New Surface Pro & Surface Laptop | Copilot+ PCs for business

Mechanics Team
13 min readSep 7, 2024

--

Redefine personal and business computing by combining powerful AI capabilities with cutting-edge hardware using Copilot+ PCs. Whether you’re developing apps that harness the full potential of the NPU or taking advantage of AI experiences like Windows Studio Effects, Copilot+ PCs offer a platform that is both robust and efficient.

Vivek Pradeep, Vice President and Distinguished Engineer at Microsoft, joins Jeremy Chapman to share how Copilot+ PCs provide the tools you need to innovate with confidence. Trust in the highest level of device security, thanks to features like Windows Hello biometric authentication and the Microsoft Pluton security processor. With new hardware options like the OLED touchscreen and Flex Keyboard, you can customize your workspace to match your workflow, all while enjoying enhanced battery life and performance.

Redefine personal and business computing.

Build next-gen AI apps for Copilot+ PC devices using APIs powered by on-device models. See it here.

Seamless, intuitive interactions leveraging on-device AI.

Check out personalized experiences with Copilot+ PCs. Click to watch.

Generate content in real-time with local processing on the NPU.

Create digital art using AI-enhanced Cocreator in Paint. Take a look at Surface Copilot+ PCs.

Watch our video here:

QUICK LINKS:

00:00 — Build next-gen AI apps for Copilot+ PC
00:44 — Personalized experience with Copilot+ PCs
02:28 — Windows Studio Effects
04:47 — AI in Paint using Cocreator
06:37 — Local AI translates speech
07:30 — Lower power consumption
08:14 — Update and deploy models
09:43 — Build apps with local AI on Copilot+ PC
11:31 — Updates to Surface Copilot+ PC devices
12:56 — Secured-core PC
13:42 — What to expect next
14:33 — Wrap up

Link References

Check out Surface updates at https://surface.com/business

Unfamiliar with Microsoft Mechanics?

As Microsoft’s official video series for IT, you can watch and share valuable content and demos of current and upcoming tech from the people who build it at Microsoft.

Keep getting this insider knowledge, join us on social:

Video Transcript:

- Copilot+ PCs are the next major shift in personal and business computing. And today we’re going to take a deeper look at the mechanics of this new class of devices. We’ll see how they come to life with Microsoft’s own Surface Copilot+ PCs, with the brand-new Surface Pro and Surface Laptop for Business with the industry leading NPU capable of up to 45 TOPs and powered by Qualcomm Snapdragon X series. And we’ll also demonstrate how you can harness this AI and compute rich platform to build the next gen of apps for Copilot+ PC devices from a single API. And joining me today to unpack everything is Vice President and Distinguished Engineer, Vivek Pradeep, welcome.

- Thank you, Jeremy, it’s great to be on the show.

- Thanks so much for joining us today. How should we think about this new class of Copilot+ PCs?

- So, the Copilot+ PCs manifest some of the best characteristics of what makes simply a great PC, with phenomenal performance and battery life. Additionally, it’s really the convergence of advances in machine learning and client computing, representing years of progress in each area to make AI-led experiences more efficient, starting with well-trained and appropriately sized AI models to solve some really complex tasks, all the way to the right set of network architectures and highly optimized Machine Learning operators implemented in hardware to execute them locally with minimal resource utilization. For example, versatile but trillion parameter sized models like the recent GPTs require GPU clusters to run them. On Copilot+ PCs, instead, we have reduced the footprint for our language models and added purpose-built machine intelligence, to power experiences that can run locally. Now, these require specialized hardware chipsets engineered with our hardware partners like Qualcomm, for powerful and efficient local AI processing on the device itself. Here the 45 TOPs Neural Processing Unit on the NPU is capable of trillions of compute operations in under a single digit watt, and can run the most powerful reasoning agents we’ve ever had, language models, and always-on non-stop mode with sustained inferencing, while still maintaining battery life. Now, this combination is highly disruptive, transforming how you interact with your device, to now be more intent-based versus task base for a more intuitive and personalized experience.

- Right, and I know we’ve had some local ML models on these devices with things like Windows Studio Effects and AI capabilities, maybe like Auto Framing that can intuitively keep you in frame as you interact with a Team’s call, but you can also do much more than this and this takes things to another level, right?

