The ARM v.s. Intel Thing - Let's Discuss It

[dbcoopernz]

I think that roadmap is for a mobile phone part. The tablet SoC's were already scheduled for 2014, IIRC?

 

[Exophase]

I think that roadmap is for a mobile phone part. The tablet SoC's were already scheduled for 2014, IIRC?

The parts you linked are for tablets, netbooks, and nettops respectively. There's no smartphone chip here.

Intel released the Saltwell based tablet part (Clovertrail) months after the phone part (Medfield), so they could be planning on releasing Merrifield by late 2013 still. On the other hand, they also released the Saltwell netbook and nettop part (Cedartrail) months before Medfield, back in 2011. So there might be no connection. I would guess they strategically released Clovertrail specifically to coincide with the first Windows 8 tablets; I'm not even sure if it's meant to go into any other sort of device.

 

[shady28]

I thought this was a very revealing article on where Intel currently stands with Atom vs ARM.

No better platform to test on that Linux, probably as close to a level playing field as we'll find.

http://www.phoronix.com/scan.php?page=article&item=samsung_exynos5_dual&num=4

Quick Summary :

They used a Chromebook with the Exynos 5 dual core A15 based processor, compared to single and dual core Atoms, and a Core i3-330M as a frame of reference.

In most benchmarks the A15 absolutely destroyed the Atom D525, N270, and Z530 it was pitted against. In most cases it was 80%+ faster. In a few areas the Atom was faster - but only marginally. The i3 naturally won every benchmark.

One example: SciMark Monte Carlo Simulation
Exynos 5 (A15) scored 167.9 MFlops
Atom D525 scored 65 MFlops

I was also shocked to see how badly the Exynos 5 Dual-Core A15 ripped up the Tegra 3's Quad Core A9.

 

[Homeles]

Tegra 3 is a joke. I can't easily do research on this phone, but I'm pretty sure none of those Atoms tested are Intel's current best. Still, I'd imagine the A15 would blow it away. We're talking about a 4 year old microarchitecure here. AMD has/had better performance than them with Brazos. If AMD is beating you, you've got a problem.

 

[Exophase]

Tegra 3 is a joke. I can't easily do research on this phone, but I'm pretty sure none of those Atoms tested are Intel's current best. Still, I'd imagine the A15 would blow it away. We're talking about a 4 year old microarchitecure here. AMD has/had better performance than them with Brazos. If AMD is beating you, you've got a problem.

Newer Atom cores are big improvements in power consumptions but don't really improve performance. Cedar Trail in particular hasn't shown any improvement over the Pineview core (D525) tested.

In a few cases the D525 is showing a much higher improvement over than the 2x cores and frequency boost would suggest. And in these cases the N270 doesn't have a tangible lead over Z530 despite having much more bandwidth, so it can't be bandwidth limited (D525 was only paired with somewhat higher bandwidth memory than N270). My theory is that at least a part of the gain is due to D525 using x86-64 while the others use x86. On Atom this can sometimes be a big deal because those extra registers can be very important for scheduling on an in-order processor.

Thing is, Medfield and Clover Trail are both 32-bit only, and it looks like Bay Trail-T will be too, meaning the smartphone Silvermont will probably follow suit. Since they're OoO they might be better suited to it, but programs can still be penalized if they're compiled for the current Atoms instead..

The part about Brazos is a shallow argument. Intel and AMD were targeting completely different power envelopes. Of course peak performance is going to be different. Atom still has better perf/W than Bobcat had on 45nm vs 40nm, and much better now that it's on 32nm. Jaguar looks like it brings some nice progress here, though.

 

[shady28]

Tegra 3 is a joke. I can't easily do research on this phone, but I'm pretty sure none of those Atoms tested are Intel's current best. Still, I'd imagine the A15 would blow it away. We're talking about a 4 year old microarchitecure here. AMD has/had better performance than them with Brazos. If AMD is beating you, you've got a problem.

The newest D2700 Dual-Core Atom is only about 10% faster than the D525 (The D525 was reviewed).

Intel would need to roughly double the performance of net-top Atoms to become competitive with the A15 from what I saw in these benchmarks.

I also looked up Intels phone "Atom" - it's significantly slower than the D525 / D2700.

