Bill Gates Discusses 64-Bit Windows, Longhorn Transitions

Microsoft chairman and chief software architect Bill Gates will announce availability of four new 64-bit versions of the Windows operating system at the company's WinHEC conference for hardware engineers, which begins April 25 in Seattle. The new Windows x64 will allow customers to run existing 32-bit applications and newer 64-bit application on the same PCs or servers, providing a bridge between Microsoft's 32-bit software environment of the past decade and its emerging 64-bit software environment for the future. On April 20, Gates sat down with InformationWeek editor in chief Stephanie Stahl and senior editor at large John Foley on Microsoft's Redmond, Wash., campus to discuss the new 64-bit product line and how it compares with the next-generation Longhorn operating system.

InformationWeek: You're about to announce availability of Windows x64. What's the business case for Windows XP and Windows Server 2003 x64? Who should get it and why?

Gates: I think the transition to 64-bit is the biggest thing happening in the computing space, and in some ways, the reason it's not getting this huge attention is because at a technical level, the way the transition's being handled, this is going to be the simplest address-space transition ever. And yet, it's the biggest address space transition ever. Going from 32 [bit] to 64 [bit] is more than we've ever done. We haven't had an address space transition since 1986. Thirty-two bit's been enough, even at the server level, 4 Gig of physical [memory] felt like enough.

For the last five or six years, people have been feeling the pressure. Not everyone, not the general purpose desktop, but a high percentage of servers and a high percentage of high-demand desktops -- financial analysis, scientific computing -- the kinds of things that were the last to flip over from Unix to Windows, the super high-end thing, where even for a time people had two machines on the desktop. They wanted Office, yet there was some application that Unix had that they didn't have. We got this nice period where at the desktop level and for most server [workloads] the standard hardware was good enough for everything. Now, the 64-bit thing, it's quite dramatic on the server because the cost of memory has come down so that putting huge amounts of physical memory on a server is economic and the demand for performance you want these Web sites to be very fast. The size of the material we have out there, whether it's the amount of text, images, click databases, everything has just expanded dramatically, so 64-bit is a huge thing and it gives us pretty unbelievable headroom.

The range of applications that benefit from 64-bit are incredible, so to go back to your question in a very direct sense, as people are rolling out new applications, as they're looking at any application they want to up the performance on, moving to 64-bit is one of the magical ways. You can simplify the application and have it be very high performance. After all, the 64-bit hardware isn't premium priced. You want to buy an Intel-based server, an AMD-based server, these things are going to be priced just the same as they were as 32-bit systems. That's pretty unbelievable. We didn't have that in the past. Many elements of the system have been sitting waiting to be enabled for 64-bit -- we've had 64-bit file system stuff for over a decade. Windows actually had that before most versions of Unix. We've had 64-bit development tools for developers for quite some time now. Recompiling your code isn't that hard. There's a little bit of a challenge to get all of the device drivers. You can mix 64-bit apps and 32-bit apps, but when you have a 64-bit operating system, you need 64-bit device drivers. During the next 12 months, there will be a lot of focus on that. I don't have any concern about it; it's just that we need to keep the message very strong that we need all those device drivers.So, whenever you buy a new server, no matter what the reason is, unless it's very lightweight use, you're going to buy a 64-bit server. In fact, you might even buy it without thinking about it because it's something like 15 months from now that Intel and AMD basically won't be shipping server chips that aren't 64-bit. Even on the business desktop, it's all going to be 64-bit. There's a few other silicon features we're getting with 64-bit -- putting in some things to help with memory protection, virtualization is on a slightly different schedule -- but it's in the same basic time frame, and those are pretty exciting things, too. But the 64-bit, this is just an amazing transition where, whether it's a high-demand desktop or almost all of the servers, they're going to have the ability to use this much memory, and it really simplifies your application. Your ability to take the disk performance out of the equation, where you've got something like SQL caching in memory or IIS or its competitors like Apache caching application stuff in memory, you're going to see really nice benchmarks.

