CES 2025: Deep Dive on Intel Core Ultra 200H & 200HX Series Processor Performance | Intel Technology
- [Narrator] Please welcome to the stage, Robert Hallock. (dynamic music) - Hello, everybody. Hello, my friends. Now, as I look out into the crowd, I'm struck by what an absolute honor it is to be here with you. This is my 16th CES, which pales in comparison to some of you in the audience I'm sure, but that's nearly half my life spent neck deep in PowerPoint over Christmas. So, it's been an adventure.
But you know what, it's worth it. It's always worth it, because I love this industry. And if you're here, you love this industry too. You love PC hardware. And deep down it's more than a job, it's a hobby. My hobby started when I was six years old.
Thank you. When I was six years old, playing an Intel 46 DX2/66. And if you're old enough to remember that, you're as old as I am. Hear some claps. But it's been a hell of a ride, because now I'm here standing on Intel's stage, nearly 40 years old, and that's some pretty cosmic continuity.
So, let's get after it. We all remain united in our love for performance. Let's take a look. As Michelle and Jim spelled out, CES '25 is a big deal for us.
The commercial opportunity has never been bigger, and I think you'll agree that Lunar Lake is an exceptional product for business systems. But if your butt is in that seat, then odds are pretty good that you wanna learn more about those enthusiast parts, the H and the HX. And that's what I'm here to do, so let's dive in. I think it's important to start with what our full mobile stack looks like in 2025.
Lunar Lake, which came out in August, September of last year was our first salvo, but that was for a market that values portability and battery life above other metrics. But at the time, many in this audience, to me directly rightly pointed out, what about multi-threaded performance? That is the 28 to 45 watt space that Jim mentioned. And we've classically addressed that with SKUs that end in H.
The H suffix. We're holding steady with that approach. Last year we had Meteor Lake H.
This year we have Arrow Lake H. Likewise, we're also staying consistent with HX. Last year we had Raptor Lake HX, now we have Arrow Lake HX, and that serves 55 watts and up, the ultimate performance segment in the market. What I love about Intel is our ability to tape out multiple chip designs that are specially targeted for very specific market and the needs of that market.
If users want a true HX part, they can have it. We can do that. If a customer wants to update their design with a new H chip with 20, 30% more performance than last year, we can do that too. So let's start with H. Naturally you'll see these SKUs in market as the Intel Core Ultra 200H family, and you should broadly expect dependent systems to be ultra portable, with a mix of both discrete and built-in graphics designs. As I mentioned, TDPs in the range of 28 to 45 watts are pretty typical, but you'll see systems above and below, and that's just the wild west of the laptop market and what makes it great.
Arrow Lake H updates, the cores, versus last year, to what we call Skymont and Lion Cove, giving us about a 20% performance improvement in multi-threaded performance versus our previous CPU. 3D graphics, raster graphics, as I like to call it, is a touch over 20% faster at 1080p as well. So good for gaming.
I'll spend more time on graphics shortly, but the first thing to know is that this is an Alchemist based graphics engine, if you're familiar with our code names. But it has been augmented with a new XMX engine that specifically helps with AI performance. When you put those engines together with the NPU and the CPU that are on this ship, as Jim mentioned, you get 99 TOPS of performance, which is second place in the market only to Lunar Lake, which is also one of ours. Speaking of NPU, we're carrying forward Meteor Lake's NPU three architecture with a speed bump now to 13 TOPS. And we think this is about the right size for camera work, voice improvement, background agents, small language models.
And this is featured on every product, but every SKU also has Thunderbolt 4, plus 20 PCI Express Gen 5 lanes for discrete attach and NVME. So this can be a very, very robust system, even if it's thin, light, super portable. This is like a really nice middle ground between the thinness of Lunar Lake and the ultimate performance of HX.
