Lunar Lake Launch: Performance and Features Overview | Talking Tech | Intel Technology
- Hi, and welcome to Talking Tech. I'm your host Alejandro Hoyos. And today, we're gonna be talking about Intel's latest mobile processor, Lunar Lake. If you are gonna be looking for like all the different details when it comes to architecture features and all the latest changes, we actually have you covered. We have different videos that go in depth with subject matter experts like architects and principal engineers that will have all this.
Today, we're gonna come in for a little bit of a higher perspective. And for that today, we actually have Manini. Hi, Manini.
- Hello Alejandro. - How you doing today? - I am awesome. Lunar Lake's launching, yes. - Yes, we are very excited, very happy about that.
And before we start, why don't you give us a quick introduction of what do you do here at Intel? - So as a product manager, my responsibility is to make sure that what we are building is meeting what we intend to deliver to the market as an expectation. So whether it is ensuring that the specifications are delivering to the experiences, we are on schedule, how we are placed with respect to competition, so all those business-related aspects is what I deal with. And ensuring the engineering part of it is aligned to the business. - Oh, that is great. Can you give us a quick overview of what Lunar Lake is and who is it for? - Sure.
So Lunar Lake is the second generation of the AI PCs category that we started with Meteor Lake last December. So it's Intel Core Ultra processor series two. It is meant to be ultimate in thin and light performance per watt, and it is meant for anybody actually who wants the best mobile experience. So mobile go-getters, could be university students, they could be independent business owners, light gamers, creators, who wants the experience of a thin and lightweight, cool, quiet, long battery life. And of course, with AI. - With AI, and yeah, with the new NPU in it in there.
- Yeah, all three XPUs through AI, and they do it really well. - Yes, yes, we will cover those in a little bit. Quick question: there were some goals specifically set for Lunar Lake. Can you give us a kind of a recount of what were the goals that were set for that? - Sure, so when we made Lunar Lake, we, first and foremost, had performance per watt architecture in mind. We wanted to bust this myth about x86 versus ARM being more power efficient. So we wanted to clear that out once and for all.
We wanted to build systems that were cooler, quieter, with longer battery life. We wanted to give the best of graphics, the latest and greatest to our end users. So we had the new Xe2 core architecture in there. And finally, of course, the AI experience, right? So all three XPUs - CPU, GPU, as well as our new NPU delivers fantastic AI use cases that are there for today's AI usages, as well as future proofing them for what is about to come. - So there's something new that we haven't talked about that it's pretty interesting and it's all about the power delivery when it comes, the power delivery for the Lunar Lake processor. Can you touch a little bit more on that? - That's right.
That is another one of the great power efficiency features. So power delivery in Lunar Lake is controlled through four power management ICs. Now, PMICs have been around in the phone industry, but for the first time on an Intel client product, the entire power delivery is being managed through these four PMICs. And what this helps us do is do fine grain power control.
So if there are certain IPs that do not need to be switched on, we can actually control and switch off the power completely for those. - So the PMICs, you are controlling different power rails, so different power delivery rails. With those four PMICs, you can actually turn off different IPs in there and do the power savings? - Correct, and in fact, there is another great telemetry feature in those PMICs.
So they give us intelligence and telemetry data back so we can make more intelligent power efficiency decisions for the SoC. - I can see like a major theme when it comes to power savings, and there's a lot of power efficiency that it was dedicated to this product. - Yes. - One of the things that you guys did for this was including the memory inside the package. - Yes. - Which to me was pretty amazing. So how does that save energy? And can you tell us also the whole story of how this came together? - Sure, so Lunar Lake is the first Intel client platform to integrate memory on package.
What this means is when you see the processor, it actually has the memory chip on it. And this does something really amazing. So first off, memory being so close to the compute, that short trace length, it reduces the amount of current needed to drive that memory.
The short trace length also means that we are able to get the full speed. So the spec that we have in Lunar Lake is LPDDR5X, and the max speed of that is 8533 mega transfers per second. And because the memory is sitting on the package, you immediately get really good signal integrity, and we are able to get that speed across the entire stack. But coming to the power question, because it's this small form factor which is sitting close to compute, it actually reduces a large percentage of SOC power needed for driving that memory. Also, another benefit of memory on package is that this reduces the board space by up to 250 millimeters square, which means smaller compact designs, it means less complexity of motherboard design, lesser number of PCP layers, as well as cost.
