AMD's last, best hope: Low-power Kabini, Temash are ready for action; could rejuvenate mobile market

Today, AMD is launching Kabini and Temash, the twin follow-ups to its popular 2011 Brazos platform. These new SoCs are built on 28nm technology, feature an expanded, more powerful GPU, and target systems with TDPs ranging from 9-25W. One thing to understand is that while we talk about two distinct code names, the underlying architecture is identical. Temash and Kabini have the same controller hub, have the same memory controller, and integrate the same features.

Temash: The tablet/netbook chip

AMD has been eyeing the tablet market for the past two years, and Temash is the company's first real chance to grab some of that market for itself. Sunnyvale is launching three Temash products at 3.9W, 8W, and 9W TDPs.

The major differences here are core counts and clock speed. The highest-end APU, the A6-1450, will offer a Turbo mode at up to 1.4GHz, with a GPU clock of up to 400Mhz. The A6-1450's low TDP compared to its brethren implies that these are the highest-binning SoCs AMD is able to make. The other two Temash processors are an 8W A4-1250 (dual-core, no turbo, 1GHz clock, 300MHz GPU, but with faster RAM) and a 3.9W A4-1200. The only listed difference between the dual-core A4-1200 and the A4-1250 is the slower Radeon core and the use of DDR3L-1066.

Compared to AMD's lowest-power 40nm part, the Brazos-based Z-60, the new A4-1200 draws 15% less power (3.9W vs. 4.5W), adds 60% more GPU cores (128 vs. 80) but cuts the GPU's clock speed by 20% (225MHz for the A4-1200 compared to 275MHz for the old Z-60). Overall performance should be better thanks to the improved Jaguar CPU core and the much-improved GCN GPU over the HD 5000-class hardware Brazos uses.

Kabini: Ultrathin to mainstream

If Temash is designed to break into tablets, Kabini is the more mainstream part that's headed for the value and mainstream notebook markets. AMD is offering a much wider range of SKUs for Kabini, as shown below:

The E1-2100 presumably exists so that OEMs that want to target a wide range of processors with a single chassis can design a 9W part into it while leaving room to scale all the way up to 25W. The "E" series parts aren't actually very interesting -- a 1.65GHz dual-core Kabini with a 128-core GPU at 450MHz is slightly better on power than the E2-1800 (1.7GHz dual-core, 523MHz GPU, 18W), but performance is likely a wash. The A4-5000, on the other hand, hits the sweet spots. This is a 15W quad-core at 1.5GHz with a 128-core GPU clocked at 500MHz.

If you're curious about performance, Anand has some details on that(Opens in a new window). The long and short of it is this: Kabini is 10-20% faster than Brazos clock-for-clock in single-threaded applications. It's much faster than Brazos anywhere it can leverage multi-threading. It beats the crap out of Clover Trail, and it does so while maintaining very nice battery life characteristics. But when it comes to gaming? Kabini just isn't good enough to drive modern titles. AMD itself alluded to these results when it sent over a slide deck with these results in it.

As soon as I saw that the A4-5000 couldn't push 25FPS in L4D2, I knew this wasn't going to be a pretty show. Granted, this refers to Temash -- the A4-5000's GPU is a full 100MHz faster, which should boost frame rates by a further 20%. What it suggests, however, is that Kabini will still be limited to older titles. This is still an advance over the current situation, but early reviews show the chip struggling to keep pace with Ivy Bridge's HD 4000. Beating Clover Trail's single-core SGX544, however, is no problem.

The big picture

Late last year, I theorized that Kabini and Temash could turn out to be precisely what the struggling x86 tablet/low-power market needed. After spending time with Microsoft's Surface, Surface Pro, and Samsung's Ativ (Tegra 3, Ivy Bridge, and Clover Trail), I concluded that the market badly needed a design that could offer stronger performance than Intel's Haswell without the accompanying hit to weight and battery life.

Can Kabini and Temash do that? Maybe. The A4-5000 beats Brazos out on core count and Intel's Clover Trail on CPU efficiency. The 128-core, 500MHz GPU will outperform the previous HD 5000 hardware thanks to its superior architecture and increased core count. I'm not sure there's much of a market for the 25W Kabini; AMD's new A10-5745M is a Richland quad-core at 25W with a 2.1GHz clock speed and a 384-core GPU. Granted, Richland's TDP figures don't include a south bridge, which tilts the math in Kabini's favor if TDP is a concern.

Still, After looking over Anand's performance figures, I feel pretty good about where Kabini ended up. If you were looking for a laptop with long battery life and performance that sat considerably nearer to Ivy Bridge than Clover Trail (at a fraction of the price), these systems could be precisely what you want. It's harder to predict what will happen with Temash, mostly for reasons beyond AMD's control. For whatever reason, x86 tablets have mostly failed to ignite at market. The problem is, AMD's window of opportunity to grab a foothold in this space while Intel is stuck on 32nm is only 6-7 months wide. Come the first part of 2014, Intel will have Bay Trail ready to go, and Sunnyvale faces a serious fight.

But Kabini, Temash, and the triple-set of licensing royalties from the Wii U, Xbox One, and PS4 are going to give it a shot.