Ask Paul: January 10 (Premium)

Happy Friday! This was a big week to start what promises to be a big year, and I’m again exhausted. So let’s kick back, and kick this off.

? Oh, Snap

gg1 asks:

Following the new PC announcements from CES, I saw much fewer new Snapdragon X PCs than I expected. Most were Intel with a palpable increase in AMD as an option. First off, is my perception correct here? And if so, what do you believe may be impacting adoption of Snapdragon X as an option in more devices?

I wouldn’t read anything into that. Aside from the fact that PC makers will offer far more x86 options than Arm for the foreseeable future, remember that these respective chip families are all on different schedules, and AMD and Intel always have new generation chips that appear in new PCs at CES. As they did this year.

Qualcomm is on its own schedule. And it’s still new to the PC market in the sense that it never needed a broad family of PC-centric chips, it just needed one chip that actually worked. And now that it nailed that, finally, it’s still ramping up from a first-gen product with a clear aim (mainstream thin and light laptops) for gen-two, which will focus on the GPU. So short term, Snapdragon X is in a bit of a holding pattern, and what Qualcomm can do to meet additional PC maker needs in the short term is what it did, by releasing a lower-cost version of the chip. This isn’t particularly exciting (to me), but it does need to happen.

Tied to Qualcomm’s slow expansion into other parts of the market, the big thing for Snapdragon X at CES was the first desktop PCs, which are all NUC or SFF PCs because of the chipset’s mobile focus. I assume that will be it until the second generation chips arrive, and then we’ll see Snapdragon X expand its footprint into other form factors and market segments. Right now, Qualcomm’s offerings in the PC space are incredibly limited compared to what AMD and Intel offer. It’s starting from scratch. These things take time to spin up.

AMD, to me, was the star of CES from a PC perspective. I had briefings with AMD and several PC makers in December, and the message from all of them was clear: The new Max and Max Pro variants of the new generation AMD Ryzen chips blew everyone away early on, and PC makers are embracing AMD like never before. With Intel hobbled, AMD has a real opening. But so does Qualcomm, of course. It has to scale out first.

?️ Desktop PCs

lvthunder asks:

Why do you think that neither Intel nor AMD has announced a desktop CPU that meets the requirements of being a Copilot+PC?

I think it’s just a timing issue. Hardware is difficult, especially chips, because these companies are trying to meet the needs of tomorrow, and that’s no longer as easy to predict. Thanks to AI, our understanding of which factors are most important to chip design have been upended. You can see the churn on the software and services side clearly because that’s easy to update and release, and it’s chaos. But hardware requires multi-year cycles. Big as it is/was, Intel moved quickly to make Lunar Lake happen so it could have a Copilot+ PC-compatible NPU in 2024, and that cost them billions of dollars and it loses money on every unit sold. That kind of upheaval is difficult. It may be impossible to pull off successfully.

And there’s a lot more to this. There has always been a tense dynamic between the chipmakers, the other hardware component vendors, the PC makers, and Microsoft (as the OS maker), with each having contrary goals and strategies in many cases. There are related dynamics to the PC industry itself, where most PCs sold are mobile now, but some desktop PCs are lucrative even in small numbers because of high margins (like gaming PCs). Add in that AI wrench in the works, and it’s an even bigger mess of conflicting strategies.

I’m not a hardware guy, but one conversation I had with a PC maker years ago really stuck with me, this notion that every product was a compromise, and that the product maker needs to review what’s possible and try to hit some nexus of usefulness for customers and profitability for them. (That’s my paraphrasing; he said it more like “spending money on the bill of materials on behalf of the customer.”) A PC is a product. But so, too, is a chip. And companies like Intel and AMD make different chips to serve different markets, obviously. Intel tried to make Lunar Lake happen because that was important strategically. It couldn’t miss out on a potentially big upgrade wave, and it couldn’t allow for it to be perceived to be behind AMD and Qualcomm. But that was for the mainstream part of the market. Desktop PCs are different.

We can pretty much ignore “mainstream” desktop PCs here, whatever that even means. Few people or companies are buying whatever mid-level desktop mini-tower PCs that any company makes at volume. This is about gaming PCs, workstations, and premium computers that content creators buy, and all of those have discrete graphics, another lucrative/value-add nexus. These PCs don’t have to be particularly efficient, at least compared to laptops. And for AI, this where we start to discuss “total system TOPS”, instead of just NPU, because the CPU and GPU contribute to whatever AI tasks may occur. More simply, the NPU just isn’t as important. And neither is AI, broadly, not to desktop PC buyers.