InformationWeek: To what extent do IT departments need to plan the move to 64-bit Windows versus it's just going to happen? Do they have applications tuning to do? Infrastructure considerations? Do they need to talk to their ISVs about 64-bit versions of applications?

Gates: Wherever they want more performance, this is a huge opportunity for them. If every application they have is running every bit as fast as they want, they don't really have much to do here because you can buy 64-bit hardware and run it in 32-bit mode. We're continuing to ship the 32-bit applications, the 32-bit operating system. There will be some time, like five or six years from now, where a lot of application updates will start to be 64-bit only, but that's quite some time off, where people will take the simplification not to have both. And maybe in some niches, like chip design, some of those high-end things, those guys will be 64-bit only even in the near term, but the people who want that kind of software will know that from the ISV. So you really should look at this as opportunity, you should say, can we make our Web site faster? Nowadays, when you look at a Web site, and you compare what it takes in terms of human programming effort and all the different costs that it would take to double the speed of the software, changing the hardware to let you run twice as fast is going to be the most economic way to do that. So everybody should go through where they're running into performance or size limitations and assume that 64-bit is here. It's mature.

We've been running a ton of stuff internally at Microsoft on 64-bit. One of the most amazing stories is we're doing this new search server farm to compete with Google, and we were able to do it with dramatically less machines -- it's still 5,000 machines -- but a lot less than they use because the individual machines have a lot more memory because we assumed a 64-bit design point. They did it back when you couldn't assume that. Our finance department runs all these simulations and the 64-bit, because that now works in memory versus having to go out to disk, they're able to complete those things like ten times faster. [For] most things, the kinds of speed-ups you see are in the 1.5 to 2 [times faster range], but you get some where it's very dramatic. People have to be careful not to think everything is going to be this very dramatic thing. But it turns out most server things are memory starved, and even client-side things, some really push those limits. So this will be much faster on the server, but an impact for some workstations even in the next year.

InformationWeek: You've made the point that Windows XP x64 will be interesting to people who are hitting bottlenecks or that have high-end applications. Do you see it having something to offer the broader set of knowledge workers or do they wait for Longhorn?Gates: I don't think you'll see broad knowledge workers change their operating system from a 32-bit operating system to a 64-bit operating system in this next year, unless you're running an application that demands an extreme amount of memory. And it's hard to list all those people, because there's even some PhotoShop users who will want to beef up their memory, but for most Office-type users, the next operating system transition will be Longhorn. And when they get Longhorn, they can decide to run that in 32-bit mode or 64-bit mode. Because the 64-bit mode does such a good job running 32-bit applications, there's almost no downside if you have the hardware, the new Intel chips or AMD chips, there's no downside to running in 64-bit mode. The big numbers on the desktop will come with Longhorn, and Longhorn's timed about to the point where Intel and AMD will almost fully have made the transition. Except for some consumer machines, they will have basically made the 64-bit transition.

InformationWeek: So, if you're looking ahead at what Microsoft is going to offer in the next couple of years, for the typical office user, Longhorn is the place to be?

Gates: Obviously, we have a message today that [Windows XP Service Pack 2], in terms of the security benefits that exist there, we're really encouraging all Windows users to be up on XP SP2. But most people will just take that and run it in 32-bit mode, they won't also do the 64-bit thing. When you're really going to make the change is when you go to Longhorn, and that's where it's easy to go to 64-bit on the typical desktop. The percentage of desktops where there's a benefit for jumping into the transition right now is less than 10%. On the server side, I'd say it's over 50% of servers that, if you want better performance, there's something that's actionable right now.

InformationWeek: Microsoft has at least two major Windows upgrades coming over the next couple of years, x64 almost immediately and then Longhorn 12 to 24 months out. What's your advice on how companies should prepare for these two waves of operating systems?

