As Qualcomm-powered Windows on Arm PCs begin appearing here at Computex, ushering in a generation of AI-infused Copilot+ laptops, it seemed appropriate to interview a major player in the push.
No, not Qualcomm. (We’ve already spoken to them.) Instead, I mean Arm, the semiconductor design company that licenses CPUs to companies like Qualcomm, Apple, and Samsung. Arm dominates in smartphones and tablets, and now, true PC contention finally seems possible.
I sat down with chief executive Rene Haas in Taipei, touching upon everything from NPUs, to how Arm solved its Windows app gap, to why Intel, AMD, and Qualcomm don’t matter to the success of Windows on Arm PCs. And he has nothing but praise for Apple’s M-series Macs, which he says “woke up the industry on the art of the possible” with Arm laptops. “I think Apple silicon has really proven that you could build a first-class laptop and have no compromises,” Haas said.
This interview has been slightly edited for length and clarity.
Arm chip and AI discussion
Mark Hachman, PCWorld: Since AI is the big thing now, my first question is basically an AI prompt. Please explain what the your recent CSS for Client processor means to a PC-centric audience.
Rene Haas, Arm: The way I might describe it is if you think about the chip that goes inside your PC, and we have CSS today for mobile phones — we aren’t announcing CSS for PCs. The way to think about it would be just a chip that’s inside your laptop that’s running all of the application software, the display or the GPU. Even the NPU that all gets designed by different blocks of separate pieces of intellectual property.
So what we ended up doing with a CSS is we take everything that’s around the computer, the CPU, the GPU, the NPU, and all of the mesh network, the interrupt controllers, and we put that all together as a finished block, and deliver that to the person who’s building the system on a chip, and then they are able to get that shipped to market much faster. An analogy for the PCWorld audience: if you think about IP as individual Lego blocks and compute subsystems as the Lego blocks that allow you to build a Statue of Liberty. That’s kind of what we do.
Mark Hachman / IDG
Mark Hachman / IDG
Mark Hachman / IDG
In the past Arm would simply license cores to companies like Qualcomm, Apple, Samsung and others. Is this an expansion of that?
It’s more that we would sell cores, but instead of PC where you might have eight big cores and four small cores, we would deliver that configuration. So what is the right mix and match of CPU cores to maximize performance? And then from implementing an actual system on chip we’ll take it all the way, with all the libraries you’d need for, say, three nanometer.
And with that we can literally assure you that you’re going to get, call it four-year performance. Everything is tuned. And the reason we do it is not only to save time to market, but we can almost assure that, built this way and configured this way, you’re going to get the maximum performance and power savings.
You mentioned an NPU before. But you don’t build an NPU, at least not in the CSS architecture. Did I miss something?
There isn’t an NPU today on the PC side. We have NPUs today on what I would call the entry-embedded line. But yeah, we haven’t gone public with our NPUs for the high end.
In your CSS announcement, you mentioned KleidiAI, which provides AI functions but for the CPU. Can you explain that to a PC audience that is just starting to understand what an NPU is?
Right. It’s a great question because the way that the software takes advantage of the NPU [today] is fairly high level. In other words, if the NPU has multiply-accumulates [a type of math] to run off a machine learning algorithm, the software will take advantage of that NPU. It will go off and run these complex instructions there.
What it’s not able to do is take advantage of anything unique in the NPU that might have been down to the metal layer or the hardware specific layer, because there’s no way for the application to know what’s inside that underlying hardware. That, by the way, is one of the disadvantages of everyone having their own NPU; the software ends up doing the least common denominator approach, but just making the most simple assumption about what’s there.
What KleidiAI does on the CPU is…. well, inside the CPU are very, very specific instructions that will accelerate performance. In the case of Arm V9, these are things like what we call SVE or SME, Scalable Vector Extensions, Scalable Matrix Extensions, these are things that can really, really speed up an AI algorithm.
But again, the software developer doesn’t really know whether the processor has SME. He maybe doesn’t know anything about it. So if you call that runtime library, the library is going to know — oh, this is what’s there. And I’m going to take advantage of it. So it allows for significant speedup of the performance of the software, without the developer having to know. That is really one of the probably the superpowers of those of those underlying libraries.
