News - Samsung's 'Exynos 5 Dual' faces-off against Intel's 'Atom N570'

[kingpotnoodle]

Intel are not dominant in CPUs. Intel are dominant in desktop/laptop/commodity server CPUs. ARM based architectures have dominated in phones for a long time, and other markets (set top boxes, embedded systems, etc etc) have different dominant architectures and manufacturers.

ARM architectures are not going to be killed off by Intel bringing Atom/Core down the power ladder and neither are Intel going to be killed off by ARM architectures moving up the performance ladder. Nobody here is an underdog they haven't always been.

This comparison is an old Atom vs a brand new Samsung Exynos, neither of which are suitable for anything else other than crappy little netbooks/nettops. Compared to a Celeron/Pentium/Athlon they are slow and compared to a smartphone/tablet SOC they are slurping power, Atom is worse at idle but both are using far too much under load for the tight chassis of a phone or thin tablet. It's an interesting comparison but it doesn't tell us anything for definite except "Samsung's new Exynos better for Chromebooks than old Atom" - which was kind of expected anyway. New CPUs tend to be better than 2 year old ones ... even AMD manage that sometimes. Note that the Bonnell (Atom) architecture is actually 4 years old and is only now being beaten on performance by an ARM architecture...

To go in a phone that Exynos would have to be several voltage points lower and lower clocks, Cortex A15 wasn't really intended for phones and only maybe for tablets (hence Qualcomm Krait and latest Apple designs are enhanced A9 not A15)... Intel could get more performance in that Chromebook's power envelope with a down-clocked Core i3, it'd be expensive yes but desktop and laptop CPUs are different in their I/O etc, tightly integrated SOCs are a different beast.

Intel have got a process advantage meaning that they can die shrink Atom and clock it higher, they have new core designs etc coming through, so whilst the Samsung Exynos is fast now it might not look so good in comparison by next year and against the same generation parts.

The only really useful and informative face-off will come when we have an Exynos and an Atom, BOTH intended for use in a tablet in the same power envelope, both built into otherwise very similar tablets - then we can test performance and battery life hopefuly using the same OS (Windows RT maybe).

[watercooled]

Started off well but:

Atom/A15/Bobcat/Nano/whatever are designed and used for far more than 'crappy little netbooks' if you just look. That doesn't exactly do justice to the massive market these ICs are intended for. Besides, Atom is more than capable of providing a decent user experience if the OEM doesn't make the stupid decision to install Vista or next to no RAM with 7, Nano and Bobcat are better again, as I imagine A15 will be.

You can't compare power use based on that article, there are lots of other components in the laptop besides the SoC believe it or not; they would be designed very differently for a tablet or similar, and the SoC would be clocked lower, as is common for Tegra 3 for example. Besides, the Chromebook is similar to a tablet in that it has no active cooling. In reality, the A15 will be using next to nothing at idle, the same can't be said for the Atom but the Atom SoCs are similar. Also remember the A15 Chromebook has a much smaller battery.

Thing is, despite the age of the uArch, Atom is what Intel still use, and will be using for some time, for everything from phones to nettops. They have their 32nm version now, which is comparable to Samsung's 32nm, and Samsung should have 28nm out very soon. Bear in mind Intel don't yet have a low power 22nm node; the constant argument they 'have a process advantage' isn't strictly true, not for SoCs at least. A15 will end up in phones soon enough, the power difference between a dual A15 and quad A9 that lots of phones are using isn't that great even on current nodes, that will improve again with shrinks.

Comparing a massive, much more expensive laptop CPU is pointless. Look at the price difference between the old and new Chromebooks, that really matters as it's hard to sell such a niche product for the price the Celeron version costs.

[kingpotnoodle]

There was some good bits in the middle, around the edges and just underneath but:

Started off well but:
Atom/A15/Bobcat/Nano/whatever are designed and used for far more than 'crappy little netbooks' if you just look. That doesn't exactly do justice to the massive market these ICs are intended for. Besides, Atom is more than capable of providing a decent user experience if the OEM doesn't make the stupid decision to install Vista or next to no RAM with 7, Nano and Bobcat are better again, as I imagine A15 will be.

Neither Atom nor this Exynos A15 in the particular incarnations compared are good enough for anything other than basic duties, the Anandtech article says as much. They're too power hungry for phones/tablets, and too slow to do more than the basic internet. They won't be stealing desktop share... I suppose they may be useful in embedded systems, set top boxes or kiosks, thin clients etc but I'm not sure they're the perfect fit there either. Sure other variations on the Exynos and Atom are good for other purposes, but not really those 2 particular chips + chipsets.

You can't compare power use based on that article, there are lots of other components in the laptop besides the SoC believe it or not; they would be designed very differently for a tablet or similar, and the SoC would be clocked lower, as is common for Tegra 3 for example. Besides, the Chromebook is similar to a tablet in that it has no active cooling. In reality, the A15 will be using next to nothing at idle, the same can't be said for the Atom but the Atom SoCs are similar. Also remember the A15 Chromebook has a much smaller battery.

Yes of course different systems are designed differently - kind of obvious no? (Hence my last sentence...) Next to nothing idle but Anandtech estimate up to 4W load, that's way too much for a phone but maybe OK for a tablet if done right. The A15 won't look so amazing (still look good I expect though) down clocked to match the power envelope of the Atoms, Tegra and Krait chips that are already in phones.

