Samsung’s dual-core Exynos 5 broke cover with the launch of a new Chromebook a few weeks ago. Now, a full suite of Ubuntu benchmarks show the chip as a credible threat to Intel and AMD’s low-power x86 processors in the tablet and netbook space. Nvidia’s Tegra 3 was similarly abused, though Tegra 4′s imminent ramp should help diffuse the situation.
Phoronix tested the Exynos 5250 against Nvidia’s quad-core Tegra 3, three Intel Atom’s (the N270, Z530, and dual-core D525), and a Core i3-330M with a 35 Watt TDP. The Exynos 5250 smokes Tegra 3 and the two single-core Atom processors, racking up wins by margins of up to 50 per cent. The dual-core, Hyper-Threaded D525 is better competition for the Cortex-A15, but the ARM chip still won more tests than it lost to that particular Intel CPU.
The Core i3-330M, on the other hand, crushed the Cortex-A15 in every single metric. While the article points out that the i3′s TDP is nearly an order of magnitude greater than the Exynos 5250′s, it failed to acknowledge that Intel has made substantial strides in this department. The i3-330M was launched in early 2010 as a 2.13GHz chip with 3MB of L3, no Turbo Mode, and no AVX support. The best current, non-embedded match would be the Core i7-3667U (2GHz base, 3.2GHz Turbo, 4MB L3, 17 Watt TDP) or the embedded i7-3517UE (1.7GHz base, 2.8GHz Turbo, 4MB L3, 17 Watt TDP).
The point here isn’t to diminish the Cortex-A15′s accomplishments, but to put its performance in the proper context. The TDP and performance-per-Watt gap between Intel’s Core products and the current low-power Atom, Brazos, and ARM families have shrunk dramatically in just two years. They’ll shrink further in 2013, when Intel launches 10 Watt Haswell hardware.
The Cortex-A15 is unlikely to challenge Intel’s absolute performance lead, but it could form a potent bulwark against Clover Trail. The D525 is a decent stand-in for the Z2760′s performance, if not its thermals. I suspect that AMD’s Brazos would have put up a better showing – it’s typically 15 to 25 per cent faster than Atom – but Brazos is a bit long in the tooth compared to either chip. As for Tegra 3, these benchmarks stress just how badly Nvidia’s product line needs a refresh.
Of the three companies we’ve mentioned, AMD is easily the most vulnerable to a strong Cortex-A15. Intel has low-power SKUs, high average selling prices (ASPs), and substantial momentum to draw on. Nvidia’s Tegra 4 is already in the pipeline. AMD’s Brazos is aging and its replacement is still half a year or more away. If Kabini slips to later in 2013, it could end up squeezed between the Cortex-A15 at the low end, and Intel’s 22nm Valleyview Atom (the first out-of-order Atom architecture) at the top.
The next 12 months will determine whether Windows RT will seize a significant percentage of the x86 notebook market, or if these devices will remain on the periphery. Based on these performance figures, I suspect Cortex-A15 designs will be the first real competitors for x86 tablet and netbook form factors. Tegra 3 may have served as a launch vehicle for the Surface RT, but the dual-core Exynos is dramatically faster than Nvidia’s quad-core design. Windows RT and Android tablet manufacturers aren’t going to ignore that.
For more on ARM’s Cortex-A15, check out our article from earlier this month on understanding the Cortex A-15, and our performance analysis looking further down the line.