- Right, yeah, our journey with the Copilot+ PCs began with Windows Studio Effects, offloading several computer vision and audio AI workloads to the very first generation Neuro Processing Units. And while we started with talking about language models, with 45 TOPs on the latest generation of Copilot+ PCs, we have really dialed up what even Studio Effects can do here. Windows Studio Effects are easily accessible in Quick settings like always, and work with the NPU to improve your online presence, regardless of which app is using your built-in camera or microphone. For example, Portrait light here, automatically adjusts the image quality to help you show up better in not so well-lit environments. Also participating in a video call using the three exclusive creative filters we’ve got here, gives you some really fun options. Here the animated option is selected, and now you can see the cartoon effect on our subject, while still preserving their facial features. Even the background is cartoonified. Now notice how the NPU is taking the load of running all of this compute on millions of pixels over 30 times a second that typically would have taxed a much beefier CPU or GPU. And then they have the improved portrait blur, with real-time depth estimation right off the RGB camera feed. As you can see, it does much better than the previous iterations of the same feature. And though not visible in this demo, the enhanced voice focus uses the NPU for deep echo cancellation and removes background noise while you are still on a call, even Zoom or WhatsApp. Now, of course, the automatic framing continues to keep you in an ideal position, which is super handy with some of the ultra wide field of view cameras. And if you combine eye contact with teleprompter turned on it’s pretty powerful. Eye contact is enhanced as you read content on the screen making you appear more engaged and natural with others on the call.

- And again because of the NPU offloading, Windows Studio Effects run a lot more efficiently, and it’s not slowing down your CPU processes. So how does the NPU then light up other intuitive experiences?

- Yeah, so this is to just about NPU running in the foreground, like when we invoke it, as we just demonstrated with the Windows Studio Effects, but also running in the background when you need it in the flow of your tasks. I can demonstrate this as I sketch and Paint using Cocreator. In fact, Jeremy, let’s work on this together.

- Sure, why don’t we also use the new Flex keyboard because it will stay connected using Bluetooth LE while we work on this together.

- That’s actually a great idea. That will allow more flexibility using the new Surface Pro. So, here I’ve started a basic drawing in Paint and I’ve started Cocreator. And my art skills are not the best, but that’s actually not a problem now. Let’s say I want an image for a website or presentation I’m working on. I’ll just describe what this is supposed to be. “An ultramodern house on the beach with a glass front.” And you can see it locally generate the image. And I’ll let you decide what you want the image style and creativity levels to be.

- Okay, sounds good. So, from here, I can actually personalize this a bit and give it a bit more detail. So I’m going to go ahead and move over to Style. And I can say I have options here for watercolor, oil painting, ink sketch, Anime, Pixel Art. I like ink sketch. So I’m going to go ahead and choose that. And that makes our preview look a little bit like it needs a bit more color again. So why don’t we try out watercolor, and that brought back some more color, but I’d like to see more. So I’m actually going to dial up the Creativity a tad all the way to the end. And now, that’s a lot closer to what I was looking for. So I just have to click on it, add it to the canvas. It doesn’t look too bad then for an original sketch.

- Yeah, and you know, all of that processing, the adjusting with the creativity settings, as well as editing the image in semi real-time, was all happening using the local NPU. Now, under the covers, multiple models in the Diffusion stack on the Copilot+ PC are deployed to interpret the original drawing, my text, and your selections for the style and creativity. And it was able to generate this unique image. And it goes beyond just text and images to speech as well. Jeremy, I’ve heard you speak a few languages? I do, so I lived actually in Germany and Mainland China, so I speak a bit of German and a bit of Chinese.

- So that’s great, you can speak all of those languages now in a single sentence. And the local AI Processing can interpret that and translate it into English captions in near real-time. And you can use this with audio in any app running in the system. I’ll play this video for the test we captured earlier.

- I was actually born pretty close to the Microsoft Campus in the Pacific Northwest, but…

- And as you saw, while Jeremy spoke in English, German, and Mandarin, it translated everything back to me in English captions in real-time. Again, this all running using local models on the device baked into Copilot+ PCs, so there’s no latency to roundtrip the data, it’s almost instantaneous.

- Right, this really adds to the things you can also do in the cloud with things like Azure AI services and using much lower kind of power consumption on the device. In fact, just to put this into perspective, the Surface Pro 11th Edition here has up to 14 hours of battery life for local video playback and the Surface Laptop 7th edition 15" device gets up to 22 hours of battery life with local video playback.

- So those phenomenal battery life numbers, by the way, are measured with our default AI processing turned on. So that amazing efficient Neuro Processing Unit definitely comes in handy. Now, speaking of AI workloads, we’ve built a very cool platform right in Windows so we can update and deploy models without needing full blown OS updates. There are dozens of models already on Copilot+ PCs running across on-demand and sustained experiences, and we have plenty of headroom left on the silicon for a lot more amazing experiences we are currently working on.