Gates: [With] servers you typically change the operating system as you're changing the application and want more application performance. It's very rare for a company to do all their servers at once. They look at what's going on with the application load running on that server. Say they want the database application to run faster; they're looking at a new version of SQL [Server]. That would be a great time to switch to 64-bits. There's not a huge synchronization there, but those individual decisions will lead to a lot of 64-bit activity even in the next 12 months. On the desktop, people tend to be more saying, OK, on the older machines, we'll run the old operating system and when the new machines come in, they'll come in with the newer operating system, that's where things trickle in with new. Or, you just try and run the same thing on everything. And there, as Longhorn comes out, people will decide, do they want to let the new machines come in with Longhorn, which we'd certainly recommend, and do they want to go back into the installed base of hardware and go for that broad uniformity. The tools we have now for deployment, compatibility testing, are dramatically improved. We're constantly saying to customers, Why aren't you using the latest, and trying to understand how do we get the marginal costs of them being on the latest Windows and Office, get those costs down, down, down. It's this huge push here.There's a lot of things that need to be done in Longhorn, a lot of things that have been done in [Systems Management Server], a lot of tools that people can use for testing -- that's a huge push for us, because people get more benefit out of our innovation if we're making it easy for them to run the latest version. So, we're very critical of ourselves in saying, 'Have we given people everything that's necessary for that?' So we hope to get a lot moving over to Longhorn. It's been typical that when people are buying machines in small batches, they'll just leave the latest operating system on there, so the consumer and small business market in some ways often moves faster to a new operating system than at least some enterprises. Now, with Longhorn we have enough of a runway to be able to say to people, 'Here's the benefits, here's how you're going to use productivity tools, here's your security architecture, your software distribution architecture.' We hope to get a pretty rapid move over to Longhorn, but it's hard to say. We did see that with Windows 2000. That was a case where enterprises did move faster than other markets. In fact, because Windows 2000 did so well in the enterprise, the delta benefit of moving to Windows XP was less than for people who stayed on Windows 95. So, we have a pretty good installed base of Windows 2000 desktops in corporations; we don't have that outside of corporations. So the Longhorn transition tools will address people on Windows 2000, Windows XP, those will be the two big bases. There's still some Windows 9x, but that's pretty small at this point.

InformationWeek: What about mobile devices or embedded software? Is there a play there with 64-bit or Longhorn?

Gates: Embedded is a very special market that we do special releases for, and the embedded space you can break down into lots of different categories: real-time embedded, certain consumer embedded things, we treat the auto market as a very special market, set-top boxes, phones. They're all very special. So, there's a whole road map on how the embedded stuff gets more configurable, and the tools get better there. Some of your readers are interested in that, but less than 10% are the engineers who are looking at, 'What do I put inside this printer, this car, this medical-scanning machine?'

Mobile is the hottest area. Mobile is where we get the fastest upgrades because we're always adding features that are particularly attractive to mobile like, even with Windows XP, the Wi-Fi capabilities meant that mobile machines went a lot faster. Mobile users also tend to be heavier users, using a broader set of applications, they've got a somewhat higher budget for the hardware because mobile hardware costs a bit more. The transition in mobile to Windows XP was pretty dramatic and even the upgrade to [SP2], because mobile machines are often outside the firewall. A machine that's protected by the firewall you can say, 'It's OK. The firewall is doing some of that blocking,' and not be quite as in a rush. We say to corporations, take your things that go outside the firewall and, if you need to prioritize, those are the things that really need to be up on [SP2]. Longhorn will have tons of features that are very unique to portable machines -- moving your files back and forth to your desktop machine, making that easier, the way we support Bluetooth, Wi-Fi, WiMax, UWB, the way we connect up to your mobile phone. There's a ton of stuff in the mobile space, and it's great to see the growth. As the whole desktop pie grows, there's a big shift to mobile, and there's a leading-edge shift to the mobile devices that have the ink support, the tablet-type machines. It's still a small slice, but we are getting both vertical and horizontal move into that. So we're really trying to push people to that -- from desktop to mobile, from mobile to tablet-type devices, sort of the ultimate tool.

