Samsung Exynos 5250 - production starting in Q2 2012

[newbieforever]

https://twitter.com/SamsungExynos

They have posted about their private event though.

 

[Nebuchadnezzar]

Exynos 5260: 667MHz
Exynos 5422: 533MHz
Exynos 5430: 600MHz

All T6XX gpus. Don't know the exact variants.

Interestingly, this is also the first empirical evidence of the existence of the 5430.

There also seems to be a Exynos 4415 who I also never heard of.

 

[Nebuchadnezzar]

Some info:

The Exynos 3250 is an interesting little 1GHz A7 dual-core on a low-power HKMG 28nm process. Not sure what GPU but it seems to be a Mali 4XX. The SoC is a mix of older Exynos4 block IP updated with some of the latest IP (MMC among them) found in the 5 series. It's what's shipping in the Gear 2.

 

[Erinyes]

Interesting little part. I wonder why we haven't seen it being used in their lower entry level/lower mid-range phones yet. A lot of their lower end phones have Snapdragon 200 with dual A5's. This would be a good step up..and they'd be using their own silicon as well.

 

[Nebuchadnezzar]

Samsung is exposing their core power coefficient values in their current IPA drivers:

5422 rev0 : 57 A15, 11 A7
5422 rev1 : 48 A15, 9 A7

As far as I know this is the first time ever we have proper C values from a manufacturer for C*f*V², and can calculate core dynamic power independently:

For rev1:

I took a mid-binned ASV chip for the voltage values.

The values are per core. The IPA driver uses the same calculation method for the power allocation.

These numbers are a massive drop from the 5410 framework numbers back from last year, even if they don't contain static power: http://forum.beyond3d.com/showpost.php?p=1732331&postcount=530

 

[Deleted member 13524]

Power consumption doubles between 1.4 and 1.9GHz for the A15, wow...

 

[Ailuros]

Power consumption doubles between 1.4 and 1.9GHz for the A15, wow...

Most of the dirty (= simple) work comes from the A7 cores, so if you consider the right of the table it balances out quite nice for average consumption. Are there even cases today where a A15 core would need to reach 1.4GHz (honest question)?

 

[Nebuchadnezzar]

Are there even cases today where a A15 core would need to reach 1.4GHz (honest question)?

Need, or do? Because there's plenty of bursty loads taking advantage of the high clocks. Web-page loading, app loading, etc are good examples. Games are really the Achilles' heel here in terms of continuous load, but I'm a bit out of the loop on the high-end games now-days and what they require. And even that is taken care of by new power management systems, GPU and CPU can never run both at max clock at the same time, the last two DVFS levels on the GPU limit the A15's to 1600MHz.

 

[Ailuros]

Need, or do? Because there's plenty of bursty loads taking advantage of the high clocks. Web-page loading, app loading, etc are good examples. Games are really the Achilles' heel here in terms of continuous load, but I'm a bit out of the loop on the high-end games now-days and what they require. And even that is taken care of by new power management systems, GPU and CPU can never run both at max clock at the same time, the last two DVFS levels on the GPU limit the A15's to 1600MHz.

In games I can see peaks of 1-1.1GHz on quad A7s tops unless I've missed anything while playing. Even if there are higher peaks they should be extremely rare. For A15 and considering the times higher IPC I'd be very surprised if frequencies would reach as high as 1.4 apart from very few exceptions. And yes it was rather "do" than "need".

 

[Exophase]

These numbers bring things closer to what nVidia claims for Tegra K1, which is about 1W per Cortex-A15 @ 2GHz, although there's still a pretty noticeable discrepancy. It could be explained by any combination of:

- nVidia exaggerating or outright lying, or their numbers not taking account for some components that Samsung's numbers are
- nVidia is binning more aggressively. I'm curious, what do the numbers look like for Samsung's best ASV?
- nVidia's implementation on TSMC 28HPM is more optimized for dynamic power consumption at the expense of static power consumption.. or TSMC just has a better process, or nVidia just has a better implementation
- nVidia is using newer A15 cores with more power consumption optimizations (A15 r3) - does anyone know what version of A15 Exynos 5422 is using?

Would also be good to see actual measured numbers on either side..

 

[Nebuchadnezzar]

5422 uses r2p3 cores as per the chip ID. r2p3 firstly doesn't officially exist, and r3p0 adds features which Samsung already has in its cores. So I doubt you'll find meaning in the revision IDs because r3p0 and r2p3 might just well be the same thing.

The best ASV bin is about 50mV lower at the high steps compared to what I posted. However the distribution of bins is highest at the one I posted and you should consider that as the norm.

TSMC's HPm is able to achieve a good 125mV lower voltage across the board compared to Samsung's current process. I'm saying this by comparing the S800's voltages to the S600 and the gains that came with it. TSMC has undoubtedly a better process.

 

[loekf]

Not sure if this has been posted here:

http://www.sammobile.com/2014/04/22...d-replace-snapdragon-cpus-on-samsung-devices/

- 20 nm process
- Exynos 5430 is an upgrade to the 5422-
- 2.1GHz on the A15’s and 1.5GHz on the A7’s
- Mali Midgard architecture (T6XX) @ 600MHz
- HEVC (H.265) hardware decoder
- Dedicated (Cortex A5) co-processor for audio encoding, decoding and audio equalization
- Minimum 600 mV Vdd

This should be put into a S5 "Premium" and pairs the SoC with an Intel-made 4G modem...

 

[Nebuchadnezzar]

- Minimum 600 mV Vdd.

I wrote that the voltage rail driver got lowered to that as a minimum (from 800), not that it's the actual minimum used. Minimum Vdd seems to be 725mV. Anyway I left out several other clues in the article which point out to 20nm, I'm 90% certain of it by now.

 

[wco81]

Well new devices with Exynose sighting:


http://arstechnica.com/gadgets/2014...wo-sharpest-additions-to-the-galaxy-tab-line/

Samsung makes so many tablets, these might get lost in the noise but the displays should get noticed.

 

[Nebuchadnezzar]

It's still the same old 5420 from last year and still running cluster migration.

 

[wco81]

Hmm, maybe they had to cut costs elsewhere to pay for the OLED screens?

 

[Nebuchadnezzar]

Only reason I would see is they have a stock of 5420's they need to get rid of, the successor tablets with 5433's are already in development.

 

[lanek]

I dont know if it is really the 5420 (( maybe i have not check ). Still there's the Snapdragon 800 version who will certainly be the international one and the most common.

Thoses new Tab S are anyway a little bit different of the other models, who was more for the middle range market. SuperAmoled, 2560x1600. 7 and 10.5" + some different rafinement we can find on the S5.

 

[Alexko]

I wonder why Samsung didn't go for the S801 on such high-end devices.

 

[Deleted member 13524]

I wonder why Samsung didn't go for the S801 on such high-end devices.

True. Those huge AMOLED panels must be pretty expensive. Why save a couple of bucks in the SoC if they're probably high-margins products anyway?



Maybe the typical consumer looks less at the performance numbers when dealing with tablets than they do with flagship phones.