Why Atom sucks

[Intel17]

People like to blame X86 for why Atom sucks compared to newer, more modern micro-architectures, but the plain and simple truth is: the underlying design is a kind of a mess and likely due to the fact that it was limited by the constraints of the 45nm process on which it was designed. I dug up this really helpful CPU optimization guide (http://www.agner.org/optimize/microarchitecture.pdf renowned programmer/optimizer Agner Fog, in which he does a really nice job summarizing why Atom is such a POS:

- In-order execution is a bottleneck
- The instruction fetch rate is really low (and you thought Bulldozer's shared fetch was a bad idea!)
- Memory access is limited to one read or one write per clock (can't do both)
- Maximum throughput of 2 instructions/cycle is only realistic if the code is specially optimized for Atom
- Memory latency is HUGE for floating point instructions compared to integer once since the memory unit is only connected to the integer cluster (ever wonder why Atom gets thrashed in the FPU part of Geekbench?)

The good news is that "Silvermont", with a much bigger transistor budget (thanks, 22nm!), probably fixes a lot of these bottlenecks that were introduced courtesy of the 45nm node. But the bad news is that today's Atom still sucks

[itsmydamnation]

a lot of those are for reduced power as well. There is always a trade off between how fast you do work and how much power you use to do work. This equation is very fluid and changes between nodes/process/etc.

As a betting man im going to bet that silvermounts OOOE wont be as flexible as jaguar and thats purely based on Target power consumption, you will likely see lots of little trade offs like this in a direct comparison between silvermount and jaguar.

That said i dont think silvermount will get very far in its target market and that has nothing to do with how it will perform. what intel is selling is trending in the opposite direction to what the phone/tablet makers are heading in. Traditional PC OEM's will likely be much happier with the intel ATOM model so i would look for tablets from them.

[Intel17]

Originally Posted by itsmydamnation

a lot of those are for reduced power as well. There is always a trade off between how fast you do work and how much power you use to do work. This equation is very fluid and changes between nodes/process/etc.

As a betting man im going to bet that silvermounts OOOE wont be as flexible as jaguar and thats purely based on Target power consumption, you will likely see lots of little trade offs like this in a direct comparison between silvermount and jaguar.

Problem is, when 32nm rolled around, Intel didn't bother doing a "tock". If it had done so, it would probably be much more easily fighting the ARM guys with the "Medfield" SoC. Actually, given how close the current Atom comes, it would probably have clear CPU leadership.

[itsmydamnation]

Originally Posted by Intel17

Problem is, when 32nm rolled around, Intel didn't bother doing a "tock". If it had done so, it would probably be much more easily fighting the ARM guys with the "Medfield" SoC. Actually, given how close the current Atom comes, it would probably have clear CPU leadership.

I dont really think so, the problem for intel is consumers dont care about CPU performance. How many people upgraded a 3GS to a 4 to 4s to a 5 because of CPU? I know none.

The next problem for them is sure they might have been able to make a more power efficient core but on G3/LTE carrier network design is the single largest factor determining standby life time. Poor cellular design and it doesn't matter how good your power saving features are your going to get sub 10 hours standby.

if intel want to be successful they will have to go to a level of openness with the phone manufactures that they never have come close to, they need to out ARM, ARM. Phone manufactures want to be in control of there component chains and products.

[meloz]

I think what bothers people is not that Atom "sucked" when it was launched (it was tolerable / adequate for intended purposes back then), but that Intel immediately went into lazy profiteering mode and refused to upgrade Atom in a Tick-Tock manner as they do with their mainstream x86-64 parts. They treated it terribly in a step-child manner, so much potential has been wasted.

So here we are in year 2013, and Atom still sucks. Ugh. When in reality it could have been the primary choice for all non-Apple tablets and smartphones, if only Intel incrementally improved it every year.

I think the failure (laziness) with Atom, and Intel's dithering on producing a decent in-house SSD controller are two huge lowpoints of Otellini's reign as Intel's CEO. He got lucky to lead Intel in this era. People will point to record profits and marketshare, but my point is that they could have made even bigger profits while still delighting enthusiasts had they nurtured Atom properly.

Oh well. Now we all await Silvermont, although frankly to me it feels Airmont 2014 is when the 'party' proper will start. Silvermont is more like an attempt to put a derailed train back on track. Then comes the Tick-Tock.

[Intel17]

Originally Posted by meloz

I think what bothers people is not that Atom "sucked" when it was launched (it was tolerable / adequate for intended purposes back then), but that Intel immediately went into lazy profiteering mode and refused to upgrade Atom in a Tick-Tock manner as they do with their mainstream x86-64 parts. They treated it terribly in a step-child manner, so much potential has been wasted.

So here we are in year 2013, and Atom still sucks. Ugh. When in reality it could have been the primary choice for all non-Apple tablets and smartphones, if only Intel incrementally improved it every year.