In my own naive way, I figured things would happen more quickly. Powerful NPUs and Copilot+ PCs happened last year, and so-called AI PCs the year before that, and I assumed that all processors would just have NPUs now and that they would all be powerful. But hardware cycles are long and difficult to change, and that’s not the case. When Intel announced Meteor Lake in 2023 and then, a few months later, a new generation of NPU-less CPUs, I was surprised. When Arrow Lake appeared this past week sans powerful NPUs, I was likewise surprised. But there are reasons, go figure. This is only obvious (to me) after the fact.

An Intel executive commented on this dynamic this past week when he noted that the company could have added a more powerful NPU to Arrow Lake, but doing so would have required it to remove some CPU or GPU capabilities (cores, basically), and that’s not what that market (for desktop PC chips) wants. This is literally just physics. The chip is whatever size, these things are all integrated on-die, and there’s no room to do more. But this is the desktop PC market, too, so we can go external for GPU and NPU. And right now, that’s all GPU. That’s what the market wants, it’s where the money is. It’s a small market, but even more so for AI PCs, so there’s no real pressure to make external NPUs. There’s not as much need for that efficiency, etc. These are battleships, not fast moving little things.

? Disaster strikes

Markld asks:

The fires in SoCal this week, got me thinking about something all of us are not immune from, a natural disaster! What do you do or what would do to prepare in terms of securing your tech in case of a natural disaster? It’s an open question, feel free to be as broad or narrow as you want.

In 1987, I was living with my father and his family in Albuquerque, New Mexico, and our house burned down a few days after Christmas. I lost everything that I owned, including a Commodore 64C, 1541C disk drive, an Okimate 20 printer, and more games and other floppy disks than I can remember.

(Actually, I have a full list of what I lost, thanks to the insurance company inspector, so I looked it up. There were over 120 floppy disks, almost 90 of which were games or applications. Geesh.)

The biggest loss was our dog, he sadly died in the smoke after hiding under a bed. But the firefighters who came to the house asked my father if he had any photos and, if so, where, and they ran in and rescued our photo albums. It was almost miraculous, and everyone in my family still has some photos that are singed and still smell like smoke all these years later. Incredible.

This was instructive in some ways. It wasn’t so much the things that were important. But photos are memories, and they are important. We’re all happy to still have them. Now in digitized form, complete with burns in some cases. (Most were just fine, which is part of that miracle.) Here’s one of my grandfather visiting the place he grew up in Ireland, for example.

It’s a different world today. Despite years of decluttering and downsizing, I’m still surrounded by stuff, in some sense, some tied to the hardware reviews I still write and to other work, but none of it is important. If this place went up in flames tomorrow, that would be terrible. But I’m not sure if there’s much in here I care about. Some things that would be inconvenient or sad to lose to some degree, but not much. The important things are all digital or have been digitized, like our photo and home video collections. And the key there is replication. Geographical separation and across multiple services.

This isn’t just about surviving a disaster, of course. This shift helps in the day to day too. I can and do move between multiple PCs every day, and everything I’m working on is instantly available on every PC. Nothing is tied to a single device–or, nothing important, nothing that isn’t temporary–because I sync through OneDrive and Google Drive. So if a PC literally dies, which doesn’t really happen, whatever. I just turn on another PC and it’s all there. Same with phones or whatever else. I don’t have a big physical book, CD, video, or whatever collection to lose in a fire or by theft or whatever else. It’s all in the cloud. As are my complete personal and work archives, which I just referenced to find the list of things I lost in the fire and the photos above. It’s all safe. The device doesn’t matter.

?‍♀️ Band on the runtime

TheJoeFin asks:

It is January and developers still don’t have have access to the LLM, OCR, or Image APIs via the “Windows Copilot Runtime” do you have any insight into why these APIs have taken so long to ship? I think a lot of dev expected more around the Snapdragon X Elite devices release, then maybe Ignite… now I have no clue when developers should expect these highly hyped AI APIs.

This was annoying and strange from the moment it was announced.

Which is why I titled my headline about it Windows Copilot Runtime Isn’t a Runtime, and WPF is Back, Baby (Premium), trying to capture the good and the bad of those announcements. Anyway, as you note, the constituent parts of this thing are just APIs, in this case, APIs that work against the NPU (technically the Windows Copilot Library), plus related AI frameworks and tools. Microsoft said they would arrive in June 2024, but this was at a time when Microsoft said Recall would arrive in June, too. And we all know how that went.

