State of mobile chip (Qualcomm Kryo, ARM Cortex A72, Intel Goldmont)

[cbn]

The Android OS update model hurts the industry even more than CDMA if we are talking about SoC (or phone vendor) competition.

At least with x86 (being a standardized platform) the end user could do the update themselves rather than waiting for the ARM device secific build to come. This especially for low end phones on prepaid where they never get updated (The OS the device ships with is all it will ever get).

 

[kpkp]

After measuring the Kirin 950's power efficiency I'm pretty certain that Kyro will have a very hard time competing against A72, and Mongoose is supposedly even better than that

Isn't this really similar to the claims ARM made for the A72?

 

[cbn]

But I doubt they will every get it due to mentality. A sell is a sell, "I managed to cheat you with my cheap crap that doesn't' work, your problem". They for sure don't have a service-oriented culture.

I doubt they will get it too, but it isn't due to the mentality. It is due to the fact that the margins on these devices are very small and there isn't really some sort of incremental future revenue from sold devices that make it worth keeping them updated.

What they are missing is that it would be worth losing money on a single halo device that is a Nexus copy just for all the positive press it would bring for their other devices.

Yes, It make sense the extra cost of updating is a problem on these low end phones and the way they are sold. And the cheaper the phone gets the less likely it is to receive attention.

 

[Face2Face]

Interesting article...if you like GB scores.

http://www.techgrapple.com/helio-x2...s-8890-vs-apple-a9-vs-apple-a9x-vs-kirin-950/

 

[poofyhairguy]

At least with x86 (being a standardized platform) the end user could do the update themselves rather than waiting for the ARM device specific build to come.

Actually with Qualcomm devices you can do that. One good thing about their market dominance is that they are pretty good on open source matters (compared to many other SoC makers) and therefore it is usually not that hard to put together your own ROM for a device if you know how to do that sort of thing. It is much better than with Exynos devices where a lot of those drivers in CM ROMs have to be reverse engineered.

 

[nismotigerwvu]

If we were more optimistic about upcoming chips would you be willing to make up your mind about them now?

Of course not, I already said I'll make up my mind when we have production hardware in our hands. I'm just not willing to dismiss something we really don't know about.

 

[IntelUser2000]

The writing is on the wall for in-house chips. I think its a natural progression for end of Moore's Law. The ever small improvements you can pull out means there's not much gain doing in-house designs anymore.

The worst is arguably Goldmont. They've suffered from 14nm delays and Intel management has completely shut their mouth about it.

The real issue is how Intel treated Atom cores.

Silvermont = 14nm A15 class
Goldmont = 14nm A57 class, I reckon. 30% improvement per clock would be good, but nothing stellar. And ARM guys will have 14nm too, but with A72 class cores.

Really what Intel should have aimed for was a year or two earlier than they originally planned. We should have had Goldmont IPC with Silvermont. Of course they seem to be afraid to cannibalize their Core line, which ironically going to turn out to kill both Atom and Core.

 

[poofyhairguy]

Really what Intel should have aimed for was a year or two earlier than they originally planned. We should have had Goldmont IPC with Silvermont. Of course they seem to be afraid to cannibalize their Core line, which ironically going to turn out to kill both Atom and Core.

Amen to all of that. Shows you what happens if you try to pull punches for too long.

 

[mrmt]

Really what Intel should have aimed for was a year or two earlier than they originally planned. We should have had Goldmont IPC with Silvermont. Of course they seem to be afraid to cannibalize their Core line, which ironically going to turn out to kill both Atom and Core.

Intel does not have a cost-optimized node to take on TSMC, so even if they rushed Goldmont it wouldn't reach the market with the right cost structure. And given that 10nm seems to be suffering from a worst version of the 14nm malaise, I wouldn't expect much of Intel mobile business.

The only way I can think of Intel becoming a relevant mobile player is to cost-optimize the hell out of the Goldmont cores and use TSMC as a foundry.

 

[StrangerGuy]

The writing is on the wall for in-house chips. I think its a natural progression for end of Moore's Law. The ever small improvements you can pull out means there's not much gain doing in-house designs anymore.

The worst is arguably Goldmont. They've suffered from 14nm delays and Intel management has completely shut their mouth about it.

The real issue is how Intel treated Atom cores.

Silvermont = 14nm A15 class
Goldmont = 14nm A57 class, I reckon. 30% improvement per clock would be good, but nothing stellar. And ARM guys will have 14nm too, but with A72 class cores.

Really what Intel should have aimed for was a year or two earlier than they originally planned. We should have had Goldmont IPC with Silvermont. Of course they seem to be afraid to cannibalize their Core line, which ironically going to turn out to kill both Atom and Core.

That's why Intel is shifting so hard to IoTs...Because there's is simply no way ARM would beat them there in an even more cost sensitive market, right?

 

[icrf]

Please forgive the thread necro, but I've looked around and can't find any decent info on Goldmont's supposed release date. The best I've found is some mention of 2H 2016, pushed back from early 2016, pushed back from mid-2015. All that info was pretty dated, too. Anyone heard anything else?

Given the imminent release of Kryo and A72, Goldmont is pretty much a non-starter. I'm mostly just really interested in a non-Pro Surface 4, and I don't see any reason that would come out without Goldmont.

