Samsung Exynos Thread (big.LITTLE Octa-core)

[Qwertilot]

Quite possible at 4w surely? 14nm CoreM not obviously so happy down there - for instance the extra headroom it gets in that zenbook really does seem to help it out a lot. Next die shrink should help a good bit.

Not that the two are actually in competition in any especially obvious way Very different design goals, market positioning etc etc.

 

[Nothingness]

There is a question of power consumption, but they tried to play tricks - see how 4W, <1W are "footnoted" by the Geekbench comparison. Thus I do not think they are applicable to Specint/Specfp, which means the above IPC calculations cannot be used in conjunction with those power consumption figures.

Programs with high IPC (most Geekbench tests) tend to consume more than those with low IPC.

 

[coercitiv]

Programs with high IPC

What's a program with high IPC?

 

[Nothingness]

A72 has to compete versus SKL-Y and CNL-Y, so good luck with that.

No, A72 has to compete against *mont. Good luck Intel :biggrin:

That comparison against Broadwell indeed makes little sense...

 

[Nothingness]

What's a program with high IPC?

A program with little cache misses and branch mispredictions.

 

[Arachnotronic]

No, A72 has to compete against *mont. Good luck Intel :biggrin:

That comparison against Broadwell indeed makes little sense...

Goldmont in this case, if A72 on 16FF+ is out in 2016.

That said, Intel is going to need to make a big leap in going from Airmont to Goldmont if it wants to hang with the A72.

 

[coercitiv]

A program with little cache misses and branch mispredictions.

From no sense to a lot of sense with just one sentence, thx

 

[witeken]

ARM claims A72 uses 4x less power than BDW-Y. So this basically means, given the slide below, that ARM asserts A72 at 28nm Planar will be as power efficient as Broadwell at 14nm 2nd gen FinFET .

Seems legit.

Edit: I didn't read the slide good enough: 28nm A72 will still be 2X as efficient as BDW-Y.

 

[soresu]

I think the graph was claiming that A72 at 28nm is 2x more efficient than A15 at 28nm, and 4x more efficient than 28nm A15 at 16nm FF+

Edit: Didnt notice the red and blue bar graph, my mistake. Also has anyone seen the KitGuru article on Eynos M1? http://www.kitguru.net/components/c...sor-core-already-supported-by-software-tools/

 

[soresu]

Im still wondering why they are taking so long to announce the Artemis core details when they basically announced its existence at the same time as A72 last year (Maia). I think that Artemis was intended as mid range, but then Qualcomm and others ran with A72 for mid range chipsets (SD 618, 620), so now Artemis has an uncertain target in the marketplace.

 

[Sweepr]

I think the graph was claiming that A72 at 28nm is 2x more efficient than A15 at 28nm, and 4x more efficient than 28nm A15 at 16nm FF+

Edit: Didnt notice the red and blue bar graph, my mistake. Also has anyone seen the KitGuru article on Eynos M1? http://www.kitguru.net/components/c...sor-core-already-supported-by-software-tools/

Nice find.

While the rumours about Samsung&#8217;s plan to introduce its own general-purpose cores have been floating around for a long time, this is the first official confirmation of the project. Moreover, addition of the support means that the new Exynos M1-powered system-on-chip already exists in silicon and is being tested by Samsung Electronics internally at the moment. It is possible that closest software partners of the South Korean giant also have access to the new SoC.

Traditionally, developers receive the first samples of their microprocessors 9-12 months before the start of mass production. If Samsung plans to use an application processor based on its first full-custom general-purpose processing core inside its next-generation Galaxy S7 flagship smartphone, then it is not surprising that the company is already testing this SoC with M1 technology.

It's almost a given that Galaxy S7 will have it, but will the custom M1/Mongoose cores make it to Galaxy Note 5's Exynos SoC? I hope so.

 

[soresu]

M1 may also refer to their ModAP line to give higher efficiency with 4G+ technologies aswell.

 

[Sweepr]

ARM claims A72 uses 4x less power than BDW-Y. So this basically means, given the slide below, that ARM asserts A72 at 28nm Planar will be as power efficient as Broadwell at 14nm 2nd gen FinFET .

Seems legit.

Edit: I didn't read the slide good enough: 28nm A72 will still be 2X as efficient as BDW-Y.

By the way, when is 16nm FF+ Cortex A72 SoCs coming out? Skylake-Y should be out by Q4-2015.

 

[soresu]

Feature set of Huawei Kirin 940 and 950 claims A72 in Q3 for 940 and Q4 for 950 (950 also has T880). I cant seem to find a process for these chips but chances are one of them may be on 16nm FF

 

[soresu]

The EETimes article on A72 has an official from ARM claiming that three functional units in the core were taken from a clean slate design high end core which hasnt been formally announced yet. The author of the article seems to think the unannounced core (probably Ares) may well compete with Broadwell for pure performance. http://www.eetimes.com/document.asp?doc_id=1326444

Edit: The main point of the article seems to be that A72 was designed for greater efficiency rather than pure performance.

