Meteor Lake Tech Tour Deep Dive
ฝัง
- เผยแพร่เมื่อ 26 มิ.ย. 2024
- Intel recently hosted a Meteor Lake tech tour at the company's Fab in Malaysia to cover Meteor Lake. In this video Mark chats with Dr. Ian Cutress of @TechTechPotato and dives deep into what they learned.
Read more on PCWorld.com: www.pcworld.com/article/20555...
*Note: Intel covered costs for PCWorld to attend but do not have any insight or control of coverage.
Buy PCWorld merch: crowdmade.com/collections/pcw...
Follow PCWorld for all things PC!
--------------------------------
SUBSCRIBE: th-cam.com/users/pcworld?sub_c...
TWITCH: / pcworldus
TWITTER: / pcworld
WEBSITE: www.pcworld.com
#intel #cpu #laptop
00:00 - Intro
00:42 - Fab Tour Info
02:49 - High Level Meteor Lake Details
06:35 - Meteor Lake Tiles - วิทยาศาสตร์และเทคโนโลยี
Miss you Gordon hope you're OK.
Based on my understanding of the video, the focus is on reviewing or showcasing mini PCs, excluding the GEEKOM brand. While I can't speak to those specific mini PCs mentioned, I have had positive experiences with GEEKOM mini PCs in terms of their powerful functionality for multitasking. The GEEKOM models I've used offer impressive processing power, ample RAM, and speedy storage options, allowing me to seamlessly handle multiple tasks simultaneously. The compact design and efficient cooling system also make them suitable for my engineering work.
Meteor lake is a revolutionary step forward for Intel. There are so many changes, this video is over an hour long and still only scratches the surface of what Intel is doing. I am more excited for Meteor Lake than I have been for ANY CPU or GPU in the last decade. Consumers are going to love the efficiency, and enthusiasts are going to love the CPU and GPU performance as well as the NPU.
the performance won't be anything special, but this revolutionary step (I do agree it is) will ENABLE big improvements in performance
Great video! Mark does a fantastic job👍
Interesting, good video, thanks! I appreciate that Ian endeavors as much as possible to use the term "Machine Learning" instead of the abused shortform "AI". Artificial Intelligence per se doesn't exist yet, but Neural Networks, Machine Learning and ML Tools do.
10:50 Gracemont, not Goldmont. Goldmont was the atom core from way back in 2016 powering Apollo Lake and Denverton.
Goldmont -> Goldmont Plus -> Tremont -> Gracemont -> Crestmont
Correction: XeSS does not use dp4a instructions in Intel's Arc GPUs; it uses the XMX cores. There is a separate implementation of XeSS for dp4a but that is only used for compatibility in GPUs that don't have XMX support.
Robert Hallock on HotHardware actually said this version of Arc does not support XMX, as confusing as that could be…
@@TypeErrorDubs While that's true, XeLPG on ARC is based on Xe architecture and accelerates DP4a faster due to concurrent execution of FP and Int/EM instructions.
the iGPU in meteor lake does not have XMX cores, or at least doesn't support the instructions needed for the highest performance and image quality. That's why meteor lake iGPU will use dp4a. Desktop GPUs use XMX cores as you have stated.
@@VideogamesAsArt thanks for the clarification, I didn't know that the iGPU version had limitations but makes sense.
waiting years for this moment, Amen
The die wafer slip was hilarious 😅
So, is Gordon still alive?
Rumors say the G-Man took him.
Kidding aside, they've said he's on leave (health reasons) and don't have more to share right now. Hopefully he's okay.
He accidentally got assassinated by catfolk.
Yeah, he is.
He died this year
I’ve been reading Gordon since the 90s. I got sad reading this, hope he is okay.
I guess the fallback for LP>E>P will be the windows Performance Profiles and whitelisting if it became a problem
Will this be the first time, that Ian actually reviews the Intel Arc GPU? they had them behind them in the videos for a while...
penang is a nice spot. lexlip in ireland near me. Kinda just fields.
Intel Meteor Lake - Core Ultra
Disaggregated Architecture
4 Main Tiles on top a Base Tile (Silicon Interposer)
EMIB for Tile to Tile
Foveros
Adamantine Silicon Interposer with 128MBs L4 Cache
Glass Core Subtrate
CPU Tile on Intel 4 featuring EUV.
SoC Tile with Low power e-cores and NPU looks very interesting
This must be challenging and fun to pull off.
Still not happy about Thunderbolt 5 not integrated in MTL
thanks for the timestamps
@@juGGaKNotEmpire welcome though not really time stamps but interesting topics relating to Meteor Lake or Core Ultra.
How will this help my gaming? Will an NPU make CPU focus less and not as much of a leap this generation?
