designed to be as low power as possible, which would have greatly influenced the choice of type and number of CPU Cores. Why have a dedicated media engine? Why have dedicated display engine? Why not let the CPU and GPU run those workloads? I believe such workloads are common enough to want to do them efficiently. A lot of the PCs are bought for content consumption and content creation, they represent a huge segment. Since the Media Engine and Display Engine find a natural home in the SOC and do the heavy lifting more efficiently for those specific workloads, including Low Power Efficiency CPU Cores enables their use without using a lot of energy. On Desktop these choices make very little sense but when running on battery it makes a lot of sense. I have seen when new laptops come out, they always market the video playback battery test; because for a lot of people the most common and significant workload is video playback. This is my best guess and like you I look forward to learning if I'm at all close to the truth. The NPU is also low power so that assistants like CoPilot can be engaged via voice commands to wake the computer from sleep and do quick and simple tasks.

[@matthewhayes7671]

Excellent introduction to a very exciting new paradigm. Thank you!

[@_modiX]

Will this also run the Intel ME rootkit?

[@IntelTechnology]

Please know that the ME toolkit will be usually updated or available for any product. However, to use it, you will need the drivers and firmware that are offered by the system manufacturers. If you wish to discuss this further, please raise a ticket via intel.ly/40zs15h with details like the purpose of your query, how many systems you have, and Intel vPro availability.

[@tipoomaster]

Can't wait, waiting on a laptop just for Meteor Lake

[@SimoBenziane]

Never buy first generation of architecture change

[@tipoomaster]

@@SimoBenziane Meh I'm not going to wait forever and this has a great many improvements all at once, good jumping on point.

[@__aceofspades]

@@SimoBenziane Literally every 2 years is an architecture change... There is zero reason to avoid it. Only AMD has issues with first gen products.

[@SimoBenziane]

@@__aceofspades Tick Tock was supposed to be each year. It's been a long time since it was the case. It's not each 2 years that there is architecture change. I always buy when they change the process node, not architectural change. You, do you, but don't spread false info

[@HDRPC]

​@@__aceofspades100% true

[@anonymouscommentator]

This is an awesome video, i enjoyed it very much!

[@chrishull8707]

What about architectural changes in the Redwood Cove and Crestmont cores?

[@xinwu5427]

Question: what could I benefit from the NPU, if I only do some office work, e.g. document and presentation, visit some websites, watch videos etc?

[@moist_ointment]

One example would be video effects in Teams calls, like background blurring.

[@__aceofspades]

Both Microsoft and Adobe and a dozen other big software companies have signed on to use Intels NPU. There are far too many benefits to list, so I suggest you watch Microsoft's recent Windows AI presentation and look into what adobe is doing.

[@anonymouscommentator]

I assume intel will work closely with microsoft to get it implemented in windows and applications. Webcam video enhancement, audio enhancement, windows copilot and so on would come to mind even if you dont use specialized programs. i do however suspect for adoption to be rather slow. right now there are tons of companies implementing their own AI accelerators and i dont see a common standard (yet).

[@kelownatechkid]

Very nicely done!!

[@Shahawir]

What IP stands for?

[@IntelTechnology]

IP stand for Intellectual Property. For information on repartition compute intensive IP, please visit intel.ly/40OwHEM If you have any further queries, feel free to share the details here intel.ly/40OU6FZ

[@urimtefiki226]

All chips are in 3D configuration now.

[@Capeau]

Very well explained. Nice video

[@HDRPC]

The best chip in the world is coming. Thanks Intel. Keep it up

[@Petch85]

Engineering is all about compromises. You always talk about the good stuff, never the bad stuff. Can you maybe explain some of the compromises. Maybe the SOC fabrics new location increases latency to some other part of the chip, maybe having all the high compute parts in one corner makes it harder to cool the chip, maybe it causes thermal strain in the silicon, maybe you need to add two low power CPU cores into the SOC in order to have the other efficiency cores in the compute part of the chip, using the same production node, so to save energy you would like to close all parts of the compute, but the two extra cores requires some space too making the SOC part a little larger. Maybe you could have splid the performance cores and the efficiency cores of the compute part, but the they could not shear the same L3 cache. I think you know what I mean. To understand your choices we need to here about the compromises, not just all the good stuff.

[@bobpritham2660]

No company talks about compromises if they just launched the product. It's the review people who have to test and talk about shortcomings.

[@netbat4399]

how to beat apple performance?

[@adamkee97]

Intel is truly back this time...

[@marancibia1971]

excellent videos. congratulations!

[@velo1337]

wish id would actually work that great. think they need one or two iterations more to get it working right

[@Altirix_]

8:00 intel glue tm

[@anonymouscommentator]

never forget marketing bs 😂

[@ItsAkile]

Can't wait for the "re-balling my Meteror Lake tiles" videos, just kidding but wished you guys went more ham on the Graphics tile.

[@machinimaaquinix3178]

Cripes, why even buy Raptor Lake refresh? I'm waiting for this. Get em out quick!

[@jubeikibagamai5323]

Guess I'll wait one more year for my laptop upgrade

[@urimtefiki226]

I have talked with people who work in IT profession for decades, and I know more things than them about chips, is interesting why no one knows how the computer works. Probably the info is hidden or there in curiosity.

[@donaldswanson9647]

great chip. im using one of these right now. big little cores with thread director is rube goldberg. npu has no applications and will get no applications. no matter. everything else looks good.

[@jenniegafoor6643]

i wish 14th gen will beat AMD latest processor easily 💪💪💪💪💪💪

[@JayzBeerz]

They won’t since they are a refresh.

[@jenniegafoor6643]

@@JayzBeerz ok

[@xanderludwig]

How's that going?

[@impuls60]

Going to chiplets is driven by rising cost of manufacturing and not the hunt for performance. The increase in latency and resistance in the substrate will generate heat losses so this might not be so impressive performance wise. Lets see if they faceplant like Amd did with their latest gpu chiplet design.

[@Summanis]

Foveros is more efficient per bit than V-cache. I doubt it will eat into power budget or produce much heat, and since they didn't serialize their interconnect like Infinity Fabric it's likely that latency is greatly reduced as well. With smaller compute dies, you can bin higher clocks, similar to how AMD's low end clocks higher relative to their top end than Intels has been. This is all theoretical, but it looks promising.

[@__aceofspades]

Meteor Lake is primarily about efficiency. Slightly more performance than raptor lake but at 50% less power.

[@bobpritham2660]

I am first. Rest are suckers. 😂

[@DanielRodriguez-fg5ll]

I heard you are an undercover AMD staff. Welcome to the force.

[@MaeVisage]

Waiting for Intel Core i9-15900k to replace my Intel Core i9-9900k.