I think the failure (laziness) with Atom, and Intel's dithering on producing a decent in-house SSD controller are two huge lowpoints of Otellini's reign as Intel's CEO. He got lucky to lead Intel in this era. People will point to record profits and marketshare, but my point is that they could have made even bigger profits while still delighting enthusiasts had they nurtured Atom properly.

Oh well. Now we all await Silvermont, although frankly to me it feels Airmont 2014 is when the 'party' proper will start. Silvermont is more like an attempt to put a derailed train back on track. Then comes the Tick-Tock.

Hey, at least when Intel finally did do an in-house SSD controller, it rocked!

Let's hope for that out of Silvermont :-)

Although really...they did mess up pretty hard by just not putting in the resources to develop it.

[Arkaign]

Originally Posted by meloz

I think what bothers people is not that Atom "sucked" when it was launched (it was tolerable / adequate for intended purposes back then), but that Intel immediately went into lazy profiteering mode and refused to upgrade Atom in a Tick-Tock manner as they do with their mainstream x86-64 parts. They treated it terribly in a step-child manner, so much potential has been wasted.

So here we are in year 2013, and Atom still sucks. Ugh. When in reality it could have been the primary choice for all non-Apple tablets and smartphones, if only Intel incrementally improved it every year.

I think the failure (laziness) with Atom, and Intel's dithering on producing a decent in-house SSD controller are two huge lowpoints of Otellini's reign as Intel's CEO. He got lucky to lead Intel in this era. People will point to record profits and marketshare, but my point is that they could have made even bigger profits while still delighting enthusiasts had they nurtured Atom properly.

Oh well. Now we all await Silvermont, although frankly to me it feels Airmont 2014 is when the 'party' proper will start. Silvermont is more like an attempt to put a derailed train back on track. Then comes the Tick-Tock.

Wow, epic lurker victory!

Less than 100 posts in 4 years, but this is absolutely spot-on.

Hopefully Intel is well into the process of fixing exactly the problems you're spotlighting.

[beginner99]

I don't know but this POS still manages not to clearly lose the the newest and greatest A15 in terms of performance/watt.

As meloz said the crappy part is that Intel didn't evolve it in tick-tock fashion because then the A15 would now probably look pretty pathetic in comparison. I don't get the ARM hype if such and outdated uArch (Atom) on an older process is still competitive.

The only problem I see for Intel is price and platform (OS).

[zebrax2]

Originally Posted by meloz

....
So here we are in year 2013, and Atom still sucks. Ugh. When in reality it could have been the primary choice for all non-Apple tablets and smartphones, if only Intel incrementally improved it every year.
....

I fully agree with this statement. While it may be true that iPhone users doesn't really upgrade because of the upgrade in CPU performance the same can't be said with android where the fight for dominance lies on being the best

[Lonyo]

Why Atom sucks:
It's a 5 year old architecture that's never been updated, and even 5 years ago it was very basic due to maximising power efficiency instead of performance.

That's why it sucks. 5 years of no progress.

[Haserath]

Originally Posted by Lonyo

Why Atom sucks:
It's a 5 year old architecture that's never been updated, and even 5 years ago it was very basic due to maximising power efficiency instead of performance.

That's why it sucks. 5 years of no progress.

Not power efficiency. They tried to maximize profit with a tiny die and really choked it too much.

Atom sucked at everything. AMD at least scored on that one a bit, though they could've probably done better.

[ShintaiDK]

Originally Posted by Haserath

Not power efficiency. They tried to maximize profit with a tiny die and really choked it too much.

Atom sucked at everything. AMD at least scored on that one a bit, though they could've probably done better.

Ironicly enough, AMD basicly did the exact same thing and forgot to shrink Brazos to 28nm. And both of them wil update their respective uarchs in 2013.

[NTMBK]

Originally Posted by ShintaiDK

Ironicly enough, AMD basicly did the exact same thing and forgot to shrink Brazos to 28nm. And both of them wil update their respective uarchs in 2013.

The Brazos shrink was on the roadmap, but apparently got canned due to 28nm low-power process being late. They would have had to launch it later than planned, and give it only a narrow window of opportunity before they launched Jaguar- unless they pushed Jaguar back, which would have been a bad move.

It'll be interesting to see what happens. Atom on 22nm will have ~4x the transistor budget of the original Atom on 45nm, and I refuse to believe that Intel won't make the most of it. I'm expecting a big leap.

[Torn Mind]

Just how large were the margins being made on an Atom compared to a budget Core architecture CPUs? I think Intel let Atom rot because they didn't want it eating into i3s, Pentiums, and Celeron chips. They still made a major miscalculation in projecting the ubiquity of small, mobile computers now though.

[jolancer]

everything said sounds true. however im on an Atom and it doesnt suck

agree with last poster, if this was any faster with this kind of power efficency it would make a lota other things obsolete. Im all about efficency so it works well for me.

