Mobility Sector Updates: Waiting for Sandy Bridge and Market Analysisby Jarred Walton & Vivek Gowri on February 7, 2011 3:00 AM EST
And in the AMD Corner…
All the talk so far has been of Intel laptops: Sandy Bridge delayed, Arrandale on the way out, etc. But what about the other big player, AMD? They’re probably the only company that actually benefited from Intel’s chipset snafu. Sure, the new laptops with Intel CPUs and Radeon 6000M GPUs are on hold, but if nothing else this gets AMD two months closer to the launch of Bulldozer and Llano. And what about Bobcat (aka, Brazos/Ontario/Zacate)?
AMD’s Brazos APU
We saw dozens of Brazos systems at CES 2011, and we expected them to go on sale right after, but they’re only now starting to hit the shelves. Anyway, constraining our view to the mobile world, HP has the dm1z available, with an E-350 APU and a starting price of $449 (with the current $100 instant rebate). That’s still $150 more than basic Atom netbooks, but you have to look at what you’re getting: a full Windows 7 install (not Starter), 3GB DDR3, 320GB HDD, an 11.6” 768p display, and a much faster processor thrown in for good measure. Let’s not forget the significantly improved video playback capabilities either! Battery life with the E-350 might not quite make it to the level of Atom netbooks, but HP claims up to 9.5 hours, which is more than respectable. These systems look to be far more interesting than Atom as an overall package, and we already have a good idea of how they’ll perform from our mini-ITX Brazos review.
A couple other E-350 based laptops are showing up at retail, but we have one concern: both the laptops we can find are 15.6” units. One system comes from Acer, the Acer Aspire 5253-BZ684/ LX.RD502.015 starting at $450. Like the HP dm1z, it comes with Win7 Home Premium, 3GB RAM, and a 320GB HDD. The other system is the MSI CR650-016US, with the same features and specs and a starting price of $500 online. Zacate seems like a fine idea as an ultraportable, but move up to 15.6” and we start to wonder if it makes sense. Battery life should still be decent, but performance relative to other larger laptops is going to be lacking. We’d suggest these types of systems more for multimedia use (think portable HTPC) rather than a “do everything” laptop; otherwise, you’d probably be better off with something like a Turion P540 laptop—much faster CPU, but lower battery life and GPU performance.
The Brazos laptops using Ontario C-series APUs aren't so prevalent. The only one we can find right now is the Acer Aspire One AO522-BZ897, but it comes priced like a netbook starting at $330. Of course, with the netbook price comes netbook features: 1GB RAM, 250GB HDD, and Win7 Starter (yuck)—though apparently the LCD is one step up from 1024x600 and goes with 1280x720. With lower CPU and GPU clocks, we expect the C-50 won’t do quite as well as the E-350 in terms of multimedia support. However, if you’re just interested in typical 720p or 1080p H.264 encodes it should work well. We hope to have one of these systems in for testing in the near future, at which point we’ll try to provide a better rundown of what to expect in terms of performance and usefulness. Can C-50 beat Atom as an overall platform, or are you better off going the E-350 route? We’ll find out soon enough.
Looking Forward to Llano
As we mentioned above, if there’s one beneficiary of the Sandy Bridge delay, it’s AMD. The SNB launch just moved back a couple months, which puts it two months closer to the Llano APU launch. Actually, that’s not even entirely accurate; we initially expected quad-core SNB notebooks in January (and we had one in hand and a couple more scheduled to arrive this week), with dual-core SNB showing up at the end of this month. Now, we’re looking at late March or early April in all likelihood, with the potential for some vendors to slip as far as May or June. What remains to be seen is how Llano actually performs.
What we know of Llano is that it will combine a K10.5 type CPU architecture with a midrange DX11 GPU (something like the HD 5650), integrated into a single chip. That may sound rather mundane, but the truly interesting part is that it will be manufactured on GlobalFoundries’ 32nm node. This is a major deal as it’s the first time we’ve ever seen a modern GPU built on a state-of-the-art CPU process node. There’s actually a lot more work involved in moving a Redwood GPU architecture to 32nm, as most of the Intellectual Property (IP) related to GPUs targets the so-called half-nodes (55nm, 40m, and in the future 28nm). It’s one reason we expect AMD to eventually move all of their CPU and GPU production to such nodes, but that's a ways off and Llano will use the same process size as Intel’s current CPUs.
