I hate to keep things from you all, but last week I was diligently working in a room at AMD’s new campus in Austin, Texas. You see, AMD wanted to give us more time with the Brazos/Zacate platform we tested at IDF ahead of its official launch. It’s too early for production worthy OEM systems and AMD wasn’t too keen on these reference platforms leaving its offices so it did the next best thing: fly us out to test the systems on AMD’s campus.

The rules were simple. We couldn’t run anything that would harm the system, but other than that we were free to bring whatever we wanted and test however we wanted. AMD dropped by our private room to check to see if we needed anything but other than that, it was all hands off.

The Brazos test platform

While I’d love to share performance data with you today, I can’t. You’ll have to wait another week or so for that. What AMD is allowing us to talk about are the specific configurations AMD’s first Fusion APUs will ship in and general impressions from the testing. Specific benchmarks are off limits unfortunately.

The platform felt final as far as stability goes. I didn’t encounter any crashes during my several hours of non-stop testing. Performance is also indicative of what will ship early next year. The system felt quick (very 11-inch MacBook Air like if you catch my drift) but you have to keep in mind that Zacate and its lower powered sibling Ontario will be used in systems priced between $299 - $549.

Meet the Brazos

The platform is called Brazos. Brazos encompasses two chips: a Zacate or Ontario APU (Accelerated Processing Unit) and the Hudson Fusion Controller Hub (FCH). This is the whole reason AMD bought ATI, Zacate and Ontario are its first CPU-GPU hybrids.

While AMD is primarily targeting the netbook/ultraportable and notebook markets with Ontario and Zacate (respectively), we’ll also see nettops/desktops and mini-ITX motherboards based on Brazos. This may be the first time we see real competition to NVIDIA’s ION platform.

AMD's Zacate APU, 19mm x 19mm package, 413 balls, 75mm^2 die

Both APUs are built from the same die. Zacate and Ontario are physically identical, they just run at different clock speeds. The 40nm die is manufactured at TSMC and measures 75mm^2. The chip is very cheap to package as well. The BGA package only has 413 balls. As one AMDer put it, these chips are designed to be stamped out as quickly and as cheaply as possible.

Both are available in single and dual core versions, although the single core variant is just a dual-core chip with one core disabled to hit various market segments. Core unlocking is apparently possible, but unexpected given the target for these platforms and the amount of effort OEMs would have to go to in order to enable it.

The CPU cores are based on AMD’s first truly low power client architecture, codenamed Bobcat. We’ve discussed Bobcat in great length already, but at a high level it looks a lot like an out-of-order Atom processor without Hyper Threading. The fundamental architecture advantage should give Bobcat the performance advantage over Atom, at least at identical clock speeds.

The GPU is based on AMD’s 5000 series architecture (Evergreen) and features 80 shader cores and AMD’s new UVD3 decode engine.

The SKUs

At launch you’ll see four SKUs, the combinations and specs are listed below:

Zacate takes the top two SKUs, while Ontario makes up the bottom two. The difference in TDP is entirely based on the clock speed of the CPU and GPU. And here is my concern. While a pair of Bobcats running at 1.6GHz are just awesome, drop the clock to 1.0GHz and I start getting concerned about performance. AMD didn’t let us test the C-50 but I’m curious to see what the margin of victory will be over Atom at that speed.

The GPU runs at 500MHz at the high end (Radeon HD 6310) and 280MHz at the low end (Radeon HD 6250). This paired with the single channel DDR3-800/1066 memory bus that has to be shared with the Bobcat cores means you should expect sub-5450 levels of performance out of these systems. Given their price point, that’s potentially not too bad. More enticing however is the fact that these APUs gain the benefit of the driver work AMD does on its discrete cards. Finally, game compatibility should be just as good on the low end as it is on the high end - assuming you meet the minimum specs for the title you’re running.

The lack of a brand for the CPU is a bit puzzling, but graphics are where it’s at these days.

The Chipset err...Fusion Controller Hub

The Zacate/Ontario APUs have an integrated PCIe interface sporting 8 lanes. Four of those lanes are used for AMD’s Universal Media Interface (UMI) - a bus that links the APU to the Hudson FCH. The remaining four can be used for discrete graphics or a combination of discrete graphics and other PCIe controllers. Both PCIe interfaces can run at gen 1 or gen 2 transfer rates.

Display output naturally stems off of the APU. You can drive two displays in tandem over any combination of VGA or Display Port/DVI/HDMI/LVDS.

As I mentioned earlier, the APU sports a single channel memory controller - but you can install up to two DIMMs on that channel. Speeds are down from the 1333MHz system we tested at IDF, DDR3-1066 is the max you’ll officially see on a Brazos system.

The Hudson FCH supplies up to 14 USB 2.0 ports, 6 x 6Gbps SATA ports and another 4 PCIe lanes. The FCH is based on AMD’s SB800 series of South Bridges. Despite the small size, this is a full featured platform.

From the looks of it, AMD has all of the right ingredients for some very competitive entries in the mainstream and ultraportable notebook markets. Now it’s just up to the OEMs to build something cool out of it.

Check back in about a week for the full rundown on Zacate performance.