Anyway, I was just looking at this, coincidentally. I’d like to figure out an AI-based Rewrite feature per Notepad (see below) as part of the coming updates I hope to make with .NETpad this year. And I thought this would be one of the APIs, or whatever, in the WCR. But it’s not. Microsoft promised text summarization capabilities last May, but the Microsoft Learn portal for WCR doesn’t even mention that. It does mention a Text Recognition API (for OCR), but that’s still “not yet available,” like all these things. (I think that one will arrive in a pre-release version of the Windows App SDK.) There’s really just not much going on there, all these months later. And Microsoft oddly announced some new WCR APIs, for images, as you mention, at Ignite.

Looking through the listed APIs now, I see that literally all of them are “not yet available.” Classic. There’s no way to know for sure, but perhaps this is just moving too quickly and Microsoft doesn’t want to put out and be forced to support immediately out-of-date APIs. Maybe I’m giving them too much credit.

(Tied to this, I also looked to see what .NET offers for AI, of course. And that’s worth reading, as I assume you’ve done. It’s broader, in the sense that’s AI generally, a mix of cloud- and on-device-based AI capabilities.)

? Hot or not?

SeattleMike asks:

Microsoft is clearly going “all in” on AI. What are your thoughts on the future of CoPilot and its usefulness within Windows and the Office suite?

The overall impact will be dramatic over time. In the short term, however, it just feels like chaos. AI is evolving so quickly, it’s hard sometimes to see the bigger picture. An example might help. Let me stick to writing for obvious reasons. Really, a subset of writing, because even that is a huge area.

Looking back 50+ years, the shift from manual typewriters to electric typewriters was a big deal. The shift to word processors that could store documents for later editing and printing was an even bigger deal. And then the shift to general purpose computers. Interestingly, even my first “computer”–the ECS add-on for Intellivision–had file storage, in that case, via a Radio Shack cassette recorder. As did my first real computer, a Commodore 64, that time via a Commodore Datasette tape drive (a cassette recorder), and then a 1541 and then a C64C with a 1541C floppy drive. (Somehow my C64C has come up twice today.) Before these things, documents were on paper, hard to edit, easy to lose, etc.

Word processor applications, huge. Fonts and formatting, on-demand printing, etc. All huge. Basic spell checking, which was originally an add-on you purchased and then it was built-in (this may have been a Word innovation). And then ever-more sophisticated spell checking. And grammar checking. And then “live” versions of both with autocorrect. On and on it goes. But today, you don’t even need a word processor to get these things. These features can be built into even a simple text editor (Notepad has much of this right now), and you can add it to any place you can enter text via apps like Grammarly. I get spelling and grammar checking in any text box anywhere on my PC now. Huge.

AI builds on all this with summarization and rewriting capabilities, among other things. Microsoft is adding a Rewrite feature to Notepad in Windows that will rewrite a document or selected text, optionally make it longer or shorter, change the tone (to be formal, casual, inspirational, or humorous), or the style, and this one is nuts. In addition to things like paragraph, list, business, and others, there’s a poem style. And the poems it makes–which read to me like something out of Dr. Seuss–are always amazing.

I know that few people need poems like this, but that’s not the point. The point is that AI is advancing our ability to write in ways that are far more impressive than the already-impressive features we got previously from word processors. But they are in the same vein, if you will, the same types of advances. Just more … advanced. These AI helpers will be, and already are, in some cases, everywhere that we can write. Especially on phones, in messaging and email, and elsewhere, where it can perform the same automatic correction and rewriting, and help us be both correct and clear. This improves everyone’s lives. Everyone’s.

We’re still coping with AI and how it will impact us. And I’m as confused as anyone, for sure. I used to use the example of AI helpers in an app like PowerPoint because it’s something I don’t use a lot, and every time I do use it, it’s like I’m starting over. So anything that improves that, makes it easier or faster, is a good thing. But AI will be useful everywhere, not just in tools you’re not familiar with, or with tasks you’re not good at. And that’s why it’s profound. I can create art, and I can use Photoshop, but I use AI to create images because it’s categorically better. Meaning, faster, easier, of high quality, and at no or little cost. I don’t use AI for writing, but I could. And if you think about the Rewrite feature above, it essentially does what tech bloggers–humans–have done manually for over 25 years: You can take a press release, blog post, or whatever, paste it in, click a button to rewrite it, and then post that online as your own work. It’s already good enough that almost no one would even notice. I suspect many sites are doing that right now, in fact (using other tools).