Thing is, despite the age of the uArch, Atom is what Intel still use, and will be using for some time, for everything from phones to nettops. They have their 32nm version now, which is comparable to Samsung's 32nm, and Samsung should have 28nm out very soon. Bear in mind Intel don't yet have a low power 22nm node; the constant argument they 'have a process advantage' isn't strictly true, not for SoCs at least. A15 will end up in phones soon enough, the power difference between a dual A15 and quad A9 that lots of phones are using isn't that great even on current nodes, that will improve again with shrinks.

Intel have Silvermont an out of order Atom core due next year, and they are generally ahead on processes but yeah of course everyone is moving forward and no the current Atoms are not 22nm but they will be with Valleyview etc.


Who's fastest/best/lowest power/value goes in swings and roundabouts anyway cos those pesky manufactures refuse to synchronise their release schedules!

[watercooled]

As I said, there are plenty of phones using quad A9 SoCs, including Samsung's previous Exynos used in the likes of the S III. A15 does not use twice as much power as A9, all else being the same, so it's perfectly reasonable to expect dual A15 in phones soon, let alone tablets. That 4W increase is not directly applicable to phones/tablets, again as I said earlier, they use different power supply designs, different busses and such i.e. the load power of the SoC is NOT 4W. 2 or even 3 watts peak is not unreasonable for a phone or a tablet and is fairly common already; people tend not to do video encoding on their phones so peak power isn't sustained for long, and lengthy tasks like video are offloaded to hardware, meaning far lower power consumption. big.LITTLE is another concept to lower power consumption during less demanding tasks, somewhat similar to what Tegra does, and does quite well.

Any of these CPUs are capable of far more than 'basic Internet', and will perform common media tasks with ease. I don't recall the Anantech article hinting otherwise, but of course, these aren't designed to compete with socketed CPUs at all. They may net steal much desktop market share, but it's more to do with the non-x86 nature posing compatibility problems rather than performance; nettops have sold plenty, and the A15 is perfectly capable of taking on that role.

A15 at current clocks, which would be reasonable to expect in tablets, is matching or outperforming nettop/netbook CPUs with a far higher TDP; instructions/clock is high, so it's reasonable to assume it will still compete well when clocked down for phone use. I don't think anyone was assuming performance would remain static when clock dropped? In other words, it's impressive at tablet clocks vs even nettop CPUs, it's likely to be impressive at phone clocks vs phone CPUs?

I never disagreed performance/node/etc varies between MFRs; I was just saying the 'Intel has process advantage' is reiterated too often in cases where it simply isn't true.

[kingpotnoodle]

http://www.anandtech.com/show/6425/g...exus-10-review

It's more half information and speculation fuel but interesting anyway.

Looks like they crammed the same Exynos SOC into the Nexus 10, looks impressive but Anandtech have said they think the extreme screen resolution might be hampering memory bandwidth and killing the CPU because it lost a couple of benchmarks to iPhone5 and the Razr-i phone (Atom). Judgement reserved on the power usage but they do wonder if it's got a little less to play with than the Chromebook. Impressive SOC for a tablet though

Worse though they reckon the Snapdragon S4 Pro quad in the Nexus 4 is throttling a lot, does look as as some of the enhanced A9 quads are a bit much for a phone so probably anything A15 (quad) based will be too on the current manufacturing. Perhaps Apple got it right with a fast enhanced A9 dual as the quad doesn't seem to add much over the dual core S4. Intel need to hurry up and get those Silvermont cores shipping next year to make it interesting.

Any of these CPUs are capable of far more than 'basic Internet', and will perform common media tasks with ease. I don't recall the Anantech article hinting otherwise, but of course, these aren't designed to compete with socketed CPUs at all. They may net steal much desktop market share, but it's more to do with the non-x86 nature posing compatibility problems rather than performance; nettops have sold plenty, and the A15 is perfectly capable of taking on that role.

OK maybe "basic internet" is harsh as they can do media consumption and stuff, but these days I think that qualifies as a 'basic' tasks thanks to all the acceleration and GPU. My point is that all the SOCs, Atom/ARM/whatever are still basic level performance wise, suitable for handheld devices but a bit slow for anything else. Anandtech did say the Chromebook began to struggle when they did certain tasks, this is what they said:

Technically the Chromebook can do a lot, but for anything other than browsing, YouTube and Google docs use I wouldn't get my hopes up. The heavier apps just don't run smoothly on the platform. Even web browsing isn't what I'd consider fast, but it's still acceptable. To put things in perspective, I got a performance warning trying to play Cut the Rope on the new Chromebook.

[ladcrooks]

the more players we have in this field - the better it is for the consumer.

[watercooled]

I think a lot of the strange results with the new Nexus products are more to do with software/optimisation over hardware; some of the results just aren't logical. Remember though, the benchmarks they fall short in are mainly JS benchmarks and as we've seen in the past, the browser can make as much difference as the CPU, and the likes of Sunspider don't seem to scale well beyond about two threads.

Quad Krait does seem a bit excessive for a phone ATM, although it's still nice to see them cracking on with upgrades, but I wonder if the cores are locked to the same clock in that SoC? I'd expect gating etc to allow single threaded performance equivalent to a dual core in the same power envelope, but again it could be more to do with software.

Tablets/Chroembooks etc basically limit what you can do with the interface and software, so lower performance isn't generally an issue. As for Cut the Rope, my ARM11 based phone runs it fine so that screams software problems IMO.

Competitions is, of course, good for the consumer, but I think people are a bit wary about Intel as they tend to not play fair and attempt to monopolise market segments, nether of which are positive for the consumer. ARM may be the standard for SoCs ATM, but the cost of the IP is relatively small and you already have competition between the several SoC MFRs, competing on things like price/features/custom cores/power/nodes/etc.