 

[Arachnotronic]

Please forgive the thread necro, but I've looked around and can't find any decent info on Goldmont's supposed release date. The best I've found is some mention of 2H 2016, pushed back from early 2016, pushed back from mid-2015. All that info was pretty dated, too. Anyone heard anything else?

Given the imminent release of Kryo and A72, Goldmont is pretty much a non-starter. I'm mostly just really interested in a non-Pro Surface 4, and I don't see any reason that would come out without Goldmont.

Look for it in 2H'16, I'm not holding my breath for 1H'16.

 

[Sweepr]

Latest info (3 months ago) mentions late Q2 launch for desktop Apollo Lake. I agree with Arachnotronic though, looks like a H2-2016 (likely Q3) launch to me.

 

[icrf]

Thanks for the info. I just thought Intel was usually a bit more forthcoming in the past about these kinds of things, and I hadn't seen anything official.

 

[ShintaiDK]

Hopefully Goldmont will be the last. Cant see any usage with Xeon D and Core M unless its for a settop box or NAS.

 

[icrf]

You're probably right, but Intel is still pretty proud of its Core CPUs and needs to keep up with budget devices, too. While I'm confident Intel can hit A72 power and performance targets with Core, does anyone really think they'll get there with a competitive price, too? That's the niche Atom fills, even if Goldmont won't be quite up to A72 standards, it still has the option of running full legacy software stacks that can help make up for it in some cases.

 

[ShintaiDK]

Core is way above A72. No ARM chips can compete with Core chips.

 

[icrf]

Absolute performance? Sure. At similar power levels, is the different that obscene? I thought the iPad Pro review showed it was surprisingly close: http://www.anandtech.com/show/9766/the-apple-ipad-pro-review/4

Sure, being 34% slower is still a lot, and the A9X outclasses an A72, but things are tightening up compared to what it used to be.

 

[Nothingness]

Absolute performance? Sure. At similar power levels, is the different that obscene? I thought the iPad Pro review showed it was surprisingly close: http://www.anandtech.com/show/9766/the-apple-ipad-pro-review/4

Sure, being 34% slower is still a lot, and the A9X outclasses an A72, but things are tightening up compared to what it used to be.

It's less than 34%. AT review was biased by icc usage and it seems Johan from AT is looking into using gcc to get more fair results: http://www.realworldtech.com/forum/?threadid=156729&curpostid=157077

 

[Arachnotronic]

It's less than 34%. AT review was biased by icc usage and it seems Johan from AT is looking into using gcc to get more fair results: http://www.realworldtech.com/forum/?threadid=156729&curpostid=157077

Although I don't buy the VISC hype/BS, I do think the A9X/Skylake comparison here is interesting:

 

[Exophase]

Absolute performance? Sure. At similar power levels, is the different that obscene? I thought the iPad Pro review showed it was surprisingly close: http://www.anandtech.com/show/9766/the-apple-ipad-pro-review/4

Sure, being 34% slower is still a lot, and the A9X outclasses an A72, but things are tightening up compared to what it used to be.

I'm convinced at this point that that libquantum score is completely unfair as a CPU comparison (while heavily influencing the average) and the others may be verging on that as well. Anandtech is showing some interest in configuring a pure GCC comparison instead, hopefully we get to see that..

And at the same time, we don't really have power consumption comparisons for these platforms.

 

[Arachnotronic]

Intel does not have a cost-optimized node to take on TSMC, so even if they rushed Goldmont it wouldn't reach the market with the right cost structure. And given that 10nm seems to be suffering from a worst version of the 14nm malaise, I wouldn't expect much of Intel mobile business.

The only way I can think of Intel becoming a relevant mobile player is to cost-optimize the hell out of the Goldmont cores and use TSMC as a foundry.

Or just use ARM cores, ARM GPU, etc. and use TSMC as a foundry

Not glamorous, but it'd be better for Intel to grab some of that revenue for itself rather than letting others have it...

 

[dark zero]

He Intel is totally screwed on mobile. And now they think that moving to IoT is the correct move... They have hardware... They lacks software...
Just like MS, they chose the wrong horse... Intel should had entered with only Windows Mobile and not Android.

 

[Lepton87]

Why isn't Mongoose included in the thread? Exynos 8890 powered Galaxy G7 is about to be released and yet there's no discussion about that SOC, especially its CPU portion. How are those cores going to compare to the A72 or the Kryo cores? Faster or slower? I currently own G6 and I'm thinking if it would be worth it to upgrade. MT performance is fine but ST could be better. If it weren't for that pesky walled garden that apple created around its iOS I would have chosen A9 much better ST performance and maybe even comparable MT performance.

 

[Systems analyst]

Qualcomm are dropping out of full-custom designed cores for mobile.
Source: analyst's question (not answer) at ARM 2015Q4 results meeting.

ARM now have two customers for Artemis POP IP, the follow-on to A72; my guess is Mediatek first, and Qualcomm second, following the above news.

It looks to me as though Apple, Samsung and Huawei will rule in the top-end mobile SoC space.

Intel are left with mid-range SoCs, competing with A72 this year, and Artemis next year. Intel would compete with A35 at the bottom end. There must be far more ARM SoC design engineers in the world, than engineers using Intel in mobile; software also is mostly ARM based. ARM have over 99% of mobile phones SoCs, they say.

What they have done in mobile phones will be attempted in laptops; remember, Intel say that they are no longer a PC company.