Intel still has a long way to go to be competitive against ARM.

 

[Arachnotronic]

Exophase

It is likely that Intel will announce the 22nm SoCs at MWC in 2013, but they won't show up in designs until 2014.

Keep in mind that Krait was announced in February 2011, but we didn't see it in phones until 2012. Z2580 was announced last year but will show up in phones this year.

"Avoton" based on "Silvermont" will be a 2013 part, though. Intel is clearly more afraid of losing market share in servers than it is worried about taking smartphone/tablet share in the near-term.

 

[shady28]

I would not count on Intel's new mobile CPU being a performer.

The Medfield CPU benched about 30% slower than an OMAP 4 4430. This is the chip found in phones like the Droid 3 and Droid Bionic from over a year ago.

http://news.softpedia.com/news/Inte...ield-with-Disastrous-Performance-293186.shtml

"The guys at gsmarena.com have already got their hands on the new Motorola RAZR i that uses the new 2 GHz Medfield and the test results are painting an ugly picture.

Compared with a rather new generation SnapDragon S4 running at just 1.5 GHz, the 2 GHz Medfield reportedly needs 202% S4’s benchmark time to complete the test.

There is a single area where Intel’s software team has managed a clear success and that’s the Sun Spider benchmark where the 2 GHz Medfield is the absolute best.

Other than that, any other benchmark is only going to reveal just how weak the computing performance is "

 

[Homeles]

The part about Brazos is a shallow argument. Intel and AMD were targeting completely different power envelopes. Of course peak performance is going to be different. Atom still has better perf/W than Bobcat had on 45nm vs 40nm, and much better now that it's on 32nm. Jaguar looks like it brings some nice progress here, though.

Of course they have different targets, but the fact that AMD had any success against Atom at all just goes to show that Atom is really old. Intel's 45nm HKMG process blows away TSMC's 40nm process, so the fact that AMD bested them gives my comparision quite a bit of substance.

Intel is currently at a process disadvantage against ARM, but next year that will change. And we'll see a newer architecture as well. When 22nm (with low power focused FinFETs to boot) Silvermont arrives, you'll see the tables turn in the opposite direction. I don't think it'll be as dramatic though.

 

[Idontcare]

Of course they have different targets, but the fact that AMD had any success against Atom at all just goes to show that Atom is really old. Intel's 45nm HKMG process blows away TSMC's 40nm process, so the fact that AMD bested them gives my comparision quite a bit of substance.

Very true, an excellent demonstration of microarchitecture triumphing over process technology and thriving despite the inherent weaknesses.

Dothan accomplished the same on Intel's fabled 90nm node, making do with what Prescott could not.

It is too bad that AMD didn't find a magic combination for GloFo's 32nm. Maybe they will with its 28nm?

 

[HypX]

Of course they have different targets, but the fact that AMD had any success against Atom at all just goes to show that Atom is really old. Intel's 45nm HKMG process blows away TSMC's 40nm process, so the fact that AMD bested them gives my comparision quite a bit of substance.

Intel is currently at a process disadvantage against ARM, but next year that will change. And we'll see a newer architecture as well. When 22nm (with low power focused FinFETs to boot) Silvermont arrives, you'll see the tables turn in the opposite direction. I don't think it'll be as dramatic though.

By next year ARM will be on 22/20nm, so I don't see how Intel will gain any sort of advantage though.

 

[Fox5]

I'll spare you what I think of Charlie's info

Intel does have a better compiler with ICC. Problem is that I don't think it's going to be getting a lot of usage on Windows 8 Metro apps. And the usage for Android is going to be close to zero.

The impact on Javascript engine JIT quality is also going to be zero.

I think I once heard that interpreters are almost entirely bandwidth limited, so the high memory bandwidth, small instruction size, and fast caches of Atom may be what contribute to their advantage in javascript.

 

[beginner99]

I thought this was a very revealing article on where Intel currently stands with Atom vs ARM.

No better platform to test on that Linux, probably as close to a level playing field as we'll find.

http://www.phoronix.com/scan.php?page=article&item=samsung_exynos5_dual&num=4

Quick Summary :

They used a Chromebook with the Exynos 5 dual core A15 based processor, compared to single and dual core Atoms, and a Core i3-330M as a frame of reference.