Extrapolate that single capability out to every little nook and cranny of Windows and Office where it makes sense, and the mind boggles. Not literally text and writing tools. But rather, AI being used to make things better. It’s profound. But it’s also not one big thing, one killer app. It’s thousands of little improvements, everywhere. And it’s happening in real time. So it’s chaotic. There are doubters, and even haters. But we’re all going to come around on this, it’s inescapable. And one day soon, every one of us will have that moment when we wonder how we ever did anything without it.

This brings up a memory. In the early 1990s, I used to stay up late at night doing things on my Amiga 500. But this was in the pre-Internet era and the computer wasn’t online in a meaningful way almost ever. I know I played games. And I did some coding and writing. But once the Internet happened, I looked back on that time and was confused how I spent so much time using a computer without the Internet. Very quickly, the Internet became the thing you did with the computer, literally, or at least the thing that you used all the time as you used the computer. You can use a PC offline, of course, and we do when we have to. But no one likes it, and it feels like you’re missing a limb or whatever. AI is/will have the same effect.

? The million dollar question

train_wreck asks:

What’s your Magic 8 ball say about the future for Intel? Seems like it’s one hit after another….

It’s been clear for a few months that Intel will not emerge from this era as the same company it’s been for decades. It’s not worth recounting all the mistakes it made, or the market changes that drove this. But it is worth speculating where Intel lands. And my take on this is that Intel sells off its non-essential businesses and investments and then splits into two companies, one that designs x86 chips and one that manufactures chips for other companies. The details are unimportant, in some ways, but this could happen peacefully, if that makes sense, or via a hostile takeover if its financial situation continues to spiral. I may lack imagination, or maybe I’m misreading this. But I don’t see any other outcome.

? The other million dollar question

simont asks:

Do you think Microsoft will extend security support for Windows 10 or will the pay for patches program be the standard going forward for home users.

And JustMe adds:

Given where W10 usage is today, do you think Microsoft will stick to their plan if, come October, W10 usage is still well over 50%? Apart from normal hardware age out (which will happen much slower than Microsoft wants), what could Microsoft do to encourage W11 adoption that isn’t dark pattern related? Is it as simple as dropping the hardware requirements to a lower level?

This is one for the ages. There is precedent. And there is no precedent.

I don’t have a good read on what will happen.

In the modern era–in this case, since Microsoft created (and then pretty much abandoned) its 10-year OS support policy–the company has extended support past the 10-year mark twice, with Windows XP and Windows 7. Both were unique cases, but then so, too, is Windows 10.

Alienating consumers isn’t an issue. It’s offering a paid extended support option for individuals, which it’s never done before. And I suspect that will be the end of it.

Businesses are trickier. There are at least three factors at play here. One is the sheer volume of business PCs that are still using Windows 10, roughly two-thirds of the market. Another is a less knowable metric tied to the relative ages and capabilities of those PCs, which is increasingly problematic because PCs are more reliable and long-lived than ever. And the third is even less knowable to those of us outside of Microsoft, and that’s which customers are where in their migrations. That is, businesses are Microsoft’s most important customers, but some businesses are more important than others. A few big accounts could sway this decision. Or maybe Microsoft just accommodates them quietly.

There’s definitely a line somewhere in time, some combination of factors that might push it to make a change.

I don’t see Microsoft changing the Windows 11 hardware requirements. This felt indefensible in 2021, but it makes a lot more sense now, given the security issues of the past year and the passage of time. But it can passive aggressively just continue to let customers install and use Windows 11 on unsupported PCs, it’s done that type of thing a lot since Windows 10 arrived in 2015.

I could see Microsoft lowering or eliminating the cost of additional support for businesses generally, or for specific businesses. But I don’t believe it will just broadly extend the support of Windows 10 to businesses or all customers.

He writes, ensuring that it will announce just that as soon as today. 🙂

We’ll see. But when Microsoft finally ended support for Windows XP, businesses had embraced Windows 7. By the time it ended support for Windows 7, businesses had embraced Windows 10. Now, it’s about to end support for Windows 10, and businesses have not embraced Windows 11 for whatever reason(s). Three years from now, it will be a different world. But Windows 10 usage heading into this year is critical. It’s just not going down. I suspect Microsoft will communicate something at some point early in 2025, or perhaps throughout the year. There’s just no way to know what that will be.