So all around, it's a big benefit. - Yeah, less PCP layers, definitely less cost and smaller. It's pretty interesting 'cause I'm pretty curious to see what different types of products or laptops we're gonna see now that we have so much reduced on the motherboard.
Now, there was something pretty interesting about this whole development of memory on package, which was that you go to this idea of having memory on package to our suppliers. What was the reaction? What was this whole story behind like, "Hey, by the way, we wanna put memory in the package." What was kind of the reception? Yeah, tell us more on that.
- Yes, initially, it was like, "So you wanna do this?" And so, "Okay, is this gonna be this one off thing, or what do you have in mind?" And as Intel always does, we democratize technology to make sure that it's a wider industry benefit and not just a one-off thing for Intel. And so we made sure that this new form factor of memory coming on the package became a spec. So I'm happy to say that this form factor is in fact a JEDEC spec. And looking at the fact that we were delivering to this kind of democratization, we had very good reception eventually coming from the memory vendors, and we have excellent collaboration with top tier memory vendors who are supplying to us for this technology.
- Ah, that's pretty neat. So pretty much, they say like, "Well, first of all, Intel, please not make it a one up shoot. Second of all, can we make it like a JEDEC spec? So we'll work together and like, okay, so we made these JEDEC spec, and now, everybody can go ahead and use it." - That's right.
- That's pretty cool. I wanna make here a quick detour, because this is pretty interesting 'cause this kind of defines the memory on the package, kind of defines our SKUs. Can you tell us a little bit more how the SKU stack looks like for Lunar Lake? - Sure, so let's start with the SKU brands.
So we have three SKU brands in Lunar Lake, the Core Ultra 5, Ultra 7, and the top is the Core Ultra 9. So the Core Ultra 9 delivers not just the full features but the top frequencies and the best performance that we have to offer on Lunar Lake. And to match that with the kind of products that we'll use Core Ultra 9, it comes with the max capacity. So the max capacity of memory in Lunar Lake is 32 GB.
So that is where we have that married, right? So Core Ultra 9 comes with 32 GB, fully featured top frequency. And then when you come to Core Ultra 7, we have two memory options. Actually, both Ultra 7 and 5 have two memory options, which is 16 GB and 32 GB. Core Ultra 7 has all the features, physical features like the eight Xe2 cores, all the six NPU tiles, but there is a delta in some frequency from Ultra 9 to Ultra 7. And then once you come to Ultra 5, there is a difference in the features. So the graphics becomes seven Xe2 cores and then NPU becomes five NPU tiles.
Now, one of the things that I kind of touched upon a little bit, but I do wanna share specifically, the entire Lunar Lake SKU stack comes with all eight cores, which is four P-cores and four E-core. So we are not gonna cut any cores for this product at all. And the entire stack will give 8533 mega transfers per second speed because we know that there are a lot of memory speed dependent usages that our end users loves. And with memory on package, we are able to deliver that. So I hope it fits, yeah. - Yeah. - Like I hope everybody loves it.
- No, it looks great, and actually, I didn't know that we were gonna keep all the cores. So P-cores and E-cores are gonna be the same across the whole three different products. - No reduction. - Oh, that's amazing. And actually, you bring a great point when it comes to memory, right? 'Cause now, we have so many workloads that are dependent on memory, especially now that we're gonna have all these new workloads from machine learning and AI that are heavily based on memory. So having it right there and having it such at great speeds just provides like an advantage for all those different type of workloads. - Exactly. - Okay,
so we just finished covering like the different SKUs, but what is actually the numbering? How's the numbering gonna look for Lunar Lake? - Yeah, so it's going to be Intel Core Ultra, and then it's series two. So the first digit is going to be two. The second digit is going to denote the SKU brand. So if you're buying a Core Ultra 9, it's going to start with an eight. If you're buying Core Ultra 7, it's going to be either six or a five.