- And I’ve actually been using and documenting some of my experiences using my personal Copilot+ PC, which is a Surface Laptop 7th edition for video capture and for editing. In fact, here I’ve actually have Davinci Resolve open, and I’m editing a 1920 by 1080 resolution video. It’s actually featuring the Surface Laptop 7th edition as well and this app is natively compiled for Windows on ARM and it has also been written to utilize the NPU. And I want to export this 1080p video and upscale it to a 4K or 3840 by 2160 resolution, and for that, I’m going to enable something called, Super Scale. So I’ll go ahead and choose Quick Export. And this looks good, so do all the settings, so I’ll Confirm. And now I just need to give it a name. So I’m going to call it Laptop_4SK, there we go. And once I hit Save, it actually starts to render. Now I’m going to go ahead and pull up the Task Manager here again to see how much it uses the NPU and you’ll see that it hovers between 75–80%, while my CPU, that’s just running at 2–3%. And this is really an example of how ISVs are building apps so they can take advantage of local resources. But, how easy is it then, if I’m an ISV and I want to build something like this that leverages the local NPU?

- Yeah, to take advantage of the Neuro Processing Unit on this class of of CoPilot+ PCs, we’ve done a lot of work to open the platform up for developers using the Windows Copilot Runtime. Let me start by showing how you can build a simple app that takes advantage of the NPU using the Onnx Runtime Execution Provider. Here, I have Visual Studio Code on the left, and Task Manager on the right. Now, this is a simple Python file for model inferencing and you’ll see that it’s using ONNX Runtime. The model is local on the device, it’s a ResNet 50 checkpoint. And for the inferencing session you can see that I have two execution providers. These are listed in order of priority. First, the QNNExecutionProvider, where QNN refers to the NPU, I want my inferencing to preferentially target the NPU. But if I do build an app and want to deploy it where there is no NPU, and it is acceptable from a power and performance point of view to fallback, I do have the CPUExecutionProvider to enable that fall back. Now, in, this particular inferencing job, normally takes about a second to run, so I have coded this example to run 10,000 times. So, now I’ll go ahead and run it. And you can see below in the Terminal, that the model is running. If I move over to the Task Manager window, and like the DaVinci example you just showed, you’ll notice that the NPU is working at around 75–80%, and notice on the left that the CPU, Memory, Disk and GPU, are minimally utilized, as I run this compute-intensive task. The NPU in this case is taking the entire load of our model runs.

- All right, in this case, you’re at your own Resnet model. And you can of course, also utilize the NPU and CPU with models that are already on the device. Why don’t we switch gears and talk hardware now. Because we’ve actually got the latest Copilot+ PCs here from Surface right in front of us. And today really marks the availability of these devices for businesses. So can we walk through some of the hardware updates?

- Yeah, so, in fact, everything that we’ve been demonstrating today are on the Qualcomm Snapdragon X series processors. Now, I’ve got this beautiful Surface Pro in front of me, and I do really like, besides all the AI, the new Flex Keyboard option that we showed earlier, while using Cocreator. You can use it detached. It’s more rigid now with carbon fiber added to its internal structure. And this is the first Surface Pro with the option for an OLED touchscreen for an even wider range of colors and 1 million to 1 contrast ratio.

- And I’ve got the Surface Laptop here, and this is the business version of my personal Sapphire Blue Daily Driver that I’m using at home, for business though, it’s available in platinum or black. When I unboxed mine, the first thing that struck me was the screen with the much smaller bezels than the previous Surface Laptop generations. It’s actually a 13.8" of display now, up from 13.5" and with that, the external dimensions have actually shrunk a little compared to previous generations. And it’s also handy to have both the USB-A and the USB-C ports, as well as an audio jack, so you don’t have to use dongles. And I’ve been using mine primarily for video capture and editing like we saw before. And I’ve been impressed with its performance and battery life. And importantly, Copilot+ PCs are also meeting the highest bar for security in Windows too, right?

- That’s right, Copilot+ PCs are Secured core PCs. So Windows Hello Biometric Authentication, using Enhanced Sign-in Security comes standard to store Windows Hello biometric and PIN information in a secure virtual enclave outside of your running Windows environment. The Microsoft Pluton security processor is also a requirement for the Copilot+ PC spec, to deliver built-in next gen security as security technologies continue to evolve. And so, when you use a Copilot+ PC, you are using the most advanced device-based security.

- And there are really so many more hardware details with this round of Surface updates that you can check out at surface.com/business. So, Vivek, what can we expect in the area of Copilot+ PCs in the future?

- I’m really excited to talk about that. So in the past few months, we’ve been really pushing the boundaries of what is possible on this class of hardware. We talked about Phi Silica during the product launch, and so I’m excited with the possibilities for a state of the art, several billion-parameter language model, that can run almost all the time fully locally. This will alter how we build software. Just imagine what we did with the Cocreator experience. But now apply it to doing other creative tasks such as writing or coding, and how in Agile, small language model, can Co-construct with you. And of course, the quality of generative image AI on these PCs is definitely going to get better. We’ll be increasingly multimodal and able to reason across audio, image, ink and text and amplify the kinds of things you’re able to do.

- And I’m really looking forward to that. Thanks so much for joining us today, Vivek. Of course, keep watching, “Microsoft Mechanics,” for all the latest AI updates and more. Be sure to subscribe, and thanks for joining us today.

--

--

No responses yet