You may have just done this, but describe for me again what the advantage would be in the real world.
The way to think about it is an analogy — it’s not a PC analogy, but it’s a fun analogy, and it applies. When [Google] Gemini came out for Android phones, Samsung’s Galaxy had two chips underneath the hood. They had a Qualcomm chip and a Samsung Exynos chip. Both of those had NPUs, but they were a little bit different.
So from a application standpoint, Gemini was only able to run at a general level, and it wasn’t really optimized for the hardware. So fast forward, what’s going to happen with these agents, whether it’s Copilot or Gemini, they’re going to be part of the operating system. And if they’re part of the operating system, they really want to understand as much of the specifics of the low level hardware so they can be performant. But as long as these things are a little bit different, you’re not gonna be able to manage that.
So we think the library approach is not only in the right place for CPUs, but I think over time, that’s what happens with these entities.
And you’re going to have an NPU for these high-end processors. You just haven’t come to market yet.
You can extrapolate that.
That’s why when you see people benchmarking 50 TOPS versus 40 TOPS, it’s a benchmark.
And TOPS isn’t supposed to be a great benchmark or definition.
TOPS isn’t a great definition; it’s a crude benchmark of tera-operations per second. But what’s more important is, is the software able to take advantage of the hardware? And that’s the story of our libraries.
Snapdragon PCs and Windows on Arm
Arm’s CSS has a “traditional” mix of extreme cores, performance cores, and efficiency cores. But Qualcomm’s Snapdragon X Elite just ditches all that for all performance cores. That’s it. OK, so what does that mean?
Qualcomm’s doing this mix and match but what we what we see is the direction of travel is such that the complexity of software is so high.
Respectfully, we work a lot more closely with the operating system vendors and the application community than we do with folks building the chip. When it comes to the software, the decisions in terms of what Microsoft or Google is going to put in the operating systems are made years in advance. Usually before the chip. Vendors decide which core to put inside it.
So the reason we think the CSS approach is going to be right over time is that it is going to allow us to work very closely with the application ecosystem, the developer ecosystem, the operating system vendors, to really ensure that we are delivering not only the most software-optimized platform, but because the time to develop these chips keeps getting longer, the manufacturing cycle times to build them keeps getting longer, that this idea of I’m gonna selectively pick all the bits and then figure out how to mix and match, no. People run out of time. And that’s why we’ve seen the CSS approach been so, so compelling for folks, it just saves a ton of time.
This is a good discussion. Let’s stick with the software. One of the historical problems with Windows on Arm is the software: it’s run slowly, with compatibility issues. Tell me how you’re helping to solve those problems.
We had a lot of benefit from the mobile ecosystem in really, really driving native applications. So just think about all the apps you run on your mobile phone: Adobe, Spotify, browsers, they now have all been natively ported to ARM and that’s the monster benefit.
When you go back go back to the Windows platform, you think about performance, because performance is really a function of both software compatibility but also software optimizations. So there are two bins.
One is the apps are there. On Windows on Arm [in previous years] there were holes. The apps were simply not there. In this day and age, that’s table stakes. If you don’t have it, it’s a big, big deal. That has literally all gone away.
And now all the the apps are tuned. And to the question, well, how did that happen? It’s a combination of working very closely with Microsoft.
When Microsoft talks about their Prism interpreter for Windows on Arm, how does that intersect with your own efforts?
We work incredibly closely with Microsoft.
So how does this intersect with say, Qualcomm, who has a big stake in making Windows on Arm work. Do you sit down with them? Does one side take direction from the other?
I think when it when it comes to the compute platform, it’s more Arm with Microsoft, than it is arm with Qualcomm, if that makes sense. Android is a good example.
If I think about Android, we work very closely with Samsung who builds chips, Qualcomm who builds chips, MediaTek who build chips, but we’re closer to Google. And it’s not anything negative about the chip guys. But the compute platform is between Arm and the operating system, between Arm and the application developers. It’s not really the chip guys.
Is that just because a rising tide lifts all boats, and it makes more sense to work for the benefit of all of the chip vendors, and not just one?
If you go back to the first Android smartphone, the number one vendor developing the app processor was [Texas Industries] OMAP. And one of the other very large guys doing chips back then was Broadcom.