The 200H family will come with five different SKUs, Spanning Core Ultra 5, seven, and nine. The top configs have 6 P-cores and 8 E-cores. And if you don't know what that is, that's okay. Suffice it to say, there is 14 cores or 16 cores, excuse me, overall, with 16 threads at the top. Graphics and NPU, we decided to keep those consistent across the stack.
So people get roughly the same amount of AI and graphics performance regardless of what CPU they choose. The rise in AI 365 that you see as the white line is AMD's H equivalent product. Officially it's 28 watts on amd.com, but frequently assorted in market up to 45 watt as you see here in this OEM system. Again, the white line. Now, at the same power, we're measuring very healthy performance leadership in single core performance.
Multi-thread also looks really good, bringing our 2025 product stack to leadership in this super, super important 28 to 45 watt space. And that leadership is a little or a lot depending on the workload, but a win is a win, and I will take it. Stepping outside of renderers into categories of workload, we continue to see leadership across video editing, photo editing, video mastering, and more. At a high level, the 200H is really just a solid performer in CPU compute. Now, we touched on how important this segment is to us, but I'll be even more blunt. It is the bread and butter of the laptop market.
Delivering CPU performance leadership in this segment is critical to our business, and I expect we'll have that lead throughout the year. Beyond performance though, energy efficiency looks super strong. Now, it's a bit of a dirty secret that 28 to 45 watts is really more like 25 to 80 when you look at all the designs that might come to market.
And that's the left to right on the chart, that's the power. And from top to bottom, we've plotted the relative performance of the very best CPUs from each vendor in this H segment. You can plainly see that versus our previous generation, the lower curve, we're about 20% faster at any given power consumption.
But when we can compare equals from Qualcomm or AMD, we're 20% better in performance per watt than Qualcomm, and 10% better than AMD. Now, let's take a look at the GPU. With Arrow Lake H, our goal was to give performance and capabilities that are similar to Lunar Lake, but wrapped in a package with a CPU that can scale to higher performance envelopes.
Lunar Lake goes up to about 30 watts, but now we know that Arrow Lake H can go way beyond that, and that's really important for this segment. To do that, we doubled the ray tracing capabilities with more traversal pipelines. So you can actually do ray tracing in a notebook, pretty cool. We doubled the size of the L2 cache to eight megabytes, so you get more bandwidth through the chip.
And the updated process technology pushes the GPU to a more performance spot on its voltage and frequency curve. And when you put all of those together, that's how we get the 20% performance improvement versus the previous generation, with the same architecture and the same core count. The infusion of those XMX extensions is critical for us in the segment. And if you know nothing else about XMX, it just helps process AI on the GPU much more efficiently than a GPU might otherwise do on its own. Now, the 200H can benefit from all the work we've done on the Lunar Lake designs, and the Meteor Lake designs and the Arrow Lake designs that are already in market. And to date, we are actually at very specifically 447 AI features that have been developed with the assistance of Intel through code optimization, co-marketing projects and so on.
But because of XMX, the GPU by itself has 77 TOPS of AI performance. And we expect about 40% of all the AI features released this year will actually use the GPU. Not an NPU, not a CPU, it's the GPU that will take on the plurality of the workloads. Now, for background agents and assistance, as I mentioned, NPU three comes back, that's the same architecture as last year, with 13 TOPS, and our CPU instructions give us another nine.
And that's how you get to 99 TOPS. With XMX, we continue to be the undisputed leader in AI performance. Competing hardware either does not run or runs extremely poorly. Models you see here include computer vision, object recognition, image classification.
These are critical to webcams, image editing, image upscaling, broadcast. These are use cases that people do every day, and they're now being infused with AI acceleration to make them faster and more efficient. Image generation models like Stable Diffusion also continue to play a huge role in the excitement around AI. Gen AI, generative AI, we're all talking about it. But it's a really great test for the compute performance of an AI accelerator.
It loves more performance, and will just keep going faster and faster. But there's one wrinkle. AI models like Stable Diffusion need a really good software framework to extract the appropriate performance.