Intel have mentioned in various write ups that they are not using Movidius as the basis of the NPU in this generation of chips. The most likey other contender is the Brainchip AKIDA IP that will go into the NPU. Intel have stated that the chiplet design of this generation will allow for "advanced intellectual properties" to be integrated into their design, and the IP from Brainchip matches the task perfectly.
No, you're wrong. I literally spoke the team lead for the NPU, and it's his Movidius team who are doing the design. It's literally Movidius v2.7 in documents.
The architecture for the NPU shown in Intel videos closely matches Movidius VPU, so I disagrees with you on that.
But intel have stated its not based on Movidius?
With the arrival of Foveros will we see more use of EMIB? Or was EMIB more of a transitionary technology?
Aren't we supposed to call all this Meteor Lake madness something else like Ultra Groovy or Ultra Smoothy CPUs? I can't keep up with the hyped up, sensationalized ultra hip lingo.
Why don’t you review the meteor lake processor that are released now?
Initially the GPU Tile was to be manufactured on TSMC 3N.
Why announce the release of Thunderbolt 5 before release of Meteor Lake which uses Thunderbolt 4?
It is a terrible idea to have Thunderbolt 4 integrated on Meteor Lake but have to use a discrete Thunderbolt 5 controller.
Intel has said that Thunderbolt 4 will continue to be the Norm for a while. Thunderbolt 5 will exist in a premium class for the next few years. Plus if they changed their minds, they'd just need to only redesign the IOE tile and they can have Thunderbolt 5. That's the benefit of a disaggregated architecture.
Because Thunderbolt development does not follow cpu's. It is developed seperately. The largest integrator of thunderbolt is Apple.
@@christopherfortineux6937 it is simply not true anymore that Apple is the biggest integrator. Since Thunderbolt 3 Intel took the lead and began shipping Thunderbolt 3 with Skylake in December 2015. Apple's first Thunderbolt 3 laptop came in October 2016.
I agree the development of the CPU and Thunderbolt is seperate but they criss-cross at some point. The Thunderbolt team has to decide which CPU they want to integrate Thunderbolt.
When Thunderbolt 3 came out it was understood that the bandwidth was still not enough because of eGPUs. Thunderbolt 3 came out with Alpine Ridge and had a mid refresh with Titan Ridge. When USB4 came out Intel simply rebranded the full spec of TB3 as Thunderbolt 4. There was no doubling of bandwidth. We expect the bandwidth to double with each generation because it is based on PCIe which doubles with every generation.
TB 3 came out in 2015/2016 and was based on PCIe 3.0 which came out in 2010.
PCIe moved to PCIe 4.0 in 2017 and PCIe 5.0 in 2019 and PCIe 6.0 in 2022. Thunderbolt 5 will move to PCIe 4.0 and comes out in 2024. It has taken almost a decade to move from PCIe 3.0 to 4.0, a decade to double the bandwidth.
I have accepted Intel has failed me and I have to wait until Arrow Lake to get the best Thunderbolt 5 experience in a mobile device.
GPU tile was NEVER N3, it was from low quality leakers than claimed N3. It was always N5.
@@davidchoi9010 It wasn't from no leakers, Intel would love you to believe that. I'm pretty confident that there was an initial desire to have the GPU Tile on TSMC N3. It would have never worked because there have been significant delays with TSMC N3.
I wouldn't count on developers doing optimization for the arc tiled GPU at first. Once it starts showing up on the Steam hardware survey then its definitely on. If the optimization matches up with the rest of the Arc series, all the better.
I would count on the possibility of game engine developers to be the best starting point for this to have already started.
I dont agree. If you look at total GPU market share, not just dGPUs, Intel actually has a whopping 75%. They have more GPUs via IGPs than Nvidia and AMD have GPUs combined. In a years time there will be more people with Meteor Lake IGPs than total AMD GPU users. Developers will need to quickly optimize for Xe, because now its coming to EVERYONE who buys a new Intel CPU (Intel has 80% of the CPU market)
With Meteor Lake embeded GPU tile, could you still use standalone GPUs? Or it's more like the Apple M chips, no need AMD&Nvidia GPUs.
99.99% sure you can, it shouldn't really have any functional differences in terms of compatible besides the fact that you're making a much larger proportion of the chip kinda useless altho not on battery tbf. So yeah u should still be able to have a dgpu.
I would like to see Intel expand on their Deep Link tech which uses all these processors in parallel. Apples Silicon is very restrictive and they do not support upgrade options. Intels Deep Link tech allows you to run the iGpu on the Cpu in parallel with a discrete A770 Gpu. So will Meteor Lake and future chips support Deep Link tech?