 

[Arachnotronic]

The EETimes article on A72 has an official from ARM claiming that three functional units in the core were taken from a clean slate design high end core which hasnt been formally announced yet. The author of the article seems to think the unannounced core (probably Ares) may well compete with Broadwell for pure performance. http://www.eetimes.com/document.asp?doc_id=1326444

Edit: The main point of the article seems to be that A72 was designed for greater efficiency rather than pure performance.

A72 looks like a solid core from ARM.

 

[lopri]

I personally believe Samsung will save their custom core until the Galaxy S7, based on the rumored performance level. In Geekbench ST;

A57 @2.0 GHz: 1500+
A72 @2.0 GHz: 1700+
Mongoose @2.0 GHz: 2000+ (estimated)

With no competition to the 7420 in sight (28nm A72 isn't one), I can see Samsung saving their custom core for the future when it's needed, e.g. against S820 that will be out for the next year's devices. I will be happy to be proven wrong, but that is what I think.

 

[krumme]

I personally believe Samsung will save their custom core until the Galaxy S7, based on the rumored performance level. In Geekbench ST;

A57 @2.0 GHz: 1500+
A72 @2.0 GHz: 1700+
Mongoose @2.0 GHz: 2000+ (estimated)

With no competition to the 7420 in sight (28nm A72 isn't one), I can see Samsung saving their custom core for the future when it's needed, e.g. against S820 that will be out for the next year's devices. I will be happy to be proven wrong, but that is what I think.

A72 on dirt cheap 28nm is a formidable competitor. When q3 hits we have 7420 performance on midrange and imo it will go to lowend already q4 as everyone and his brother will print their own a72 at near zero cost. What does 10mm2 cost on 28nm. The impact for highend is hard. This is not octo core a53 marketing performance. This is a8 real world performance dirt cheap. What will samsung and qq do to that?

I simply dont know but the obvious answer is their own cores on finfet - fast ! What else?

 

[lopri]

Higher clocked A57 or Samsung's own A72..? Most optimistic competition on 28nm would be 2+4. I frankly do not see 4+4 on 28nm from Mediatek, Rockchips, Kirin, etc. Even if they somehow go for it, they are still competing against 4+4 on 14nm.. They won't match performance and frequency of the revised 7420.

Again, I would like to see the custom cores, too. I just think it may be too early for Samsung for various reasons.

 

[Sweepr]

Qualcomm is doing 4+4 A72/A53 at 28nm with the S620 (mainstream SoC).
LG supposedly cancelled their 4+4 A57/A53 SoC due to overheating and will launch a 4+4 A72/A53 SoC at 20nm/16nm FF+ later this year.
Huawei is rumoured to pack 4+4 A72/A53 at 16nm FF+ inside their Kirin 940/950 SoCs in H2-2015.

And the list goes on. I have no doubt Sammy could deliver the best 2.4-2.6GHz octo-core A72 SoC later this year but I'm expecting a bit more from them if they want to diversify and maintain their performance advantage relative to the Android competitors and Apple.

They adopted A57 before everyone else and the only 14nm phone SoC belongs to them. I have high expectations for their next SoCs.

 

[krumme]

Some collection of gossip of Mongoos samsung next core
http://m.gsmarena.com/samsungs_new_custom_cpu_cores_45_faster_than_a57-news-12059.php
Here we have this 45% faster than a57. Note via and source in bottom. Phonearene have other sources (basically its old wine). But hey

 

[Sweepr]

A bit OT but the fact that LG chose the Snapdragon 808 instead of S810 for the G4 (after adopting S810 with the G Flex 2) looks like the ultimate proof of the reported power/heat issues.

 

[krumme]

A bit OT but the fact that LG chose the Snapdragon 808 instead of S810 for the G4 (after adopting S810 with the G Flex 2) looks like the ultimate proof of the reported power/heat issues.

And the g4 808 a57 freq is still very low. And performance is too:
http://m.gsmarena.com/lg_g4_hands_on-review-1245p3.php

Damn...
Qcom 820 cant come soon enough. This is eaten alive of the new cheap 28nm a72.

 

[kpkp]

A bit OT but the fact that LG chose the Snapdragon 808 instead of S810 for the G4 (after adopting S810 with the G Flex 2) looks like the ultimate proof of the reported power/heat issues.

Nah...
LG has been Qualcomm launch partner for most of the high end SoCs in recent years.

They just like to work together.


About the bench, LG phones don't have a history of high scores, it seems they don't pay much attention to the thermal dissipation of the SoC.

 

[altairauditore]

How will Snapdragon 618 (2xA72+4A53) compare to 808 (2xA72+4A53)?
618 will have a slower GPU, might be slower CPU frequencies, old node, worse (ISP, other on board features).