TOPS is a promise, but software is how you actually deliver that. OpenVINO, which is our AI framework, is clearly up to the challenge. But the frameworks recommended by other vendors are unfortunately coming up short. Right, so when we talk about AI compatibility and when we talk about AI performance, this is the competitive landscape we're talking about, and this is the kind of leadership that Intel is driving in offline and local AI. Finally, large language models and small language models run super, super well on the 200H.
We have a nine X lower first token latency or a result from your prompt in about half a second. And it's nine times slower. You have to wait almost five seconds for the next fastest competitor. Qualcomm doesn't run at all. And our token rate, which is strictly governed by the amount of memory bandwidth in your system, we're actually out scaling our competitor.
We're getting about 5% more tokens or words, punctuations, spaces for every gigabyte per second of memory bandwidth. That efficiency, being able to extract more from the memory bandwidth you have, is a function of our software. That is what OpenVINO can do.
It makes better use of the underlying hardware. Okay, so that's a lot of AI. I think other people will be like, "All right, Rob, show me some gaming. Let's talk gaming. This is this massive bar chart."
Okay, gen over gen, the Intel Core Ultra 9 285H, our flagship part, is actually on average 22% faster. That's the raw number, 22% versus our previous generation. We also see some really disproportionate improvements in certain titles, modern titles that have newer graphics APIs. So "Cyberpunk 2077" keeps getting updated. Heck of a game, love to play it. That's 36% faster.
So there's some real highlights in this mix that go above and beyond the 22% average. And I feel really great about the performance that we're offering gen over gen. But when we compare to the competition in similar platforms, we do see a tie versus H equivalent AMD AI 365.
We win in some games, they win in some games, and on average we're on par. But we all saw a new entrant to the laptop market last year. And so versus Qualcomm's 84-100, that's their flagship part, we are significantly faster. In fact, by about 58%.
The 84-100 continues to struggle in games, and they have not had a driver update for most systems since August. But with Intel, you can download a monthly driver every month for our integrated graphics, just like you can a discrete graphics card. We treat these GPUs in these chips just like a gaming GPU. So not only is it great performance, but it's also the performance improvements and the feature improvements of new software, regular software.
With this product, you also get access to XeSS super resolution, which is our AI based upscaling technology. It's supported in over 150 games. This helps boost the performance of a game by using AI to upscale the image, restoring more performance back to you.
And now, we see really great performance with this technology on this part, which is about twice as fast when you turn XeSS on. And these numbers are actually higher than what you may have seen for Lunar Lake or Meteor Lake, and there's a few reasons for that. First, the CPU in the Core Ultra 200H is more efficient and more performant than our previous designs.
It gives more performance headroom to reach higher frames per second. And the built-in GPU makes use of those new XMX extensions to boost performance even more. So XMX is not just for AI processing and conventional workloads, we can also use it to help make games faster as well.
With Battlemage or ARC B-Series that Michelle mentioned, we've recently launched a newer version of XeSS called XeSS 2, Very straightforward. XeSS 2 combines super resolution, frame generation, and low latency optimizations to deliver not only great performance, but also a really great gaming experience. Even though 200H is not a Battlemage card, it is not a B-Series part, I'm happy to tell you that we have back ported that feature from Battlemage to this GPU.
And that's because of the XMX technology, we were able to make that feature work. Now, when you turn on XeSS 2, the performance uplift is pretty extraordinary. This is an example from "F1 2024." Great game if you're an F1 fan. XeSS 2 allows us to change the in game preset.
What this really means is that you could go from high image quality setting now to ultra, where you wouldn't be able to run ray tracing on this game before in a laptop like this, you can now. You can make the image quality better, you can turn on features that were previously inaccessible, and then you can boost performance out the other side. The combo of these settings allows us to get over 90 frames per second. Well, about 90 frames per second in "F1 24" versus just 24 frames per second natively with all the eye candy turned on. It's an intense game even for a fast laptop, so I'm thrilled to get this kind of performance.
