Samsung SoC & ARMv8 discussions

Thought about starting a new thread since the old one is getting outdated and needs a clean start.

Recent patches released by Samsung points out that the GH7 SoC is an 8-core A57 server chip with dedicated networking co-processor(s).

I wonder what will happen with it since they never officially announced the 5440 server chip.

That's all I got for now.

 

If one was being pedantic, one might say that a posting about a server chip with networking co-processors isn't exactly the best of clean starts for a thread in the "handheld technology" forum.

 

Handheld server?

 

I assume you mean 8 A57 cores and not 4 A57 and 4 A53? I saw another post of yours saying they had first silicon back in December. Did you mean this server chip or were you talking about a consumer chip with presumably 4 A57 + 4 A53?

If one was being really pedantic then sure...but given that the same core and a similar SoC would be used in the mobile space..it would definitely have some bearing to the handheld technology forum.

 

That's what I wrote.

Not sure which one was referred back in December, just that it's ARMv8.

Then don't be pedantic about it. It has direct effect on their mobile lineup timeline, as Erinyes said. Plus we're in the embedded forum and I don't see any server sub-category.

 

There is a reason there is no server sub-forum, its because the parent forum is the embedded 3D forum.

 

http://community.arm.com/servlet/JiveServlet/showImage/38-3323-8512/Glenn.jpg

Points out they taped out a production chip on (I assume) 14nm this May.

Source:
http://community.arm.com/groups/soc.../dac-breakfast-14nm-is-real-and-ready-for-you

If I'm not mistaken the nominal voltage for 14nm is 0.8V, so their DVFS range goes from 0.6 to 0.9V in the chart. Obviously the Y axis unit seems to be watts going by the scale of little/big cores.

 

Looks like the little cores don't have a big perf/W advantage on Samsung's 14FF. Compare with what ARM was claiming for A15 + A7 big.LITTLE, presumably determined for a 28nm (TSMC?) process:

http://cdn.arstechnica.net/2011/10/20/arma7-dvfscurves-4ea0421-intro.png

Even if the Samsung is hitting the minimum voltage for the big core and could only scale down linearly from there it still wouldn't be that much worse than the little core. Kind of makes you wonder why they're bothering anymore.

I don't think it's necessarily watts, more likely it's normalized against the UD2 point, while the performance axis is normalized against the lowest little core figure.

 

There still seems to be something like a 30% power difference at the UD2 point; that's nothing to sneeze at. Plus, there appears to be a bit more overlap, i.e. the little core can go higher. But that's hard to say without real units for the scales.

 

Not sure if this is the right thread but it seemed to be the closest to the topic. Read two interesting articles which I haven't seen posted on here yet.


TSMC announces its first 16nm FinFET networking chip: 32-core ARM Cortex-A57 - http://www.extremetech.com/computin...finfet-networking-chip-32-core-arm-cortex-a57

Some noteworthy points:-

1. Built by HiSilicon. While not explicitly stated, it appears to be working silicon based on a quote from HiSilicon President Teresa He - “We are delighted to see TSMC’s FinFET technology and CoWoS solution successfully bringing our innovative designs to working silicon,”. No mention of volume production.

2. According to the foundry, the HiSilicon device has twice the gate density of 28nm HPM, is 40% faster at the same power consumption, or can reduce power consumption by 60% at the same speed.


TSMC announces first 16nm FinFET results, unveils 10nm roadmap - http://www.extremetech.com/computin...irst-16nm-finfet-results-unveils-10nm-roadmap

1. “Silicon results on 16FF show the ‘big’ Cortex-A57 processor achieving 2.3GHz for sustained mobile peak performance.”

2. TSMC also claims that Cortex A53 on 16nm FF will consume just 75mW “for most common workloads.”

3. TSMC is also discussing a new type of FinFET process (FF+), which is expected to start ramping up late this year and which will deliver a further 11% performance gain for the Cortex-A57 and a 35% reduction for the Cortex-A53 when running in minimal power consumption mode.

 

funny

http://www.gsmarena.com/samsung_announces_the_exynos_7_octa_is_the_old_exynos_5433-news-9946.php

Exynos 5433 = Exynos 7

 

Then I guess we might see the Note 4 getting an update to 64bit when Lollipop comes around, after all.

 

Notice the complete lack of any 64-bit mention in their announcement, so I still stand by my claims that they won't update it to AArch64. Hopefully I'm wrong but I doubt it.

 

Do you have a theory why? (sorry if you've posted one already)

 

Like I said in the other thread, that they're shipping the Note 4 with an ARM32 kernel and the chances that they will be doing a new ARMv8 platform just 2 months afterwards are from a software development and Q&A standpoint close to nil.

 

http://www.android.com/versions/lollipop-5-0/

 

Well i dont think they will do it too.

The only possibility i see is if they want to release a new version of the Note4 later .. other than that, this will mean they have 2 versions on the market, one "unlocked" 64bit bit, and the Snapdragon one in 32bit and thats the international version. That is not really good as a marketing situation.

 

Linux 64-bit kernel for 5433. I guess that doesn't mean we'll see Note 4 with 64-bit Android alas.

arm64: Add the support for new 64-bit Exynos5433 SoC

 

http://www.fudzilla.com/home/item/36478-samsung-gpu-in-summer-2015

I beg your pardon?

 

This is strange.
Samsung should be able to purchase i.e. Vivante and enter the market much faster.
Why create a GPU from the ground up? And who is working on it?