Having just released a lot of architectural information about the Bobcat based netbook APU AMD would appear to be on a roll, so it would be a real downer for them if Intel pre-announced an intention to directly compete in this market segment, however there is more than one Bobcat based APU planned and this blog questions whether Intel is able to challenge both.
With what do they intend to bridge the 5W to 18W divide?
According to AMD’s Fusion marketing director, John Taylor there will be two Bobcat based products:
- An 18-watt TDP APU codenamed “Zacate” for ultrathin, mainstream, and value notebooks as well as desktops and all-in-ones.
- And a 9-watt APU codenamed “Ontario” for netbooks and small form factor desktops and devices.
So where does Intel fit into this?
At the low-end Intel has the Atom SoC based on the pineview platform with a TDP of 13W plus another 5-10 watts for the Ion2 GPU required to make it useful. This has two problems as noted previously; an underpowered in-order CPU that is really designed to supply the MID/phone markets, and a thoroughly inadequate GPU which also fails to support OpenCL for GPGPU tasks.
At the high-end Intel has the Core i3 platform which combines a powerful and efficient CPU architecture in a power envelope that can be brought down to as little as 18W. This has the problem that although it is a power efficient CPU, it is not designed as a low-power architecture and so will struggle to reach down to 18W without extensive power-binning that pushes the cost well outside the netbook bracket. Its GPU is also underpowered and currently does not support OpenCL either.
Thus does the gap becomes obvious, Intels Atom is insufficient for general computing tasks, and is not able to compete with Ontario on performance, and its Core i3 architecture is unable to compete with Zacate on price.
Intel is poised to release its new Sandy-Bridge architecture, which unlike the early Atoms and the i3 platform is a genuine single-die APU, rather than a single package consisting of separate CPU/GPU portions. It will have even better performance-per-watt than Core i3, and a much improved GPU that in its single core iteration should provide performance comparable to a 40 shader ATI 4450 GPU of the previous generation, with OpenCL support to boot.
There will no doubt be a new Core i3 equivalent product based on the Sandy-Bridge architecture, almost certainly sporting a dual-core CPU and the single-core GPU of its bigger brother. However, all is not rosy, for Core i3 already has its CPU manufactured at 32nm, so we should not expect TDP to drop any on that account, and although the GPU is now on-die at the same fabrication process it is also more powerful than previous designs so again an unlikely candidate for power savings.
So Intel will be able to manufacture a Sandy-Bridge based product with an 18W TDP, but for it to be badged as a Celeron with an appropriately cheap price tag will necessitate disabling hyper-threading and portions of the cache as well as down-clocking the GPU even further, and this still doesn’t get Intel below 18W. Worse still, this powerful architecture uses a lot of transistors making it far more expensive to produce than Ontario/Zacate, which can only be described as sub-optimal when you consider that so many of them will be disabled for reasons of market segmentation.
Intel can compete against Zacate, and without doubt provide superior performance in most metrics, but is left with nothing to compete against Ontario in the price and power sensitive netbook segment. Equally, Intel will be unable to move Atom’s performance upward as its in-order design is fundamentally targeted at the ‘device’ market rather than the computing market, and even there it will struggle as ARM’s architecture is fundamentally better optimised for low-power, whilst it performance is rapidly approaching that of its x86 competitors and the arrival of SoC products sporting Cortex A15 CPU’s should be particularly worrying for Intel.
AMD’s Ontario product should prove to be a roaring success, and matters will only improve once they transition to 28nm for Intel will not only be competing in a significant market sector with an inappropriate architecture but they will be doing so without a huge fabrication process advantage.