Hello and welcome to Espeaks, the podcast where enterprise tech gets real. I'm your host, Corey Nolles, and uh today we're going to dig into a shift in workplace computing, AI ready PCs built with Snapdragon X series processors. Joining me today is Craig Talian, director of field application and customer engineering for enterprise compute at Qualcomm Snapdragon. How are you, Craig? >> I'm doing well. Thanks for asking, Corey. [snorts] >> Good. So, today what we're going to do here is we're going to kind of unpack how this new architecture can boost productivity slashit costs and even trim your carbon footprint in the process.
So, let's dive in here. So Craig, why does Snapdragon believe 2025 is the tipping point year for AI ready PCs? >> Yeah. No. Uh great question Corey. So there's a couple of kind of forces at nature right now. Um you have Windows uh 10 is going end of life uh for uh commercial customers uh in October. So uh pretty much if you're not already on Windows 11, you need to make sure you have a plan to get you on to Windows 11 by the end of the year. So, uh, naturally, there could be scenarios where customers will refresh their PCs.
What we're likely expecting is customers will, um, not just do an in place upgrade, but they'll physically buy new hardware. Yeah. >> So, there's an opportunistic, there's an opportunity right there where, um, if you're looking to futureproof, uh, your environment and, um, you're going to lean in and, and really consider a AI capable uh, PC. Um and then you also um you know you have a lot of hardware that was consumed during the pandemic and um you know you're looking at now that hardware being four or five years old.
So that's another critical moment for PC refreshes as well >> which in an AI world may as well be a decade. >> Yeah, absolutely. >> That's interesting. You know, you hear a lot about them and they're they're all over the place. So, Snapdragon X series processors run on what you're calling Windows on Snapdragon. Uh, and it's the first, if I'm correct, non x86 on Windows. Um, so what's fundamentally different about the Orion CPU and NPU design compared with the traditional x86 systems that most enterprises run on today? >> Yeah.
No. Uh, so we we are definitely a a a unique architecture. Um we're a riskbased architecture. Uh it's a reduced instruction set. Uh it's um our Snapdragon uh silicon is um what is prevalent in uh most of your high-end premium Android mobile handsets today. And when you have to build uh around a low power island where you have um not you know uh you have limited real estate for thermals, limited real estate for uh power um and just in general you want to keep the devices thin and light.
Um you need to kind of rethink uh how you architect uh the processing power and and so what we've been able to achieve in the mobile handset space uh we're now bringing to uh to the PC. And when you compare that to x86, uh, you know, when Windows PCs first came out, they were desktop computers. They were connected to the wall. There was no such thing as running on battery, no such thing, um, as a laptop or a clamshell. So the ability to um take what was a traditional PC experience and make that be more mobile.
Uh that's where the x86 architecture is really starting to you know run into some some limitations or obstacles that we naturally um being that we come from the mobile space um are better equipped to handle and and so um part of that infrastructure part of that design is where we we basically incorporate dedicated engines for different types of compute processing. Uh so your CPU is obviously the the brains of the operation that does most of your system. um processing the applications and so forth.
But then you have a terminal as a GPU, graphics processing unit. Uh and that is something people are very familiar with. But then the third engine here is an MPU, a neural processing unit. And that's where all of those um AI workloads are handled. And um by decoupling those three engines, we're able to only uh consume the power when we need it for that specific engine. And it's different than x86 architecture where you pretty much throw everything at it.
And when you throw everything at it, the challenge is is uh you're going to have to throw a lot more at it to cool it. And then obviously once you unplug from the wall, you're throwing a lot more battery power at it, which leads to challenges uh longer term. >> That makes a lot of sense. I hadn't considered that. So what what kind of AI tasks run locally on the NPU today? Why is that why is that valuable from like a privacy standpoint or latency cost? >> Yeah.
No. So, so um I would say a lot of the the the first wave of AI experiences uh from Windows um that Microsoft delivered was it was anchored on the video and the audio stuff. And um when you when you kind of go back to the pandemic period, everybody was working remote. So, it was natural to turn on your video, right? turn on the camera and the challenge was um you didn't have a way to efficiently process um things like background blur and uh autoframing or um noise suppression or echo cancellation.
