20nm - Exynos 5430

[jdubs03]

This was just posted: http://www.anandtech.com/show/8382/samsung-announces-exynos-5430-first-20nm-samsung-soc

No 64-bit functionality, maybe that was saved for the 5433? (20nm too?, I would assume it is, otherwise that doesn't make sense). Also it makes it seem like 14nm is a further year away. I doubt they'll go to 14nm only 6-9 months after 20nm. Such a shame, everyone is going to 64-bit, looks like we'll be seeing that next year (unless the 5433 has A57 cores).

This at 20nm compared to TK1-32bit at 28nm isn't even fair in terms of performance (at least based on Antutu, -4000 compared to TK1). Obviously, Samsung needs to keep their attention on process improvements; bringing their chip designs to greater fruition seems to be a massive undertaking. Though one would think they would like to wean themselves off Qualcomm, that isn't going to happen for quite awhile.

 

[TrulyUncouth]

I think its cool that we might have somewhat of an apple to apples A15 big/little comparison from both processes. Should be interesting.

 

[witeken]

Only 33% higher performance per watt from 20nm.

 

[TrulyUncouth]

Only 33% higher performance per watt from 20nm.

Is that the projection? or have actual benchmarks been done?

 

[ShintaiDK]

So more or less stagnat clocks and ARMv7 with a 25% power reduction acording to Samsung.

ARMv7 is dissapointing. And the question begs on where ARMv8 would be power consumption wise with vanilla cores.

 

[NTMBK]

Android L isn't ready, so no rush to 64 bit I guess.

 

[witeken]

Is that the projection? or have actual benchmarks been done?

A projection from Samsung. You really need Tri-Gates at such small geometries.

 

[jdubs03]

Android L isn't ready, so no rush to 64 bit I guess.

Android L will be coming out this year though with the Nexus 9-Denver (maybe we'll get a glimpse at IFA?); but I can see 64-bit not being a necessity with phones (even if Apple has done it). Tablets make more sense in this regard. The next logical 64-bit product from Samsung would be the Galaxy S6; maybe that will have a 20nm ARMv8 Exynos in it, for sure it'll have S810.

 

[witeken]

It's interesting to see that the clock speed of this chip is lower than the 28nm version (1.8 vs 2.1). 20nm process not as good as expected?

 

[kpkp]

It's interesting to see that the clock speed of this chip is lower than the 28nm version (1.8 vs 2.1). 20nm process not as good as expected?

Or maybe is a decision based on a stricter thermal constraints with a smaller battery, I wouldn't complain when they finally do something rational...

 

[dawheat]

It's interesting to see that the clock speed of this chip is lower than the 28nm version (1.8 vs 2.1). 20nm process not as good as expected?

Yeah in this case it's hard to separate the process from the end device they're shipping it in. It's the thinnest Samsung phone with the smallest battery in years. Performance regardless seems equivalent to S801 powered phones - so it also could have been intentional to just launch with parity performance.

 

[NTMBK]

It's interesting to see that the clock speed of this chip is lower than the 28nm version (1.8 vs 2.1). 20nm process not as good as expected?

Could just be that they're going for battery life over flat-out performance.

 

[CakeMonster]

Samsung probably just went with something really simple for the first model to improve yields and be able to tweak the process during the coming months. Then why not make something simple and put it into the really unimpressive (hardware wise) Galaxy Alpha and you get to run that experiment on the cheap too.

 

[Exophase]

This isn't 64-bit because it is, for all intents and purposes, a shrink with some minor modifications. Samsung did the same thing with the 32nm Exynos 4s, before moving on to Exynos 5. It's still possible we'll see a 20nm Exynos 6 later this year.

 

[NTMBK]

This isn't 64-bit because it is, for all intents and purposes, a shrink with some minor modifications. Samsung did the same thing with the 32nm Exynos 4s, before moving on to Exynos 5. It's still possible we'll see a 20nm Exynos 6 later this year.

Shrink first, refine later. Tick tock...

 

[CakeMonster]

At least Samsung has a real 20nm product out, that's a feat despite the product being unimpressive.

 

[witeken]

Could just be that they're going for battery life over flat-out performance.

Highest clock speed shouldn't impact battery life. Highest clock speed impacts highest burst performance. Maybe 6.7mm is simply too thin to make sense for higher frequencies.

 

[Ajay]

A projection from Samsung. You really need Tri-Gates at such small geometries.

Yes, suddenly it become abundantly clear why Intel went with a MuGFET @ 22nm. 20nm planar gives a nice increase in xtors/mm2 but a poor reduction in perf/watt. 14/16FF will likely give a nice increase in perf/watt, but a very poor improvement in xtors/mm2.

David Kanter @ RWT had said in a post, some time back, that he wasn't expecting allot from 20nm or 16nm from the fabs. Now it's clear why - each one is more like a 1/2 node step in one of the two of the most important parameters meaning it's taking fabs two new nodes to achieve the benefits of a 'classic' full node shrink.

 

[jdubs03]

Yes, suddenly it become abundantly clear why Intel went with a MuGFET @ 22nm. 20nm planar gives a nice increase in xtors/mm2 but a poor reduction in perf/watt. 14/16FF will likely give a nice increase in perf/watt, but a very poor improvement in xtors/mm2.

David Kanter @ RWT had said in a post, some time back, that he wasn't expecting allot from 20nm or 16nm from the fabs. Now it's clear why - each one is more like a 1/2 node step in one of the two of the most important parameters meaning it's taking fabs two new nodes to get the achieve the benefits of a 'classic' full node shrink.

Remember MuGFET's are two-gates. There is no gate on top of the fin.

Edit: This has been in the news: http://www.cnet.com/news/samsung-galaxy-note-4-specs-seemingly-leak-online/ 5433 is indeed 64-bit. My guess is it'll be 20nm too, or else their naming sequence is nonsensical.

 

[witeken]

MuGFETs have 2 or more gates.

 

[jdubs03]

True, the MuG is the acronym for Multi-Gate. Thanks for the clarification. From what I have seen though most refer to three or four gates as FinFET/Tri-gate, or GAA/Surrounding Gate.