You can use dGPUs. Ideally an Arc dGPU as you can pair them together to increase performance, but Nvidia and AMD GPUs will work with Meteor Lake too. You can also just use the new IGP for gaming, its 2X the performance of last gen, while using less power.
@@__aceofspades interesting
Where is Gordon.
Is it still lga1700 socket?
No, it's a totally new architecture.
14th Gen Desktop will be Raptor Lake Refresh, which will be on LGA1700. Meteor lake is only for Mobile. Arrow lake will be on LGA1851.
laptops don't have desktop sockets as we know them, so no
Throttling and terrible battery has always been why x86 laptops I’ve bought over the years all sucked. My M1 MacBook gets me 1-3 days without charge depending on what I’m doing. I don’t see how these Meteor Lake chips will come close to Apple Silicon.
Impressive what intel is doing. But No tb5 on MTL means I’m Skipping it.
I use graphics demos now instead of games that can be manipulated
When manufacturers say "ai" I hear "b.s."
Are there blockchain and nft tiles too? Where's the self driving tile?
But overall I like what they've done here. Excited to see how the implementation works out. Its a very complex processor architecture.
Seems like the meteor lake cores are basically a refresh of alder lake’s cores, but on a new node? So then basically, meteor lake is late, and should’ve been what we got instead of raptor lake ?
Its not a refresh at all... Meteor Lake uses brand new core architectures. Same heterogeneous idea with P-cores and E-cores (although there is a new ultra low power E-core island), but new IPC and power consumption due to new architectures.
@@__aceofspades thanks. the meteor lake redwood cove ipc increase over raptor lake’s raptor cove is what, 3%? Crestmont is what 6% more ipc vs gracemont? Intel has ticks and tocks. A tick is basically the old core on a new process and tock is a completely brand new core architecture with much bigger ipc increase. Just because a core has a new name doesn’t mean it is a tock. Alder lake’s golden cove was a tock vs rocket lake/comet lake.
But with meteor lake, redwood cove apparently isn’t much faster in terms of ipc vs golden cove/raptor cove. So the question is why? Is redwood cove essentially golden/raptor cove but on intel 4? Or in other words a refresh on a new node?
And then look at the timing. Meteor lake is launching 2 years after alder lake. Why release a product that is a tick now, after raptor lake was itself a tick, and it already refreshed alder lake ? I think raptor lake wasn’t initially on the roadmap but was released to compete with zen 4, given that meteor lake was late.
I guess arrow lake is the tock.
@@__aceofspades The P-cores in Meteor Lake don’t use a new architecture, they’re the same Raptor Cove cores of 13th gen but with more cache. The E-cores are the ones with an improved architecture and an IPC uplift 4-6% compared to Gracemont
yes, originally the plan after Alder Lake (2021) was to have meteor lake in 2022. That's why it's the same design as raptor cove but ported to Intel 4, since they later designed raptor cove on Intel 7 because meteor lake was delayed.
Is that Coca Cola?
Shocking that you’re drinking all that sugar!
Sponsored by Coca-Cola /s
Gordan is quite quitting
เซเว่นเซเว่น จะนำเสนอคุณสมบัติแบบล้ำๆเหมือนเล่นมายากล เราจะเปิดมันทีละนิดทีละนิด
Something is wrong with the color management or grading in this upload. the shadows are really crushed if you look at Ians shirt.
I am close to building a workstation (but also play csgo competitively etc). So it will likely be i9. Intel didn't talk about desktop CPUs at all. I am also a researcher in computational linguistics... so I got a lot of language model inference to do for my research. And Intel had nothing for me. I am not building a home server with Xeons. And I want a accelerator card (can't buy Gaudi2). there is no large NPU PCIe card. You can't even buy GPU Max PCIe for workstation (because sleep state power?).
So.... Do I just buy RTX 6000 ada instead? Intel has nothing for me. They tell me to use IDC.
mb
I'm sorry but without Gordan your videos/content lack that rebel pizzaz he brought to the table, they now bore me and I have stopped watching, Gordan was PCWorld to myself and millions others
This dude Ian needs to get off his high horse, If I were mark I would also look to correct him in everything he says.
I will be honest, I’m not a fan of Ian and will not watch any video he is featured on, I’m sorry but he truly comes off as arrogant and smug. Please find someone else to feature PCWorld.
No, you just aren't willing to put in the time to get to know a figure who is not instantly charismatic. Ian was regularly amused by Intel's framing of things and instances where Mark boiled it down in a way that would make Intel annoyed (though Mark was doing the right thing in doing so). If you care about technology and not just personalities in technology, I'd recommend you try again. Your favorite reviewers were reading Ian's work to cross-reference when he was at Anandtech.
Straight shooter who knows what he's talking about and isn't afraid to call a spade a spade. Rare to find people that can keep up with him in the tech TH-camr world tbh.