What ended up happening was is the Windows and the applications were having to push the full system and and GPU resources and that led to devices overheating uh devices slowing down from performance standpoint and so forth. So Microsoft sort of took a step back and said, "How can I rearchitect this capability so that I offload some of those tasks to a a system that's not the GPU or the CPU?" And that's where the neuroprocessing unit, the MPU, comes into play.
The MPU is able to handle a lot of the inference models. When I say inference models, it's basically knowing, you know, knowing when to blur my background and knowing that I'm an actual human and when I move my hands, it knows what to blur and what not to blur. knowing when to suppress um the noise of a dog barking in the background, right? Those things are in essence being handled within a local language model, a small language model that's running on the device, but the key is is that that processing doesn't have to go through the CPU or the GPU.
It can get offloaded to the MPU. And so for us with Snapdragon, it's terrific because regardless of what size number of cores of your CPU you have from Snapdragon, you have the same NPU uh performance of 45 tops which can handle those seamless tasks like background blur, noise suppression, echo cancellation and so forth. So that's table stakes that typically >> is you know 10 tops or 15 tops uh of required processing power in the MPU. Where it gets more interesting is as you start to introduce more complex inference models that can run locally, that's where you start to exhaust what is capable um of a of sort of an entry-level AIPC.
And um you're going to start to see um more experiences from ISVS, from Microsoft get introduced that um I can push the envelope of my MPU without having to compromise my CPU or my GPU performance. Um so yeah so Corey if I can embellish a great example would be um a security application that is running in the background like a security agent that's trying to detect fraud among the user trying to detect where data may get compromised that needs to run constantly it never turns off >> well if it's always running constantly and it never turns off think of the the strain that's going to put on my CPU >> it's going to tax my CPU it's going to make everything slow down things are going to get hot.
That type of a task can now get offloaded to the NPU which only requires drips of power and it doesn't impact any over other overhead that the CPU needs for the key productivity applications that you're used to running. >> Oh wow. So your GPU stays your GPU and CPU stay free to do what they what they need to be working on what they're good at. >> Absolutely. Yeah. Absolutely. And that's where a lot of the innovation is going to come is it's going to come from these very uh surgical specific experiences from not just you know firstparty Microsoft like you know like within the office suite or teams or zoom for example but it may come from a lot of these security ISVS that they typically have to always be running low-level processing >> consistently they can never turn off. >> Yeah.
Oh wow. So, Snapdragon says that your X series processors can hit the same CPU scores as Intel Core Ultra while drawing 70% less power. What does that translate to in real OPEX and refresh cycles? >> Yeah. So, um so, so first of all, you know, you think about again where we come from. When you have a a smartphone and you're doing tasks on your smartphone, it doesn't run any faster when it's plugged into the wall versus when you pull it out of your pocket, right?
It runs the same. Performance is uncompromised whether you're plugged in or unplugged. >> And so for us, that's what we've been able to achieve with Windows on Snapdragon, which is >> wow, >> an incredible feat. But then what happens is is when you look at the traditional x86 architecture, they're able to achieve, you know, fairly comparable with the current gen Lunar Lake and AMD uh stricks, they're able to achieve close to that same performance curve when you're plugged in, but the second they unplug, what ends up happening is is they need to slow down in order to conserve battery life.
So the performance threshold drops down dramatically. In some cases, it's 50% or more >> because it's trying to manage your performance. >> Right. Right. And and there's always going to be a compromise with x86. If you want higher performance, it's going to come at the detriment of battery life. Yeah. Um if you want longer battery life, it's going to come at the the detriment of high performance. We have a zero compromise architecture, which is phenomenal.
But more importantly, we're we're pulling the when we pull power, we're pulling it at literally half the amount of wattage of most traditional um x86 architectures out there. So, um, when I do my electricity bill at the end of the year, if I am an ITDM, um, if I'm pulling 10 watts of power versus maybe a x86 PC that's averaging 30 40 watts, I am advertise that over a year over 5 years, my build a K Edison or a local electricity company is going to be a lot higher in my x86 architecture.