[ShintaiDK]

Originally Posted by Torn Mind

Just how large were the margins being made on an Atom compared to a budget Core architecture CPUs? I think Intel let Atom rot because they didn't want it eating into i3s, Pentiums, and Celeron chips. They still made a major miscalculation in projecting the ubiquity of small, mobile computers now though.

Margins on Atoms is on the level as Celerons and Pentiums. Atoms being cheaper to produce gives it the win.

The whole margin thing is just a myth. The point is that Atoms serves a purpose(power consumption) thats still unreachable for Core.

[HeXen]

I had an Atom on an MSI Wind netbook and it ran just fine. I played most every console emulator with ease, Titans Quest ran great, no problems with the Windows apps i used on it.

[happysmiles]

THERE
IS
A
MAJOR
ATOM
UPDATE
COMING
THIS
YEAR

complaining about a 4+ year old design seriously.

[jpiniero]

Originally Posted by ShintaiDK

Margins on Atoms is on the level as Celerons and Pentiums. Atoms being cheaper to produce gives it the win.

I seriously doubt that. If that was the case, they would have pushed Atom more instead of letting it rot.

[Lonyo]

Originally Posted by Torn Mind

Just how large were the margins being made on an Atom compared to a budget Core architecture CPUs? I think Intel let Atom rot because they didn't want it eating into i3s, Pentiums, and Celeron chips. They still made a major miscalculation in projecting the ubiquity of small, mobile computers now though.

Margins are %, even if the margin is the same, it doesn't mean the actual contribution is.
Sell 50 Celerons at $50 with 50% margin, you have $25*50 = $1250
Sell 50 Atoms at $40 with 50% margin, you have $20*50 = $1000

Why push the cheaper product even if it has the same margins?

[ShintaiDK]

Originally Posted by jpiniero

I seriously doubt that. If that was the case, they would have pushed Atom more instead of letting it rot.

The doubt can be cleared by taking a cruise on ark.intel.com.

You cant really push Atom harder if the segment aint there. Atom first slowly started the smartphone venture (And we all know the 4-5 years development cycle.). Besides that it only served a purpose in NAS and Netbooks to put it shortly. And people decided they wanted laptops instead of netbooks, despite the longer batterylife in netbooks. Tablets actually suffers the same situation as netbooks. So the endgame will be both laptops and smartphones. Atom can only apply for one of those.

[ShintaiDK]

Originally Posted by Lonyo

Margins are %, even if the margin is the same, it doesn't mean the actual contribution is.
Sell 50 Celerons at $50 with 50% margin, you have $25*50 = $1250
Sell 50 Atoms at $40 with 50% margin, you have $20*50 = $1000

Why push the cheaper product even if it has the same margins?

Most atoms aint cheaper than celerons at all. Some are even more expensive than pentiums. However they serve a completely different power consumption segment.

[jpiniero]

Tablets actually suffers the same situation as netbooks. So the endgame will be both laptops and smartphones. Atom can only apply for one of those.

Oh tablets will be around, people love the form factor. There does seem to be a trend for smartphones to get bigger to the point where having a phone and a tablet might not make sense however.

[Intel17]

Keep in mind that Intel does not want to lose the micro-server space to any ARM player. At the launch of the Atom S1200, Diane Bryant (who runs that server group) said,

"We will not hold back the performance of any of our product lines"

[Exophase]

I highly value and respect the reverse engineering Agner does but sometimes his conclusions about performance implications seem a little off to me. For instance he thinks HT often offers little benefit or even slowdown in Atom, when in the real world that never seems to be the case. He's also said weird things about how programs would often hit Bulldozer's restriction about not being able to access two cache banks in the same cycle if they're 256 bytes apart, seems to me that'd almost never be a problem..

I also don't think the 8-bytes/cycle fetch is particularly unbalanced for the rest of the processor, especially when you consider the most important Atoms are x86 only (no REX prefix) and don't have SSE4+, much less AVX. One read or write per cycle is par for the course for this complexity/power consumption class - if you move up to read + write like Bobcat and Cortex-A15 you pay for it. It's nice to have though.

I actually think Atom isn't that bad, not as bad as I originally thought anyway. But what I do think is it needs x86-64, preferably even "x32", since 8 registers really holds it back. Try coding some ASM optimized for it sometime, I think you'll see what I mean. Some other ISA features hurt more for a narrow in-order processor, like having to deal with more moves, but the full read-modify-write pipeline is nice. Agner is right though, software has to be well optimized specifically for Atom. I wonder how much that is the case for software currently executed..

As for all the stuff about how it's a 5 year old core with no changes, Saltwell at least brought about some changes (outside of being ported to 32nm of course). What we don't really know is how much Intel has worked at changes that could reduce power consumption but not necessarily be visible as functional differences. Given the huge advancements they've made here I wouldn't be so sure that it was down to the process improvement alone.