In most benchmarks the A15 absolutely destroyed the Atom D525, N270, and Z530 it was pitted against. In most cases it was 80%+ faster. In a few areas the Atom was faster - but only marginally. The i3 naturally won every benchmark.

One example: SciMark Monte Carlo Simulation
Exynos 5 (A15) scored 167.9 MFlops
Atom D525 scored 65 MFlops

I was also shocked to see how badly the Exynos 5 Dual-Core A15 ripped up the Tegra 3's Quad Core A9.

So bascially they compared the newest and greated uArch (A15) on the newest process to a 5 year old uArch on a 5 year old process? How can you even take something like this seriously? (without even checking the actually benchmark and that site always seems to be biased towards the underdogs)

Once Atom gets a new uArch and is actually made on the newest process node...then we can do a fair comparison.

Also that A15 probably uses a lot more power than previous ARM designs. Hell, todays smartphones suck in that regard. My basic 5 year old phone with a very, very used battery still lasts a lot longer than any smartphone on one charge.

 

[Arachnotronic]

By next year ARM will be on 22/20nm, so I don't see how Intel will gain any sort of advantage though.

Unlikely. TSMC/GloFo are just ramping 28nm. Qualcomm is not even using the HKMG version of 28nm.

Samsung's A15 design is on 32nm.

 

[krumme]

Yeaa and Qualcomm have the same market cap as Intel.

How many percent of the sold phones is even on 28/32 nm as of now?

 

[HypX]

Unlikely. TSMC/GloFo are just ramping 28nm. Qualcomm is not even using the HKMG version of 28nm.

Samsung's A15 design is on 32nm.

TSMC missing the 20nm node for all of 2014 would be quite the disaster, and I don't think anyone is predicting that.

Unless of course you forgot what year this is.

 

[ShintaiDK]

Yeaa and Qualcomm have the same market cap as Intel.

How many percent of the sold phones is even on 28/32 nm as of now?

1-2%?

 

[sontin]

So bascially they compared the newest and greated uArch (A15) on the newest process to a 5 year old uArch on a 5 year old process? How can you even take something like this seriously? (without even checking the actually benchmark and that site always seems to be biased towards the underdogs)

Once Atom gets a new uArch and is actually made on the newest process node...then we can do a fair comparison.

Also like Clover Trail vs. Tegra 3?
Intel needs to get their time to market together. They can't come to the market after one year. They seem on track for the smartphone market but in the tablet and low-end they are to far behind the competition.

 

[Nemesis 1]

In these Atom topics we have slides from intel these slides look like past INTEL slides . We have post with many slides the intel ones say intel on them and the sheme remains the same as past graph. But suddenly we have these new graph. that are not normal intel colors nor is intel on any of these graphs . Lately there has been a flood of these none issued intel graphs . I suspect the armies of hell are at work here . LOL these do not look like intel graphs .

Google merrifield and the info on you find pay attention to what a real intel graph looks like when talking roadmaps . Is it possiable intel slipped . Yes it is . I will tell you a thing if intel did slip and they don't report it to the public I will switch to ARM . IF intel changed its roadmaps and publicly don't say so . I turn my back on them . Anyone can slip but owning up to it is a big deal . If intel doesn't Give a report on atom at CES . Arm will work for me. If I turn on our online hype machine we will do intel great harm . I will do it to.

 

[Nemesis 1]

I would not count on Intel's new mobile CPU being a performer.

The Medfield CPU benched about 30% slower than an OMAP 4 4430. This is the chip found in phones like the Droid 3 and Droid Bionic from over a year ago.

http://news.softpedia.com/news/Inte...ield-with-Disastrous-Performance-293186.shtml

"The guys at gsmarena.com have already got their hands on the new Motorola RAZR i that uses the new 2 GHz Medfield and the test results are painting an ugly picture.

Compared with a rather new generation SnapDragon S4 running at just 1.5 GHz, the 2 GHz Medfield reportedly needs 202% S4’s benchmark time to complete the test.

There is a single area where Intel’s software team has managed a clear success and that’s the Sun Spider benchmark where the 2 GHz Medfield is the absolute best.