So, that's one. The other thing is when you think of running on battery, I'm draining my battery less frequently. It takes me longer to go through the the battery cycle, >> which means the cycle count is occurring. Um, I'm doing fewer cycle counts over the span of a year, 5 years. So, I now stretch the lifespan of my battery because most batteries have a known a known life expectancy defined as a cycle count. >> Yeah. And so it will take me twice as long or three times as long to hit that cycle count threshold where the battery begins to degrade versus a traditional x86 is going to hit that cycle count probably in the first year or two where you traditionally have to replace a battery. >> So in addition to longer battery life I mean in addition to longer charge you're also looking at this extension of life of battery. >> Absolutely.
Yeah. So, it's a great carbon footprint uh you know, eco green story because being so more so much more power efficient, battery efficient, I'm stretching the lifespan of the battery itself much longer. >> Oh, wow. Yeah, that's interesting. I had never considered that, but that makes a lot of sense. That's some really interesting stuff you're doing. Can you share any lessons from Snapdragon's own roll out of a 100,000 of these Windows with 11 PCs and the the BDS connected solutions field pilot? >> Yeah.
So um so obviously uh the architecture is different, right? Um it's still Windows, but um the OS kernel is not the same OS kernel that an Intel or an AMD uses because they're an x86 architecture. Ours is a a riskbased architecture. And so um part of that requires that an IT environment they need to understand how do I manage and support a different a different Windows architecture because what happens is is there are going to be certain applications that will need to be specific to uh the Snapdragon environment that won't necessarily be compatible with the x86.
And those are typically your security level applications. Things like um encryption, uh data loss prevention, endpoint protection, VPN, those scenarios need to basically be identified and addressed. So typically, um you won't see a customer go completely, you know, full in on Snapdragon. It's a slow burn where they want to be able to get comfortable that as they introduce this new architecture into their environment, they can validate and confirm that they have all the proper, you know, processes in place to manage and deploy and support any driver updates, application updates, and so forth.
So, how are Microsoft Copilot, Dell, Lenovo, and other OEMs shaping this next wave of Snap Snapdragon powered enterprise devices? >> Yeah. Um so so what we've seen is um with the introduction of Snapdragon X series uh late middle of last year um Dell, HP, Lenovo, Microsoft Surface, Samsung, Asus, Acer, they've all started to um enhance their portfolio along their x86 offerings. And in in most cases, you're looking at multiple uh device options, form factors available um from that OEM.
And so we're at a point now where when I look at mainstream PCs, your traditional clam shell environment, um, anywhere from the 12 or 13inch size screen up to maybe a 15 in, >> um, we have at least, you know, one or two offerings running Snapdragon from each of those set OEMs. Um, and that's that's going to continue to grow over time as we introduce next gen of silicon and so forth. And it's really it's a way to um offer an alternative to um what has been a um a fairly successful um momentum that Apple's built with their own, you know, M1, M2, M3, M4 architecture where when when end users need a device with high performance, great battery life that's thin and light, um some of them have actually turned to Apple.
So, uh they would prefer to stay on Windows if they can. um but they haven't been able to find a viable alternative um um from x86 and and that's where we come into play and um for that user profile for that scenario where uh the individual um isn't sitting at their desk 24/7 with unlimited power where they care about thin and light, great thermals, a great battery life, great performance, um we become a a great option for that enterprise to consider.
I can see that. That makes a lot of sense. So, so speaking of Apple, u where does Snapdragon bring that unique value versus say Apple Silicon or the latest x86 pieces without getting into a full, you know, spec sheet war. >> Yeah. Yeah. And you know, you're going to always have um your community of folks that love Apple and that um are fully uh committed to the ecosystem across, you know, iPhone, iPad, and and Mac. Um what what we've seen with the majority of enterprises that we speak to is >> if if they've started to embrace Apple in their environment, it's >> it's begrudgingly because they couldn't find an alternative from from you know Windows but the preference for it is to stay on Windows because it provides a much uh more streamlined uh known way to manage and deploy uh their you know there typically is going to be a trade-off when you have to introduce Apple into your environment. you may have to use a separate tool like JF.
Uh you may have to limit the uh the security tool set that is available. Um so you know Apple definitely has a great story with regard to performance and what they promise. The challenge is is Mac OS is a completely different environment than than Windows and most enterprises want to be able to stay within the Windows ecosystem and and that is obviously something Apple can't provide. um that that we can with Snapdragon. >> Um so yeah, it's been, you know, I would say it's interesting as um both Intel and AMD have uh started to enter the the fray with their own copilot plus PCs.