Fast forward to today, it’s MediaTek and Qualcomm. And if you go back to the handset vendors back in 2009, there was HTC and LG, Ericsson, Nokia. And you look at it and say well, Nokia is gone. Ericsson is gone. LG is gone. HTC is gone. Ti is gone. Broadcom’s gone. Yet, Arm is still dominant in Android. Why is that? Well, it’s because we create the environment. And you create the opportunity for different handset manufacturers and different chip people to enter the market.
I like this conversation and I want to continue it, but a rumor springs to mind: that there has been an exclusivity agreement in place between Microsoft and Qualcomm for Windows on Arm, and that agreement ended this year. Is this true?
Everything I’ve heard is that that is true. Okay. I’ve heard that rumor, exactly. I have also heard that times out this year. And that will allow other players to enter the market, which again, all the rumors I’ve heard, that’s true. And I think it’s going to be great because it’s going to allow for choice and it’s going to allow for diversity, which is kind of the theme of the Windows ecosystem.
How do you view the potential for Windows on Arm for Arm, versus something like, automotive?
Gosh, without getting into the [details]… for Windows on Arm, it’s a pretty significant revenue opportunity. Because 200 million units, and our market share is approximately zero.
Well, maybe not zero. But smallish.
Yeah, so I think there’s only upside for us there. And if I look at the other ecosystem for PCs, and I look at what Mac OS has done, at the silicon, it has been amazing. And the products are amazing.
Do you think Apple helped validate your approach, by making its transition from X86 to Arm?
They were, they were a great help. They were great. Apple’s a fantastic partner. And I think Apple silicon has really proven that you could build a first-class laptop and have no compromises.
We’re learning that Qualcomm has promised monthly drivers for Snapdragon X Elite PCs. That’s their commitment. Do you help out here?
Were they referring to GPU drivers?
I believe they were referring generically to monthly driver updates.
If it’s an OS driver, that’s actually Arm and Windows. So when you get that annoying. security patch — “Windows needs to update your machine” — that’s Arm and Windows, meaning Microsoft. Qualcomm is not involved in anything relative to the OS first.
From where you’re sitting, is the Windows on Arm community providing the right messaging to consumers and potential buyers?
Again, respectfully, I think the world has kind of moved on relative to Intel, AMD, Qualcomm inside and there’s probably less of a buying decision for folks any more.
I think the AI PC is good liftoff because it’s obvious with what Microsoft’s doing with Copilot and what runs locally and what runs at the cloud. And it’s obvious that AI is creating all kinds of differentiated use cases.
Let me say it this way. If there was no AI, I think it might have been a little harder for Microsoft in general to create buzz around this new category. I think AI PC gives a great kind of tailwind. And I think on top of that, that it creates the window for new machines.
I think the Windows on Arm machines are going to be when people say oh, I need it. It starts with AI PC. A new opportunity. Now let me see, what are the options I have with AI PC? Oh my gosh, these machines here look pretty good. The battery life is great. The thermals are great and mechanicals are great. I think it’s less about oh my gosh, what brand is it, Intel? AMD? No.
So let’s say Windows on Arm is a resounding success. What does that mean for future development of Arm processors? Or does CSS for Client anticipate that success?
Yes.
It’ll be good for us.
OK, final question. I know I’m probably going to get a biased answer, but did you expect what I’d characterize as a warm reception for Windows on Arm this time around, versus before?
I did. I did. I think it was time. And again, I would give thanks to the folks in in Cupertino [Apple]. I think they I think they woke up the industry on the art of the possible in terms of what can be done with an Arm-based PC.
I think a lot of things come into play relative to the right time. Microsoft making the investment. So maybe the timing is right. I mean, as I mentioned in the keynote, I was personally involved in the very first Windows on Arm PC. I was at Nvidia at the time; I was the GM [general manager] that was running that business.
I lived Surface RT. It had a kludgey version of [Microsoft] Office. It had no enterprise support whatsoever. If you were a CIO, an IT manager there was no way to do anything with it. All that’s gone.
Further reading: Surface VP sitdown: How is AI going to change Microsoft’s PCs?