Other than that, any other benchmark is only going to reveal just how weak the computing performance is "

Thats the most bias article I have ever seen very unperfessional . But that normal for them . Why is it that AT review shows Atom phone in a good light , Why is it that aton dual core is faster on tablets . Why did you post the worse bias article you could find . Why? Why? Why?

 

[Nemesis 1]

If it is "ARM vs Intel" within the context of "who will end up dominating the smartphone/tablet (non-PC, so no laptops) market" (and foregoing all technical "which arch is better" discussions), the only real limitations I have in mind are those of Intel's own making.

Bluntly, they are not approaching every fight with both hands untied and ready to do "whatever it takes". We can imagine that with the tech at their disposal and the billions in budget, they can pretty much kick ass in every related field. And this would be very true, if Intel had no reservations about dominating every related field.

Unfortunately, they are hamstrung by the gross margin requirements. It is completely within the boundaries of tech-feasibility and market-feasibility for Intel to produce a chip that will go into every smartphone/tablet, be superior in all tech metrics, and be cheaper than those of any competitor. Some of us would think of that as an absolute good thing for Intel, an obvious course of action. But for Intel, if it would end up severely damaging their their accustomed >60% gross margins, it simply wouldn't be worth the investment. Their history shows this. The board wants it. CEO's job depend on it. It is not a cycle that someone in Intel can just break by saying "Screw the gross margins! We must dominate smartphones no matter what!".

This is the short and quick of it. If the ARM / smartphone / tablet makers make it an industry with margins at 30% (completely pulled out of air, just for the sake of exposition), Intel may simply leave the market, as long as they feel that the company is in no danger of extinction, if it ends up affecting their overall gross margins significantly in a negative way. If it were not for this particular limitation that has Intel quite literally fighting with one hand tied behind its back, I don't see how ARM (the design company, not the arch) can actually compete against Intel if Intel wanted to take ARM's cake.

Intel margins . LOL at the ignorance . Take intel break it up into fabs and Design house . Intel buys from x-intel fabs . The fabs need at least 30% margins to exist. Intel the design house needs at least 30% margines to exist . As a whole that would be 60% broken up that becomes less than 30% Intel has other businesses also that add to the bottom line .
The reality is intels 60% margines are not that good . Look at the money Intel poors into the US economy . We need intel to fall . So America can be world class 3rd world economy

 

[simboss]

1-2%?

Iphone 5, galaxy S III, are on 32nm, most snapdragon S4 are on 28nm.
The only chip not yet on this node is Tegra3 which in phones is not sellig that much.

It's pretty difficult to get a real figure, but I would bet at least 30% of smartphones are on <=32nm node.

Atom marketshare is probably lower than 1% though :hmm:

 

[Idontcare]

Welcome to the forums simboss :thumbsup:

 

[Nemesis 1]

...which was what I meant by Intel hitting a frequency wall.

The question is if ARM will hit that wall to at around 4 GHz. Being a RISC CPU it may be able to reach higher frequencies than the Intel CISC CPUs.

Thats not a wall . My Sb and IB cpus easily overclock to 4.2 ghz at stock voltage . The TDP does not go over 125 . Intel doesn't want this they want less tdp . Look at haswell higher TDP than ib . People go Oh my thats not good . Did intel add something to the die area that would cause this??? Yes they did . System total power usage will be lower for haswell . You can't play poker unless you understand the read of both the cards showing or the person setting across the table . If you can't do these to things . Playing poker if you win a hand its pure luck no skill . At the end of the night your a loser.

 

[ShintaiDK]

Iphone 5, galaxy S III, are on 32nm, most snapdragon S4 are on 28nm.
The only chip not yet on this node is Tegra3 which in phones is not sellig that much.

It's pretty difficult to get a real figure, but I would bet at least 30% of smartphones are on <=32nm node.

Atom marketshare is probably lower than 1% though :hmm:

Samsung S3 Mini for example is 45nm. Its essentially only the top end phones that are 28/32nm.

Plus non smartphones still outsell smartphones with a huge margin.
I think for Samsung the ratio was 360/60 million for 2012 with a total of 420 shipped..