Single page Print

AMD's E-350 Fusion APU on the desktop

Zacate takes on Atom—and the world
— 11:45 PM on February 20, 2011

I'll admit it. We are about to commit a terrible crime against the holy writ of product segmentation. We're about to consider the merits of AMD's Brazos and Intel's Pine Trail—incredibly small, low-power, and inexpensive PC platforms—versus a whole range of full-grown desktop CPUs, a ridiculously unfair and inappropriate endeavor if ever there was one.

Before we dig into our heinous crime against carefully calibrated marketing, we should take a second to explain how we've arrived at this sorry state of affairs. First of all, you should know that we have some history of violating product segmentation lines when we've found it interesting to do so. After all, running a Pentium M on the desktop seemed at one point like a pretty decent idea. Since Brazos and Atom serve as the basis for a number of small-form-factor desktop systems, asking how they compare to larger, more expensive options felt like fair game to us.

Then there's the fact that AMD made us do it. You see, back when the firm revealed the first details about the Bobcat CPU architecture used in Brazos, it predicted Bobcat would achieve an "estimated 90% of today's mainstream performance in less than half the silicon area." Now, such statements are famously slippery, but those words did conjure up some rather dreamy expectations inside of our heads. We likened Bobcat's potential performance to 90% of the Athlon X2 255's—then arguably a "mainstream" part as these terms tend to be used—and noted that "the X2 255 is more than up to the task of running modern games" and "should be plenty adequate for the vast majority of everyday computing tasks." With prospects like that, a comparison seemed to be in order.

Also consider the fascinating fact that x86-compatible CPUs span the sheer range that they do, from tens of dollars to thousands, from a handful of watts to well over a hundred, from tiny "APUs" to behemoths comprised of over a billion transistors. Faced with this amazing dynamic range of cross-compatible processors, we couldn't resist a little exploration. Particularly when we already had a fresh set of results for desktop CPUs spanning from $87 to $999 created for our Sandy Bridge review.

So yes, we have thrown AMD's newly minted Brazos platform into the ring against Intel's Pine Trail (the latest Atom platform) and forced the two of them to fight it out against a backdrop that includes the full range of today's desktop processors. We think it's an intriguing exercise. Of course, we've also taken time to investigate the question of adequacy for basic computing tasks, which will be critical success of these low-cost, lower-power platforms, regardless of how they match up to the big boys.

Brazos busts out
Let's back up a second, for those who aren't yet fully acquainted with AMD's Brazos platform, and do some quick introductions. We've already previewed Brazos and taken a preliminary look at its performance, in case you missed those articles. In sum, Brazos is a low-cost, low-power, PC-compatible computing platform that aimed at a number of places where traditional PC processors won't always easily fit: netbooks and ultraportables, compact home theatre PCs, ultra-slim desktops, embedded systems, and perhaps even tablets. You'll grasp the basic concept behind Brazos if you compare it to the most common variants of Intel's Atom, although AMD has promised superior integrated graphics and video decode acceleration for Brazos—key pain points for a lot of Atom-based systems.

The key pain point for Brazos, by contrast, may be all of the code names, brands, and acronyms being thrown about in reference to it. Perhaps we're just a little dull after too many late nights with Infinity Blade lately, but it all seems rather bewildering to us. To help you keep track, here's a quick Brazos-related decoder ring.

  • Bobcat — This code name refers to the architecture of the CPU cores that drive the main chip in the Brazos platform. Bobcat is an all-new, clean-sheet CPU architecture tailored for power-efficient operation. Bobcat supports many of the modern extensions to the x86 instruction set, including AMD64 and Supplemental SSE3. Unlike the Atom, the Bobcat core is capable of out-of-order instruction execution, potentially giving it the ability to execute and retire more instructions per clock.
  • Ontario — The code name given to the first chip to incorporate a Bobcat CPU core. Ontario has two Bobcat cores, DirectX 11-class Radeon integrated graphics, and an onboard memory controller. Ontario chips are rated to fit into a 9W thermal envelope.
  • Zacate — Somewhere along the way, AMD decided it needed a second code name to describe the variants of Ontario that will fit into an 18W power envelope, and Zacate was born. The bifurcation of marketing code names was likely deemed necessary because the 9W parts and the 18W parts will serve different markets. The Brazos system we're reviewing today is Zacate-based.

  • Logical block diagram of the Brazos platform. Source: AMD.

  • Hudson — This is the I/O support chip that, together with Ontario/Zacate, forms the basis of the Brazos platform. Hudson handles most of the traditional duties of a core-logic chipset, although like many "chipsets" these days, it's just a single chip. AMD calls Hudson an FCH for "Fusion controller hub," a variation on the more widely used "I/O controller hub" or Intel's "platform controller hub."
  • Fusion — Back when AMD bought ATI and brought Radeon graphics into the fold, the top executives involved talked a lot about how CPUs and GPUs were on a collision course. They projected a future where the responsibility for the execution of programs could be shared between different types of resources. Branchy, sequential threads could run on traditional CPU cores, while problems ripe for parallelization could run on arrays of FLOPS-intensive execution units—in other words, highly evolved GPUs.