There is another game coming, and I'm pleased to announce that "Marvel Rivals" later this week will be updated with XeSS 2 as well. And if you haven't played that game, it's a blast. You gotta check it out, it's a lot of fun. Stepping back, Arrow Lake H is a super great choice for the 28 to 45 watt segment. It has the fastest cores, fastest multi-threaded performance, fastest AI performance, the best energy efficiency, and a great GPU. On the back of those metrics, you're gonna see over a hundred designs based on the 200H.
Many will be on display here at CES from our wonderful partners here in the front. And again, from all of us at Intel, thank you for your partnership. Thank you for your support.
Now, as fun as the H segment is, HX is even more fun. All right, this is a performance segment that Intel created, with products like Alder Lake and Raptor Lake. These are hugely multi-core, high performance, they're just a lot of fun. And so later this quarter, we will keep that goodness going with the launch of the Intel Core Ultra 200 HX processor family. These are mobile first chips built for TDPs 55 watts and up. Now, the words and up does a lot of heavy lifting here, because OEMs in this market will happily put up to 150 watts into the CPU if it keeps yielding performance.
And that's what HX is designed to do. It's designed to give a desktop's worth of performance in a notebook. But we also know that total graphics power or how much power the GPU consumes is the primary driver of gaming notebook performance. So one of the key benefits of the 200HX generation is that we've improved power efficiency by about 40% versus our previous product in this space, which gives the GPU more of that energy budget to do what it does best.
Plus, of course, you can get an entire desktops worth of expandability on platform, a hilarious amount of USB ports. It's impressive. The HX dye contains a small GPU, up to four cores. And that's really just for lighting a display, for power efficiency, for desktop productivity when you're not using that big GPU that all of these systems tend to have. Like Arrow Lake H, it has the same 13 TOPS to NPU. Again, size for camera, voice improvement, super resolution, background agents.
And then you also get technologies like Wi-Fi 6E, DDR5, with both SODIMM and (C)SODIMM support. And if you don't know what that is, find me after the show. I would love to talk to you about (C)SODIMM, fun technology, really great. But we both know that CP performance is pretty much the whole reason why people buy HX. So the 285HX is our flagship.
It has eight P cores and 16 E cores, for a total of 24 threads, with clocks up to 5.5 gigahertz. Now, if you're the kind of person who would like a copy of these slides, you're gonna get it. Plus a giant wall of text with even more specs, those are coming.
I'll give you just a sec to scan the SKUs and then we'll move on. All right, with that many threads in flight, we're actually still quite efficient in this segment. I mean, all things are relative, but 50% more performance per watt than Qualcomm at the same power, 40% more than our previous generation at the same power, and 35% more than AMD's Strix Point HX.
And yes, this chart is saying that it continues to scale all the way up to 150 watts. And you know what? I'm the kind of guy who loves a CPU like that, so I'm thrilled. Now, we're gonna get to the brass tax of performance. We're about 5% faster than competitors in this segment on a single core or single thread basis.
But on a multi-core basis, the two blue bars become super, super important. Teal is the 285HX at about 50 watts, which matches the fastest and most powerful competing systems we could find and measure in literally the entire world. Now, being 30 to 60% faster, it's pretty cool, pretty cool. But that bright blue bar is from an upcoming 285HX customer chassis.
A real system blasting out 150 watts of CPU capability. And that kind of system clearly pays dividends in workstation, scientific and creative workloads. Now, I'm up to two and a quarter times faster than Strix HX and Qualcomm's best. Two and a quarter times. For completeness, here is another view of multicore, with AMD's Dragon Range chip in the mix. Where that CPU tends to top out at about a hundred watts.
