Intel Airmont: First benchmarks

[Fjodor2001]

The First Benchmarks of Intel Corporation’s 14-Nanometer Airmont Atom Don’t Look Promising

http://www.fool.com/investing/gener...benchmarks-of-intel-corporations-14-nano.aspx

Interestingly enough, Cherry Trail (based on the Airmont core) showed up in the Geekbench database. Let's take a look at how the Cherry Trail chip, which the benchmark claims is running at 1.60GHz, compares with a quad-core Silvermont running at 2.33GHz:

According to the test, the Airmont core in this particular platform is 2.3% faster on a per-core basis and about 10.6% faster when all four cores are being used.

Now, leaks have suggested that the Airmont CPU core as part of the Cherry Trail platform will run at 2.7GHz maximum frequency. So, this might initially lead one to believe that at "full" performance, the Cherry Trail will perform significantly better. However, if we were to scale the results of the "1.6GHz" Cherry Trail to 2.7GHz, we'd get about 1500 for the single-core score and 4915 for the multicore score.

This would be too large of an improvement to expect generation on generation (that multicore score is roughly in-line with an Ultrabook-oriented Broadwell chip), so my guess is that the "1.6GHz" refers to the base frequency and that the chip can "turbo" up to a higher frequency if the thermal headroom exists. This is exactly what the Silvermont core inside of the Atom Z3580 tested here does. (Although Intel does not publish the base clocks for the Z3580, the Atom Z3795 -- built on the same process and using the same core -- runs at a base clock of 1.59GHz and can turbo to 2.39GHz.)

Airmont looks like a mere die-shrink of Silvermont

Although we should wait for final production tablet performance numbers to be sure, these initial numbers don’t suggest a large improvement for Airmont over Silvermont.

This is very reminiscent of the Haswell to Broadwell transition for Intel’s big-core line, where Broadwell offered minimal architectural enhancements over Haswell, with the performance-per-watt gains seemingly coming from the transition to the more advanced 14-nanometer manufacturing technology.

While this should be "good enough" for the company's Braswell processor for low-end PCs and for midrange tablet chips, I think Intel will need the next generation core -- known as Goldmont -- in order to make real strides against the competition in the high end of the mobile market.

 

[NTMBK]

If I remember rightly, Airmont is the die shrink and Goldmont is the redesign, yes?

So long as it maintains decent CPU performance, a die shrink will free up more thermal budget for graphics, accelerators and modem- it's not really CPU performance that has been Intel's downfall in mobile, it's Everything Else. We'll have to wait and see how good its new modems are.

 

[witeken]

We'll have to wait and see how good its new modems are.

They're launched in Samsung Note 4 and Alpha. More will follow this year (e.g. Zenphone 2 I guess), and new 7230 will also launch at the end of next year.

 

[Arachnotronic]

They're launched in Samsung Note 4 and Alpha. More will follow this year (e.g. Zenphone 2 I guess), and new 7230 will also launch at the end of next year.

I believe ZenFone 2 is XMM 7160. XMM 7360 is expected to sample in Q1 2015, with commercial deployments in 2H 2015.

 

[Sweepr]

Already posted in BT/CT thread and those scores are not at 1.6GHz, Geekbench lists base clocks, not turbo clocks.

 

[Fjodor2001]

Already posted in BT/CT thread and those scores are not at 1.6GHz, Geekbench lists base clocks, not turbo clocks.

So the conclusion in the article is correct then? I.e. Airmont is not much of a performance improvement over Silvermont?

 

[coercitiv]

So the conclusion in the article is correct then? I.e. Airmont is not much of a performance improvement over Silvermont?

What exact frequency is the Airmont sample operating at?

 

[jdubs03]

So the conclusion in the article is correct then? I.e. Airmont is not much of a performance improvement over Silvermont?

i think it's too early to come to any conclusions. but, i don't think we should get our expectations up too much. especially considering the roll-out is woefully late.

broxton better be something significant because a72's could be coming out (nvidia?...), along with custom qualcomm's and of course newer exynos and apple's likely new uarch+"14nm" finfet.

i think intel is going to wind up losing this low-power battle. i'd feel better if airmont is at least 25-30% faster at the same power, but i don't know if that will be the case, so i cannot be confident in their competitiveness against the ARMy licensees whom have demonstrated solid per gains. the apple a9(x) i venture a guess will be class leading on the cpu side getting close to core m in ST.