And if you buy a Core Ultra 5, it's going to be either a three or a two. Okay? - Okay. - And then the third and final digits is going to denote the size of the memory. So if it's six, that means you've got a 16 GB memory on package, and if it's an eight, then you've got a 32 GB memory on package.
So three digits. Always the first will be two, the second is going to be the core SKU brand, and the third is going to be the memory on package size denoted. And finally, there's gonna be a V for all Lunar Lake SKUs. So definitely look for the V. - (laughs) Awesome, thank you.
So Lunar Lake, it's also part of this new technology, this advanced packaging that we have. Can you tell us a little bit more how it kind of works? - Sure, so the tiles themselves, the CPU as well as the platform controller tile are through an external foundry, but it's been put together with Intel-based eye as well as our latest Foveros packaging technology, which is one of the best in the industry. - So speaking about all the different cores, so we have the P-cores and the E-cores, there has been a lot of changes done to it. Can you give us like a high level of what the quick changes that they have done to those? - Oh sure, so Lunar Lake actually has brand new cores, both P-cores and E-cores. So our P-cores are the latest line cove processors that have overhauled micro architecture. So you'll see double digit IPC growth, you will see a double digit performance improvements gen over gen.
So across the board performance per watt has been the focus. One key thing that has changed from previous generation is that we have removed hyperthreading and this helps us get a really great performance per watt per area. And this has been done also because we have such great E-cores, which I will talk about next, that actually, we get really good performance and we don't need hyperthreading anymore. And when it comes to E-cores, they are rockstar E-cores right? You will see the numbers, but like if you compare them to Meteor Lake LPE cores, it is industry shattering numbers, right? And architecturally, they are also sitting separately from the P-cores. So what happens is that when you have workloads coming in, and if you are able to execute them just on the E-cores and you can, in many cases, because they're just that good, you can completely switch off the P-cores and that means that you're getting a wonderful user experience in low power envelope with lot of power saving.
So that's the high level. - That's great. So that also kind of ties in with the Thread Director, right? 'Cause the Thread Director is the one who actually helps you figure it out where the workload, the thread that's gonna land. And that's all tied together for that power savings. - Exactly, and thanks for the segue.
So what you were alluding to, in fact, is one of the great power saving and efficiency features that we have brought in with our latest generation of Thread Director. So the default behavior of Thread Director is going to be that they will first bring in the workload to be executed on the E-cores and see if we can contain it there. And of course, if either their workload needs more power or if the user or the OEM has set, made settings where they need the P-cores to start off, that is when the Thread Director will direct it accordingly. But yeah, with the excellence performance per watt experience of the E-cores, working nicely with the new Thread Directors algorithm, we expect really good performance per watt on Lunar Lake.
- So if you guys want more details on that, we actually have a couple of videos that's go into that. There's fully detail in there, so please go look for them if you are curious about that. Also, okay, so we have all these different improvements. But at the end of the day, why will my grandma go ahead and buy Lunar Lake besides the point of his grandson being worked on the project? - Yeah, actually, that is also why for the first, like my answer number one would be, of course Grandma's gonna buy that, because my grandma would buy that for the same reason. Hashtag, love your grandma. But why we would recommend Lunar Lake to our grandmas is because, well, it is a beautiful thin and light system that the battery's gonna last her for the entire day so she doesn't need to carry the charger along with her everywhere.
It's going to be cool and quiet and very light, so it doesn't hurt her wrists. When she puts it anywhere, it doesn't get hot. And then for usages, it's gonna be excellent in performance. So anything that she wants to do online, like do a video call with you, it's gonna be awesome experience for her. - Yeah, I owe her a couple of video calls, so shes definitely gonna be happy about that. - We're gonna do that right on there. - Yeah, we'll definitely
do it together. (both laughing) Okay, so for Lunar Lake, how about we cover all the different key features that it brings to the public? So we have gone through a few, but there's others that we haven't covered. So how about we just go through the whole list of what it brings to the table? - Sure, so some of the key highlights I think we covered, first of all, starting with memory on package very visible, you see the two chips, right? I think we covered a little bit on the P-cores and the E-cores. Both of them are new.