That 150 watt capability allows a 285HX to keep pulling forward in performance, into a leadership position. Arrow Lake HX can scale to much higher performance envelopes than any other HX system we could buy and measure just last month. Blender, Cinebench, video encoding. Now, these are hallmark tests for a CPU like this, but what about Spec Workstation? If you've never heard of this workload, it contains 20 different workloads spanning science, finance, simulation, machine learning, programming. We're another 23% ahead of our competitors' fastest at this time, and we're excited to deliver a top tier workstation experience you can toss in a backpack. As for gaming performance, you'll be hearing from us later this quarter with data, briefings and systems.
I'm sure you know what I mean. Stay tuned, these systems are gonna be really exciting. The 200HX systems can expose on chip Wi-Fi 6E, 10 gig ethernet, which is still pretty expensive as a discreet card, 44 total PCI express lanes, and even more ridiculous quantity of USB ports. You really do get a basically a desktop in a laptop. It's super cool.
Our customers are ready to rumble with 200HX, and you should expect to see over 40 designs from all the top OEMs. You may even see some of these systems at CES, but more designs and even more vendors will be coming later this quarter with their own announcements in a few weeks. Wrapping up, I'm reminiscing about the times many of you in this audience asked me what Intel plans to do about multi-core performance in a notebook. And I'll be plain, 200HX straight up doubles the highest performance we have ever seen from a Snapdragon X Elite CPU. It blows away any other HX design we could buy and measure. And it's a full 40% more performance per watt than our previous chip, which frees up an enormous amount of power budget for those new flagship GPUs we're all waiting for.
But wait, there's more. We're also expanding Arrow Lake on the desktop with new SKUs. Now, this launched in October, and there was just a couple CPUs available at the time. But we're also expanding with new options, new chip sets, even new options all the way down to 35 watts. So let's jump into those.
Now, you're gonna get a full list, if you want it, of all the new SKUs and all the specs. But these three are my focus today. I've been asked about these a lot online, and I'm pleased to say that Core Ultra 5 and what we call non-K SKUs are launching today. Now, it's gonna take a few weeks for these processors to flush into the channel so you can buy 'em in retail, but they're on the way. And in total you will see 17 different Arrow Lake S or 200 S SKUs for the desktop PC by the end of this quarter.
So 17 total desktop processor choices available in the market. I'm also thrilled to say that new chip sets are on the way. The other big question that I get all the time. Now, the mainstream B860 is probably going to be a crowd favorite. Has four dim slots, a Thunderbolt 4 port, and 20 CPU attached PCI express lanes for a big GPU and NNVME. That's what most people need to build a great gaming system, and that's what this chip set delivers.
But we're also introducing H810, which is our entry level chip set, a ton of value for the money. Two dim slots for up to 64 gigs of RAM in the DDR5 era, 16 lanes of PCIe Gen 5 for a big GPU. a Thunderbolt 4 port still, and NVME off the chip set.
The big question, price. Every ODM will be launching a wave of new motherboards, starting at just a hundred bucks for H810 and about 130 bucks US for B860. So with these releases, we have a full chip set lineup and a full desktop CPU lineup all coming into the market this quarter.
Whew, okay, that was a lot. Let's recap. Lunar Lake for business, unlocking massive opportunity for Intel. You heard from Michelle that there's TAM growth this year.
You heard from Jim, a massive vPro upgrade. This is a core part of our business that we're supercharging with the Core Ultra 200V series. Arrow Lake H and HX complete our vision for a top to bottom stack based on our Lion Cove and Skymont cores, each with strong performance in their segment, with gaming reviews and additional data on gaming coming later this quarter when systems are available. And finally, a wave of new chip sets and new desktop CPUs. It's gonna be a great CES, it's gonna be a great 2025.
And I'm thrilled to spend another year in this awesome industry with you. I've grown up with many of you. You keep me passionate. I'm grateful, thank you for coming today. And from all of us at Intel, have fun, stay safe, and we'll see you out there.
Thanks very much. (audience applauds) (dynamic music) (bright music)
2025-01-10 22:02