 

[Roland00Address]

Intel is having no problems competing with the cpu performance for the next 12 months or so. Where they are failing is in gpu performance and feature sets/integration

 

[jdubs03]

Intel is having no problems competing with the cpu performance for the next 12 months or so. Where they are failing is in gpu performance and feature sets/integration

we're talking one benchmark here so there's that. but when you compare that score against 28nm tegra k1-32 bit, can you really say that intel won't have a problem competing? 20nm 64-bit tegra x1 is likely going to smash that score, and be a top-end performer. the only competitor on the cpu side will likely be the exynos 7420 which has the advantage of finfets.

 

[AtenRa]

Intel intentionally castrated the 14nm ATOMs because Broadwell-M would look terrible.

 

[ShintaiDK]

Intel intentionally castrated the 14nm ATOMs because Broadwell-M would look terrible.

Time for some edvidence, else you are just trolling :thumbsdown:

 

[AtenRa]

Time for some edvidence, else you are just trolling :thumbsdown:

Did you see the OP ???

 

[ShintaiDK]

Did you see the OP ???

And its relevant to your claim how?

http://www.dell.com/us/p/dell-venue-8-7840-tablet/pd

 

[mikk]

So the conclusion in the article is correct then? I.e. Airmont is not much of a performance improvement over Silvermont?

This is a dumb question since nobody here knows turbo frequency of either this prototype nor the planned x,x Ghz on finalized Cherry Trail-T. All we know is there is a protoype with 1.6 Ghz (base probably) with a performance similar to the best Silvermont-T models in Geekbench.

 

[ninaholic37]

This is a dumb question since nobody here knows turbo frequency of either this prototype nor the planned x,x Ghz on finalized Cherry Trail-T. All we know is there is a protoype with 1.6 Ghz (base probably) with a performance similar to the best Silvermont-T models in Geekbench.

http://en.wikipedia.org/wiki/No_such_thing_as_a_stupid_question

 

[AtenRa]

This is a dumb question since nobody here knows turbo frequency of either this prototype nor the planned x,x Ghz on finalized Cherry Trail-T. All we know is there is a protoype with 1.6 Ghz (base probably) with a performance similar to the best Silvermont-T models in Geekbench.

CherryTrail is up to 2.7GHz.

 

[witeken]

Did you see the OP ???

It's a testament to Intel's Tick-Tock pragmatism. GPU is up by 2X, though.

 

[witeken]

CherryTrail is up to 2.7GHz.

Interesting, a >1.5 year old roadmaps shows that Broxton goes up to 2.7GHz, which I didn't know (not too surprising though).

 

[Arachnotronic]

Interesting, a >1.5 year old roadmaps shows that Broxton goes up to 2.7GHz, which I didn't know (not too surprising though).

2.7GHz in phones; Windows tablet (possibly slightly higher power envelope) not listed although I wouldn't be surprised if it were also 2.7GHz.

 

[witeken]

But MGF and WT are both canceled, probably because of the 14 kilopicometer catastrophe.

 

[mikk]

CherryTrail is up to 2.7GHz.

This is just an early estimate or aim. We have to wait for final specifications.

 

[Arachnotronic]

But MGF and WT are both canceled, probably because of the 14 kilopicometer catastrophe.

Morganfield/Willow Trail are now just called Broxton, it seems

 

[witeken]

Morganfield/Willow Trail are now just called Broxton, it seems

False. Broxton is one unified tablet+smartphone platform. Like Herman Eul said, it will make his life easier because he won't have to take care of 2 separate platforms anymore.

 

[Arachnotronic]

False. Broxton is one unified tablet+smartphone platform. Like Herman Eul said, it will make his life easier because he won't have to take care of 2 separate platforms anymore.

...right. Morganfield was the Goldmont/Gen. 9 smartphone platform, Willow Trail was the Goldmont/Gen. 9 Windows tablet platform, and now they are both simply called Broxton.