AMD's first desktop Fusion processors were detailed on Thursday, and the company marked their debut with a confident blast at its megacompetitor, Intel.
"We win – that's our competitive positioning," Sasa Marinkovic, AMD's head of desktops and software product marketing told a press gathering when the A-Series APUs (accelerated processing units) were previewed at the AMD Fusion Summit in Bellevue, Washington earlier this month.
The two quad-core CPU/GPU mash-ups that Marinkovic is so confident about are the A8-3850 and A6-3650. The A8-3850's CPU is clocked at 2.9GHz; its GPU, with 400 Radeon cores, spins at 600MHz. The A6-3650 has a 2.6GHz CPU, and a GPU with 320 Radeon cores running at 443MHz. Both have 4MB of L2 cache and 100W TDPs, both GPUs are DirectX 11–capable, and both were fabbed in GlobalFoundries' 32nm process.
Elaborating on the A-Series' competitive advantages, Marinkovic noted that despite the fact that each of the two chips are quad-core, the GPU takes up a third of the total silicon. This isn't just for pure graphics performance, he noted. Instead, he emphasized the theme that AMD has been hammering home for some time: that GPUs aren't just for graphics anymore.
This brave new "heterogeneous computing" world is one in which increasing amounts of processing is offloaded from the CPU cores to the GPU cores. "The reason we have such a big GPU inside," said Marinkovic, "is because of the applications that are being developed, and how we want to provide this stuff for the software guys to really develop, and really take the best computational power out of the silicon, and really apply it to their applications."
As Marinkovic said in a company blog posting: "We know the future of microprocessors will not be just traditional CPUs, or even the combination of CPUs and graphics processors (GPUs), but instead will consist of all sorts of 'heterogeneous' processor cores working together."
The A8-3850 and A6-3650 – which will retail for $135 and $115, respectively, when they become available on Sunday – may be designed for this hoped-for heterogeneous future, but their CPU cores still have a foot in the past. Those cores are based on AMD's Phenom technology, which you could characterize as either aging or tried-and-true.
The first Fusion processors to use AMD's next-generation Bulldozer core won't appear until next year, when the company's "Trinity" processor – which was demoed at the Fusion Summit – is scheduled to appear.
But the new A-Series parts do support the current fave foundations of consumer-level heterogeneous computing – or, at minimum, GPU-assisted massively parallel processing: OpenCL 1.1, C++ AMP, and DirectCompute.
As The Reg noted on Tuesday, motherboards ready for the A-Series APUs have already appeared from Asus, MSI, and Gigabyte. AMD has updated that list, adding ASRock, Biostar, ECS, Foxconn, Jetway, and Sapphire.
Considering Intel's deep pockets and marketing muscle, Marinkovic's "We win" effusion may be a bit brash. In the microprocessor market, size does indeed matter. Intel's market capitilization as of this Thursday morning was $113.15bn, nearly two-dozen times the size of AMD's paltry $4.78bn. Chipzilla's mere bulk can inspire customer confidencer in terms of assured distribution, multi-tier deal-making, and we'll-be-there-tomorrow security.
And then there's the fact that Intel's latest desktop processor line, "Sandy Bridge" (aka the 2nd Generation Intel Core Processors) has one fine compute core – even if its graphics core, although improved from past stumbles, isn't quite up to Radeon-class performance.
And so the question arises: at the consumer level, where visuals are more important than pure number-crunching, does the A-Series' better graphics trump Sandy Bridge's more-modern compute cores?
Heterogeneity, shmeterogeneity – from a consumer perspective, the real question is how well will the A8-3850 and A6-3650 play the new DirectX 11 version of Crysis 2, in comparison to a Sandy Bridge integrated-graphics system?
Oh, that's right – Sandy Bridge doesn't support DirectX 11.
Never mind. ®