AMD's Financial Analyst Day: Brazos 2.0, 28nm CPUs, Future SoC's
Events in desktop, mobile, and server are more interesting.
Note the debut of what's being called "Brazos 2.0." It's confirmation that the Krishna/Wichita chips that were originally meant to succeed Brazos were indeed cancelled. Their disappearance leaves AMD with a significant gap in that product segment. Despite the 2.0 moniker, the update is a modest one; the E2-1800 that replaces the E2-450 is said to be 100MHz faster with a GPU clocked at 680MHz rather than 600MHz, support for DDR3-1333, and the already listed USB 3.0 and Turbo Core. The GPU also gets a re-brand, from HD 6320 to HD 7340, despite the fact that the GPU inside Brazos is actually based on HD 5000-era technology.
Hondo is a Brazos chip that's been rearchitected for low power operation. Unlike AMD's first tablet chip, codenamed Desna, Hondo is not a standard Brazos that's been binned for ultra-low power, but a physically different chip. Unlike Desna, which had a 5.9W TDP, Hondo's TDP will be 4.5W. Other features are currently scarce.
This year brings the debut of Trinity in both desktop and mobile configurations; the chip is expected to sample in laptops this quarter. Trinity will ship in three configurations -- 17W, 25W, and 35W, with the first aimed at AMD's "ultrathin" initiative. Sunnyvale hopes to challenge Intel's "Ultrabook" products with systems designed to similar physical specs but at a much lower price and claims that its Trinity CPUs at 17W will offer similar performance to 35W Llanos at half the power.
In desktop, AMD will launch a set of CPUs built on the same technology with Trinity APUs dominating the mainstream segment.
Things change considerably in 2013. AMD's new 28nm low-power platform "Tamesh" debuts, powered by new Jaguar CPU cores in concert with a GPU derived from the current Southern Islands family. Kabini takes over the market Wichita was originally supposed to dominate with 2-4 Jaguar cores, while Kaveri -- an APU based on next-generation "Steamroller" CPU cores -- debuts at the high end. All three chips are SoCs rather than conventional APUs, and all three are built on 28nm rather than 32nm.
In desktop, the aging Piledriver will anchor the high-end until the end of the year. AMD's presentations from last year's Fusion summit indicate that the company's long-term goal is to integrate CPU and GPU until the graphics core essentially replaces the CPU's floating point units. It's possible that holding off on a new higher-end CPU core is AMD's way of delaying until a new GPU is ready to take on that role in the highest end of the market. Such a chip would conceivably debut in 2014, beyond the scope of these presentations.
Everything Hinges on Piledriver
It's no exaggeration to say that AMD's ability to inspire any degree of investor confidence hinges entirely on the performance of Trinity and the Piledriver core at the heart of that chip. AMD's graphics technology may be first-rate, but the company's GPU revenue since 2008 looks like this:
That means everything relies on Piledriver delivering strong performance per watt (graphics performance is a given). Despite AMD's claims, however, we're leery of the CPU. AMD has said next to nothing about the chip's CPU performance, all of the demos and product slides have focused on the GPU. We haven't even seen any slides showing performance gains in CPU-centric workloads compared strictly to Llano. AMD has claimed that Trinity (the APU) will provide up to 20% more performance than Llano -- all we can say is that we hope so.