    Such talk got us computing enthusiasts all hot and bothered, so AMD decided to bottle the lightning. Soon, "fusion" came to be applied to a very different sort of animal, a product that incorporates very traditional CPU and graphics cores on the same chip for the sake of cost and power savings. This concept was much less exciting, but the fusion name granted the whole enterprise a halo of futuristic awesomeness.

    Finally, with the introduction of Ontario/Zacate, Fusion has become an official AMD brand name encompassing both the Brazos platform and future products based on an upcoming, much larger, Phenom-derived chip code-named Llano. Neither of them has any of the hallmarks of the fully realized fusion concept, such as a programming model with shared memory (or even shared cache) between the CPU and GPU cores.

  • APU — Stands for "accelerated processing unit," AMD's term to describe its CPUs with integrated graphics. Ontario/Zacate is billed as "the world's first APU." Intel calls its similar integrated chips CPUs, and given that the programming model hasn't changed for either, we're betting this new term won't stick.
  • Vision — AMD's all-encompassing brand for its PC platform solutions. Near as we can tell, Vision is AMD's new flagship brand, supplanting Radeon and Phenom in importance, judged by the placement of Vision stickers on laptops and desktops.
  • Radeon HD 6310 — The name given to the integrated graphics processor (IGP) in Zacate. The lower-clocked version of this same silicon in Ontario has been dubbed the Radeon HD 6250.
  • Uh, none, really? — The official brand name of the Zacate/Ontario APUs. The Zacate model we're reviewing today is known as the "AMD E-350 APU with Radeon HD 6310 graphics." There's no snappy name like Phenom or Athlon involved, just some alphanumeric designators. The E-series APUs are based on Zacate, while the 9W Ontario APUs are in the C series. The table below lays it all out for you.
Model CPU
E-350 2 1.6 GHz HD 6310 500 MHz 18 W
E-240 1 1.5 GHz HD 6310 500 MHz 18 W
C-50 2 1.0 GHz HD 6250 280 MHz 9 W
C-30 1 1.2 GHz HD 6250 280 MHz 9 W

Phew. So yeah, those are the Zacate models, and the handy list above will help you navigate the minefield of Brazos platform terminology for days and possibly even weeks before it all changes again.

The Hudson FCH (left) and Zacate APU (right)

Now that we've defined our terms, we can talk in a little more detail about the Zacate APU. The chip itself, shown on the right above, is a very compact 75 mm². It's manufactured at TSMC on the same 40-nanometer fabrication process used to produce the latest Radeon and GeForce GPUs. Each of its two Bobcat-based cores has 32KB L1 data and instruction caches, along with 512KB of L2 cache.

The built-in Radeon graphics processor looks to be very similar to the discrete "Cedar" chip used in the Radeon HD 5450. In other words, it's based on the same fundamental technology as other Radeon HD 5000-series GPUs, only scaled down to fit this application. The IGP has two SIMD engines, each with eight execution units that process five-wide vectors. That works out to 40 ALUs per SIMD and 80 ALUs total for the IGP overall. The texture units aligned with the SIMDs can sample and bilinearly filter up to eight texels per clock cycle, and the IGP's lone ROP unit is capable of outputting four pixels per clock. At 500MHz on a Zacate APU, those specs don't work out to a tremendous amount of graphics processing power. Because the IGP has to share a single DDR3 1066MHz memory channel with the two CPU cores, it probably won't be as fast as a discrete Radeon HD 5450. Still, an integrated GPU anywhere in that neighborhood is a huge step up from what Intel has been fielding in its comparable Atom solutions.

Perhaps even more important than the graphics horsepower on offer is the full DirectX 11 feature set—this thing can do tessellation and high-precision texture formats, for gosh sakes—and the related fact that the IGP will be supported by AMD's Catalyst video drivers. Even if the IGP is a little puny for some tasks, its basic compatibility should be quite good.

Aside from graphics, the IGP packs one of Zacate's most notable features: the UVD3 video processing block, the same third-gen unit present in the latest Radeon HD 6000-series GPUs. Aboard a low-power system like this one, this logic's ability to offload H.264 decoding chores from the CPU is absolutely critical to a good user experience, especially since the web is now so saturated with H.264-encoded videos. With proper support for UVD3, applications like Windows Media Center and Adobe Flash 10.2 can help Zacate play heavily compressed HD videos fluidly—something this class of CPU, even in dual-core form, may struggle to do. Witness the droves of Atom-based slim desktops and netbooks whose frustration factor has skyrocketed with YouTube's rise. Dedicated video processing hardware should be more power-efficient than CPU-only software routines, as well.

Zacate also includes the necessary bits of glue to hold everything together, such as an internal bus, a memory controller, a four-lane PCI Express interface, and an interconnect to the Hudson I/O chip. All told, that's an awful lot of capability packed into one little chip.