Samsung SoC & ARMv8 discussions

So any cheating done by Samsung in benchmarks or any evidence in that direction yet?
A lot of people are claiming that (arbitrarily)
Not that synthetic benchmarks are everything but they do help differentiate between a ...potato and something not a...
(JK alright I like potatoes )

 

It was 5433.

 

http://www.anandtech.com/show/9111/samsung-galaxy-s6-and-s6-edge-preview

Well not bad I guess, but I think it falls a bit short of expectations for me (especially on the cpu side). I also suspect a finfet series 6xt would probably be king. Are these gpu benches first run or after x loops?

Unrelated, but can someone buy Josh a 6+?

 

If it's about GPU performance just add 1 more fps on top of the iPhone6 scores in Manhattan

 

Found this over xda forums:
http://samgukji.net/won/link/?item_no=783753&ckattempt=1

Will quote a few sentences (Using google translate):

So the article claims a custom core samsung soc codenamed jjirasi (Mongoose) and a 2240 score in single threaded geekbench
Another thing :

what i know is meerkats hunt snakes even though google translate gives you barely coherent statements so it looks like a interesting codename.
Also there are many claims over other geekbench scores which look more like calculations/speculation.

 

jjirasi is not a code-name, Mongoose is. It looks like jjirasi is basically a slang form of "Tabloid".

"News about Samsung's custom core Mongoose is out"
"By the looks of it, having leaks now means starting(design) few years ago and code-name indicates awareness of a competitors product"

Krait is a snake, so Mongoose is supposed to eat that? A reference to Qualcomm he says.

The picture says: Geekbench 3 ST
2.3GHz = 2216/2117/2240

(Does some extrapolations of Integer and FP score based on some assumptions)

His extrapolations say,

Mongoose: 1974 points/GHz
Exynos 7420: 1470
Apple A8X: 2357
(The above scores are not Geekbench scores but Integer and FP scores without the memory score)

"2.3GHz clocks indicate 14nm part"
"Can't assume its a 10nm part because its too early"
"Probably at least quad cores"
(Some more extrapolations based on number of cores and ST scores)

 

About Mongoose: the info comes from a Korean guy which has a relatively correct track record. No reason for doubt for now.

http://www.chipworks.com/en/technic.../resources/blog/inside-the-samsung-galaxy-s6/

Chipworks is reporting 78mm² for the 7420. That's an insane reduction over the 5433 at 110mm² given it has 2 more GPU clusters. I'm really intrigued on what they changed in the SoC. This could really make them very price competitive against Qualcomm which seems what has happened in the S6, S810 issues aside.

 

I seem unable to get any kind of translation at all.
Could anyone help out?

 

Wow, so Samsung makes the SoC, RAM, flash, PMICs, modem, radio, image processor, wifi module, and (although it's not listed, but AFAIK) the display. They're getting really close to complete vertical integration, or at least probably closer than anyone else has.

 

Wi-Fi module is SEM but still uses a BCM4358 inside. Pity their 20mp sensor didn't get the design win otherwise this would have been 95% Samsung BoM.

 

Could have never understood that from google translate alone.
Anyways Kyro cores vs Mongoose cores would be interesting to see , especially if they release around the same time.

 

QCOM's custom CPU is codenamed KRYO and not Kyro.

 

Couldn't edit that typo without posting again...
(Even though I can edit this post somehow

 

The power of the new S6 Soc , still not an A8x but is the fastest android socs on the market

http://browser.primatelabs.com/geekbench3/2314492

 

Camparing it to a tablet soc is unfair, if you compare it with an A8 is faster in a lot of thing and in particular multithreaded code

 

Any comparison is somewhat questionable exactly because you can't run the 7420 under iOS either. To be even more fair though the 7420 is one process and technology generation ahead of the A8 too and heck it's no surprise either when you barely surpass it in a couple of a spots, when it should actually spank it green and blue across the board. The GPU despite its huge size isn't cutting the cake exactly and when after an hour or so in T-Rex when heavy throttling starts kicking in, the A8 is one of the last that would have to hide: http://gfxbench.com/compare.jsp?benchmark=gfx31&D1=Samsung+Galaxy+S6+Edge+(SM-G925x,+SC-04G,+SCV31)&os1=Android&api1=gl&D2=Apple+iPhone+6+Plus&cols=2

I don't know what Samsung and in extension QCOM is planning in the longrun when it comes to GPUs, but it might be one thing to consider that the consumers aren't as big idiots as they would want to think; when I see N performance in a synthetic benchmark I would like to have it as much as possible sustained.

 

Has there been extensive testing of performance degradation on the S6?

 

No idea, but I'm looking forward to Anandtech's S6/S6 Edge review.

 

http://www.anandtech.com/show/9146/the-samsung-galaxy-s6-and-s6-edge-review

No real degradation testing though. That's literally the most important benchmark! I don't care about a SoC's burst performance! I want to know what performance I can expect after playing a game for 30+ minutes. WTB: looping off-screen benchmarks...

 

There are some performance degradation tests on page 3.