InformationWeek: We're guessing Indigo and Avalon will become available on Windows XP x64, so how different will Longhorn be?Gates: If you turn on Longhorn, the visual look will be different. It's the way we've chosen to advance the user interface and exploit the latest graphics chips. Longhorn, you won't mistake it. It's more of a change than anything since Windows 95, in a very good way in terms of the richness that's there. When you move to 64-bit, you don't see any change at all. It just happens, your 64-bit applications run. There isn't a pixel that's different when you're running 64-bit. Longhorn is a feature release to make all end-user tasks dramatically simpler than they are today, things like the way we integrate search, the way we integrate some performance stuff, the way we advance the user-interface capabilities. So that's where you get features; Longhorn is lots and lots of features that will be very relevant to end users.

You're right that the message to developers is slightly more complex because we've given developers the added benefit that some of the things that are built into Longhorn, they can actually get a redistributable from us that will allow them to have the runtime pieces running down to Windows XP. So if they want to use Indigo to write service-oriented applications, we give them a redistributable that contains these down-level things. So, for the developer, he gets to decide, 'Do I want to use Indigo, or Avalon, or the new storage stuff,' even if he's targeting people who run on XP, whereas the end-user decision is, 'Do I want these features that are there?' You get a lot of top-down enthusiasm for these things because the ability to manage and monitor and secure desktops is substantially advanced by Longhorn.

InformationWeek: There's been some commentary that Longhorn isn't going to be the big change once envisioned. How do you respond to that?

Gates: Well, Longhorn's going to be a very big change. Every time we do an operating system, we consider doing more than we end up doing. But all the big things in terms of performance, user interface, search integration -- those things are happening. The timing of Longhorn in terms of what people want for management, in terms of simplified user interface, Longhorn timing is super good. There's a lot of end-user things that have built up since we released Windows XP. It's been some time since we released XP, so the interest in having new and interesting things is very high. We've bunched these things up. So, where XP was sort of the completion of getting the [Windows] NT technology to be super mainstream, there was a lot of work in XP that was under the covers, not so much end-user features. You really have to go back to Windows 95 to look at a release that had as much in terms of, 'Wow, there's this new way I can find things, there's this new way I can manage a set of desktops.' There's no shortage of capabilities in Longhorn. We've got WinHEC, which is very focused on the 64-bit, but we'll talk about the hardware-related features of Longhorn and give people an update there. We have the PDC in the fall, and then during '06, we'll be doing a lot of betas and getting very close to the final release and everybody will have will have a very clear sense of what's in Longhorn as you get through '06.

InformationWeek: You obviously work on a lot of things at Microsoft. To what extent do you continue to be involved in this operating system work, especially x64 and Longhorn?Gates: Longhorn, there's a lot of decisions about features and things like that, and I'm very involved in that. Longhorn's the product I spend the most time on, more than even Office or Search, which are great things, Longhorn's the biggest thing, probably because everything else builds on top of Longhorn and assumes getting Longhorn right is the most important thing going on here. 64-bit, once we decided to do it, I was involved in some discussions with Intel and AMD, encouraging them to bring the chips along and make sure it matched what we're doing in the software, but the specification for 64-bit is pretty straightforward, so we didn't have to go to a lot of meetings to discuss trade-offs, choices, things like that. I'd get mail from the team saying some of the code they were getting from another group, you know, that group wasn't prioritizing the 64-bit work highly enough, so I'd get involved in that. The amount of time I've had to spend on 64-bit is quite small compared with Longhorn because it's just got a very pure definition -- everything in Windows works 64-bit. It's a huge effort.