And just to clarify, right, Microsoft defines a C-pilot Plus PC >> as an AI PC with an NPU that has tops of at least 40 uh or higher. Um, okay. There are other PCs that are called AIP PCs, but they're not delivering 40 tops. They're delivering maybe 10 or 15. >> And in order to deliver the enhanced set of functionality that we spoke about earlier with these language models, you need an MPU with capable with at least 40 tops or more. So, um, so I would just say like for for Snapdragon, we're we're able to deliver across the entire Snapdragon X series line 45 tops of MPU performance, whether you're in an 8 core, 10 core, or 12 core version of our silicon.
In the case of of x86, their AI PCs, the mileage is going to vary based on whether they're running Lunar Lake or whether they're running a prior gen. And then you end up with the with the challenges we spoke about which is performance cores, efficiency cores, trading off battery for performance. There's always going to be a compromise there that you got to deal with. >> It is and and this is a thing I think it's important for people to kind of learn as we go forward because you know AIPCs are still a relatively new space and uh really seeing a lot of new stuff coming out and this is this is a big deal. >> Yeah, absolutely.
So, if I'm an IT director planning, you know, a major refresh in 2026, what's the single pilot project you'd start tomorrow to validate Snapdragon in production? >> Yeah. like so I'm gonna I'm gonna give a a little bit longer answer here because I want to make sure um the the first thing is is I want IT uh directors to start to identify what is the company's overall strategy around AI because what we're finding is is that the I the the IT client compute team isn't necessarily talking to the AI developer team and what I mean by that is is that the AI developer team is working on stuff in the cloud and they're thinking about everything just goes into the cloud.
Um, and I have unlimited AI resources that I can leverage in an always connected environment. And the reality is is that's not sustainable. That's going to lead to challenges around, you know, power and energy consumption for the hyperscalers that are delivering that AI workload. But then it's also going to add to significant cost for um for that company that's using it. And then there's also concerns around do I want to put everything into the cloud? um for security privacy reasons there could be issues around latency where the network may not be as fast.
So I would say every IT director on on the client compute side needs to have a plan around how am I going to embrace AI at the edge? How am I going to run AI experiences on my PC? And if you can start to put together that plan, then that will impact the decisions you make today. Because even if you don't have a solution to run locally on the edge today, you will in the next 2 years or 3 years. >> And so you need to make sure you're futurep proofing your PC uh acquisitions today so that they have enough capability on the MPU side to handle those AI workloads if and when you start to deploy them locally on device.
You know, if I could just embellish Cory for a second, like we have, you know, some customers that are already on that journey with us where they're looking at um a concept known as RAG, retrieval, augmented generation, and they're basically saying, >> I can't allow chat GPT or C-pilot to be my single source of truth. I want my own company data to be the single source of truth. And so the RAD concept basically lets me say, okay, here is a PDF document or here is a set of of unstructured data that is only proprietary to me as the company and I'm going to point my language model to that data.
Well, if I could have that data all run locally on my PC and process it and now my employee can do all those workloads whether or not they have a good internet connection um and they can all do it um in the safety of that encrypted PC that your company's managing and you don't [snorts] need to be paying out money to the cloud hyperscalers for all those workloads and so forth. So um to bring this all back um the single pilot project uh I would say is what is your you know strategy around AI edge compute on device and make sure that um not only you identify what that strategy is but you're futurep proofing your PC investment today with you know AI capable PCs with at least 40 tops. >> I agree.
I think uh we're at a point where that time is here and it's smarter than ever right now to be >> thinking a couple of years ahead >> and looking at what you have access to today and and assuming that will continue to grow and that your need for it will as well and and planning accordingly and I think this is a great way to do that. >> Yep. >> Well, that wraps us up for today. Craig, thanks so much for joining us. It's been fascinating. Yeah. No, my pleasure, Corey.
Uh, love to do this again. So, >> absolutely, we we sure will. Well, thanks to Craig and Snapdragon for the insights and to you, our listeners, uh, IT leaders across the US and Europe even to for tuning in today. Uh, if today's episode sparked any ideas for your next device refresh, subscribe to Espeaks wherever you get your podcasts, drop us a rating. Till next time, stay curious and stay ahead.
This transcript was generated automatically from the
video's captions and may contain errors.