Besides the shrink in process, AMD has certainly had opportunity to better tune the K10.5 architecture for power efficiency. If Llano gets a healthy dose of clock and power gating, even though K10.5 may not be an all-new architecture like Bulldozer or Bobcat, it could be highly compelling. We expect Llano will be a reasonably small chip that offers plenty of performance—particularly for graphics programs and games, along with the potential for GPGPU programs—and the price should be attractive as well. We thought the Acer 5551G had a lot of potential at the $600 price point, and Llano should enable better performance at a lower price. We expect Llano to hit the market round about June, give or take, and if AMD can push it out as early as May they could really steal some of the Sandy Bridge thunder, at least as far as moderately priced gaming laptops go.
What About Bulldozer?
Of course, there’s still Bulldozer to discuss. We really don’t have much to go on as far as performance information goes, but let’s look at the architectural design for a moment. AMD is putting two full Integer cores inside each Bulldozer core, which means the design should excel at heavily threaded integer workloads. The concern is that heavily threaded integer work may not be that useful for most users. We already have difficulty taxing four integer cores without resorting to heavy multitasking scenarios, and multithreaded tasks like video encoding and 3D rendering generally need more floating-point performance. Bulldozer is the successor to the Opteron legacy, which gives a clue as to where it should really shine: servers.
So what does that mean for Bulldozer derived chips for notebooks? Performance is a huge unknown right now—there are just too many factors involved (i.e. issue width, cache size, branch prediction, and other architectural elements) to do anything more than guess at performance right now. It could be an amazingly fast architecture—we certainly hope that’s the case—or it could be only moderately faster than the current stuff. I’ve heard rumblings of performance targets 50% faster clock for clock than K10.5, though, so let’s take that as the goal.
If Bulldozer can provide a 50% performance increase relative to the current K10.5 designs (or even 25%), it will certainly compete in the server and high performance desktop arenas. From there, it should eventually find its way into mainstream desktops and eventually notebooks; however, mobility isn’t a major focus in the initial rollout. My bet is this will play out similar to how the Hammer architecture launched.
The first CPUs were for socket 940 servers and workstations in June 2003 (i.e. Sledgehammer Opteron CPUs), and then we saw Athlon FX “enthusiast” systems in September 2003 (still SledgeHammer on socket 940). It wasn’t until the Athlon chips on socket 754 in December 2003 that we finally got mainstream K8 processors (ClawHammer and Newcastle), and then in June 2004 we got socket 939 (still ClawHammer and Newcastle, only with dual-channel unbuffered memory support). From there, mobile variants of the ClawHammer and Newcastle still took time to appear, and while the performance was good there’s still the question of scaling that down to a reasonable power envelope.
Bulldozer (Orochi) is now slated to show up first on high perforamnce desktops, followed by servers, but it appears these will both be eight-core (four Bulldozer module) designs, and the six-core and four-core variants will come later. As for notebooks, to be competitive with Sandy Bridge, we’d need maximum power draw of 35-45W for the CPU, idle power draw down in the 3-5W range, and “typical” power draw under a light load (i.e. surfing the Internet) well below 10W. Given the apparent server and workstation target of Bulldozer, that’s asking a lot, but it could still happen.
There haven’t been any roadmaps for mobile CPU-only designs, as AMD looks set to move all of their mobile products to Fusion APUs. That means CPU-only mobile Bulldozer offerings might be limited to DTR notebooks packing desktop CPUs (e.g. the AMD equivalent of the Clevo X7200). Whatever performance Bulldozer brings to the table, it's likely that notebooks won’t see such processors for at least a quarter after the desktop launch. Anand guessed at a Q2/Q3'2011 launch for desktop Bulldozer, which means Bulldozer might not join the mobile party until Q4’11 or perhaps even 2012.
The official mobile Bulldozer product is an APU dubbed Trinity, combining a DX11 GPU with 2-4 Bulldozer cores, and that's slated for sometime in 2012. So AMD looks set to concede the highest performance laptops and notebooks to Intel, choosing to focus instead on the (much!) higher volume entry and midrange parts. Let’s hope Llano and Bobcat can hold down the fort for AMD’s mobile division, because we'd really like to see more competition for Intel in the mobile space.