This other thing that the team highlights very much is they took the 64-bit release and, some of the way they do the engineering practices and making sure the thing is very locked down for security, they put a lot of energy into that. So it's not like they just did the 64-bit release, it's also the foundation, and maybe this is more an internal thing than an external thing, but a ton of stuff that they did in that 64-bit release is what lays the great foundation for Longhorn, because they did all that security stuff that we knew we had to do for Longhorn, they just went ahead and did it in the 64-bit release. In fact, if we hadn't done all that, we could probably have had this thing out four to six months earlier, but the timing's actually perfect in terms of when the chips are coming out. And everybody who cares has had versions of the 64-bit going back more than six months. That's why MSN Search is running today using 64-bit bit, and we have a bunch of customers using it. The timing between us and the chip guys is really about perfect.

InformationWeek: Windows is about 20 years old, and clearly you're spending a lot of time on Longhorn. What else are you spending time on?

Gates: Let's see, Windows 1.0 is, what, 1986? I think that's right. We did the announcement of Windows more than 20 years ago, but we didn't ship it. And Windows 1.0 wasn't exactly a barn burner. It had, like, four applications that ran with it. We totally rebuilt Windows with the NT technology about 10 to 12 years into the Windows history, we really rebuilt what was underneath, and now we're starting some of those same things. Some of this virtualization technology lets us have ways of doing compatibility with all the Windows applications, it's obviously a big asset, but also having in the system, taking all the Web services work we've done and building that down really into the heart of the operating system to have something that's very focused on this new generation of applications, so you can really get the best of both worlds. That's why we've been pushing this virtualization hardware both for server loads and client loads, and over the next three of four years, virtualization is a very big thing. It gives up the opportunity to do the new operating system design where the compatibility is pretty straightforward.

So we actually have three things that are new operating system-type things. One in research, one in an avant-garde product group, one in the Windows group itself, looking at different ideas, and we're really coalescing those. So we're going to call this thing, we could have changed the name of Windows when we went with the NT technology, not a line of code survived by the time we all through with it, it was all completely new, it just happened to run those same applications; they even ran DOS applications. That's a case where we did change the name. We could still be calling this thing DOS, but once we got the GUI and it looked so different, we changed the name. As we go to this next one, we don't really have a time frame, over the next five to 10 years, it will probably keep the same name because the value of running those applications is there, and we have been bringing in support for these new service-oriented applications with things like the Indigo runtime. There's never one day where, boom, overnight things are different, the newer applications will just run better. So we are renewing the technology. There are a lot of ways that systems are very different than they were 10 years ago when NT was such a timely, important transition for us. A lot of people questioned, would we get through that? Could Windows be both a broad high-volume system and have all the power that people, back then, associated only with Unix? We achieved that.InformationWeek: How long will Microsoft continue to develop 32-bit software? And how long will it continue to be around running in customer environments?

Gates: It will be around for ... infinite. There are DOS applications running today on top of these systems, so we're not going to have a version of the operating system that doesn't run DOS applications and doesn't run 32-bit Windows applications. Part of the Windows proposition is compatibility. Fifty years from now, somebody may say, now's the time, but not in my lifetime is that likely to change. In terms of how much, when you do new features, will you always do 32-bit things? For the foreseeable future, but we'll sit and talk to customers. Take the extreme case for us, SQL server. Is it possible that five years from now all of our SQL customers would say, we just want 64-bit? It's possible. But it's in that kind of time frame for the applications that are very 64-bit oriented. Remember, we're very high volume, so Microsoft needs to serve a higher variety of customers than anybody else. We ship things in the hundreds of millions, so we need to keep 32-bit around a long time. For some ISVs, they may have a quicker transition where they tell people, hey, let's get over to 64-bit. I bet if I talked to the ISV group, there's some of those very high-end ISVs that are looking over the next several years, for their new features. They'll keep supporting their old software, but for the really new stuff, they'll say to their customers, get on 64-bit machines. So it's not imminent for anything but very high-end software.

Photo of Bill Gates by Christian Lambiotte/AFP/Getty Images