The E-cores are actually sitting separately. We have new memory side cache, eight MBs that is shared across IPs. We have a new generation of Xe2 graphics engine that is fantastic, very big graphics with a great performance. We have the new generation of NPU, very, very large IP. So giving you 48 tops. And then a whole bunch of IO improvements, right? So starting with integrated wifi seven and Bluetooth 5.4.
Many usages, again, architecture depth is there, available. We have PCIe Gen 5 support. There is eDP 1.5 support. And also, some new display and media features that, like VVC or H.266, hardware decode is now available. I think that's, um, and we have Thunderbolt 4, three ports capability.
So yeah. - Yeah. There's a lot of new features in coming in there. Also, we have like new features when it comes to the security engines in there. - Thank you, so there is also a brand new partner security engine that enables third party security usages to be enabled on Lunar Lake, like Microsoft Pluton. - Yes, and another thing that I was pretty excited too, so the Xe2 cores and the whole new architecture on the graphics also comes with the XMX library extensions for- - AI engine. - [Alejandro] Yes, for the AI engine.
- That's right, so within the new graphics IP, we have a brand new XE matrix extension AI systolic array AI engine that gives a massive 67 tops. We also have a new XE super scaler kernel. We have eight enhanced ray-tracing units.
So massive boost in the experience and performance of graphics. - Oh, that's great, yeah. No, these are all different features that we have there that provide an amazing an experience at a platform level for everybody.
- That's right. - So a couple days back, we were talking, you and I, and Lunar Lake is actually something very special for you. Why is Lunar Lake kind of special for you? - Well, this is my first product that I delivered, and I was handed this with a lesser scope when I started. It was something that I was supposed to learn the ropes with. And then once the AI revolution kicked off, it became front and center and just floodlights on it, right? And we had to make sure everything new that was happening in the industry, in the competition, usages that users were starting to pick up very quickly, those were getting incorporated in the product. And I'm very pleased to say that the engineering fired on all cylinders, and we have an amazing product which will give you the usages of today as well as future proof you to some extent of what is about to come.
- So you were telling me earlier off cameras that there was this pretty interesting thing that you guys see happening when it came to the whole schedule and the execution for Lunar Lake. So tell us a bit more. - Sure, so in any project execution, as you know, we have milestones. And usually, we have them defined in time windows, right? So we keep some buffer and we have the left edge and we have the right edge by which we want to deliver. And when we started Lunar Lake initially when the scope was lesser, we met initially the left edge, and we were like, "Nice."
Like we seem to be executing fine. And then we had all of these scope changes and asks from the industry to improve the Lunar Lake. And so we thought, "Well, maybe we are gonna shift a little bit." But engineering was like so good in execution that even with those changes, they delivered on the left edge of the window, and that sort of fired everybody up. It became a positive feedback loop.
And so it was amazing to see that throughout the program, we pretty much, all milestones, we hit the left edge. And effectively, if you add all of those buffers that we did not need, we actually delivered the product four months earlier than where we anticipated was absolute late that we could go forward with it. So it's truly a marvel. When engineering decides to fire on all engines, magic happens. - So we have talked a lot about Lunar Lake, all the different things.
What are you most proud about when it comes to Lunar Lake? - I think, of course, Intel makes all kickass products, but with Lunar Lake, we have really knocked it out of the park, and we have completely busted the myth of dependency on instruction set architecture. x86 versus ARM discussion, it's done. Performance per watt is the hero for Lunar Lake, and that is what I'm super duper proud of. It's going to be one of the best things. It's a historical leap for Intel that we have been able to achieve. I'm looking forward for it to come to the market.
- Yeah, that's great. Do we know, have an idea, more or less, when it'll be we ready for the market? - Sure, so just after launch, the pre-orders will start. So look out for those. And starting end of September, early October, you will start seeing them on shelf. So yeah, it's gonna be there for holidays. - Awesome, awesome, so like, if you're looking for a great present for someone, so holidays, keep a lookout for Lunar Lake.
- Yeah, give yourself some love, give your loved ones even more love, and go buy a Lunar Lake. - Alright, thank you, Manini, appreciate it. Thank you so much. - Thank you very much, Alejandro. (upbeat music) (light music)
2024-09-09 17:56
