Intel has just released their press release about Atom “Silvermont, to a handful of tech sites only, which makes me very suspicious about their already pretty unbelievable claims about the performance and battery efficiency of the new chip design. But let’s explore some of their claims anyway.
Intel claims the new chip is “3x faster” than the old one, which may or may not be true, depending how you look at it. Most people will see this as meaning that it’s 3x faster than the tablet chip, Clover Trail. That’s certainly not true. What they mean is that the new quad core Atom is 3x faster than the old (original) Atom core, launched in 2008. They are basically comparing the architecture here.
Then the claim could be true, because ARM chips have improved many times more than that. Just look back at where ARM was in 2008. The highest end ARM chip was a 420 Mhz ARM11 chip, such as the one found in the iPhone. There may have been some 600 Mhz ARM11 around back then, too. Now compare that to the 1.9 Ghz Cortex A15 we can see today in Tegra 4, and the 2.3 Ghz in Qualcomm S800 (yes, both are not shipping yet, but neither is Silvemont, and these two will likely ship before it does).
With a 1.1 DMIPS/Mhz performance for ARM11, and 3.5 DMIPS performance for A15, and with the difference in clock speed between the two, we get about a 10x improvement. And that’s without even considering it’s a quad core chip, just comparing the single cores. It’s obvious ARM chips have evolved much faster in performance. Atom also needed to focus on power consumption, because it was already much faster than ARM back then.
Anandtech claims the architecture performance should be around Cortex A15 level, but Anand is usually a little “too” optimistic about the expectations for Intel’s future chips, and considering Silvermont is dual-issue, while A15 is triple-issue, I have a hunch A15 will still have a lead on it, at least at the same clock speed. If Intel manages to pull ahead in clock speed, and beat Cortex A15 in overall performance, that’s a whole other story. But that remains to be seen.
Surprisingly, no Intel slides sent by Intel to these several sites, are showing the GPU performance. But we do know it has 1/4 of the cores in the IVB GPU, which means 4x lower performance by default. That could put it behind the iPad 4 GPU (launched last year), although Intel will probably increase the clock speed a bit. As they did with Clover Trail+ this year, I expect them to be a full generation (a year) behind in GPU performance compared to just about everyone else.
Atom’s weakness has always been GPU performance, by far, and it looks that even with the IVB graphics, it won’t be able to catch-up. Intel themselves have said that the GPU only supports up to 2560×1600 resolution, while every ARM chip launching this year from Tegra 4, to Adreno 330, Mali T628 and PowerVR Series 6 will support 4k resolution (2x more pixels/performance). So my prediction that Silvermont’s graphics will be about a whole year behind (or 2x slower) is most likely true.
While Intel’s Silvermont will be made at 22nm, and ARM chips are mainly made at 28nm these days, I believe it will only have about a few months advantage in terms of process technology, as next year (probably early next year) we should start seeing 20nm ARM chips. Also Silvermont, will only arrive this year in tablet version (Bay Trail), and it won’t arrive in smartphone version (Merrifield) until next year. By then it’s very possible the 20nm ARM chips will be ready.
Considering Intel doesn’t seem to compete in GPU performance, Intel’s Atom could once again prove to be not competitive enough. But even if it is competitive from every point of view, it should still be significantly more expensive than the ARM chips, and will still have to compete for the OEM’s attention with many other ARM chip makers. So whatever the performance of the chip, Intel should continue to struggle in the mobile market for the time being.