Samsung Exynos 5250 - production starting in Q2 2012

[french toast]

It actually saves power.

Saving area matters less than saving power

Right before I even start a reply to this nonsense I will say you are getting a little personal lately..If you have a gripe with me or my writing style..fine no problem with that...just PM me with it.
- If for nothing else that to saves everyone else wasting time reading it. (I find it mildy funny)

Right..nvidia prism..yes you are correct that is more nvidia marketing for their display technology. .and not the variable SMP core swapping set up I was referring to..although I dont take much stock in nvidias marketing...and besides im sure ive read sites that have also got that term mixed up...not suprising considering nvidia adds marketing terms to everything it can.

I never said tegra variable smt was the same thing as big little... I know fully well they are different technologies... (pointed out to me by exophase/metaphor last year)
However if you bothered to read the content of my hypothesis, I mearly put forward the idea of either 4 eagle cores and 1 a7 or vice versa as being more efficient than throwing 4 of each in....to my knowledge big little has that flexibility to be able to do such a scenario. ..
..
Quite a feasable and reasonable scenario put forward..yet you are more interested in childish ridicule before properly digesting the content.

You are not seriously telling me we need 4 cortex a15s @1.9 ghz just load web pages are you?? How much faster would that scenario be over 1 or 2 cores at around the same speed or higher?

Something like this I linked..ironically a few pages back, practice what you preach
http://blog.gsmarena.com/samsung-pr...big-little-arm-architecture-of-exynos-5-octa/

Edit Apologies. Quote meant for nebuchanezza.

 

[french toast]

Sorry rpg 314..that post was supposed to be quoting nebuchanezza.. damn phone app.

 

[french toast]

I don't know the requirements are, but the A7s are validated if any of its threads make more sense to run on A7 than an A15, not just all of them. What I do know is this:

1) Most games don't have the means to vary their per-frame CPU requirements.
2) Most games work okay on hardware with lower end CPUs (like iPhone 4S: 2x Cortex-A9 @ 800MHz) even if they also have relatively powerful GPUs (iPhone 4S again), which a 1.2GHz Cortex-A7 can match.
3) Consoles in particular have driven game development to utilize many more threads than they have in the past, and this trend will only continue.

Yea?...so why have 4 a15s AND 4 a7s if you and turtle say a7s and gpu are capable of doing gaming??
You pour scorn on my questions yet you havnt explained why we need all those resources. ..quite the contrary. ..you seem to agree we dont :/

Edit..correct me if im wrong... last year weren't you one of the people questioning whether we woukd ever need more than a dual core for phones??

Edit..this..
http://blog.gsmarena.com/samsung-pr...big-little-arm-architecture-of-exynos-5-octa/

 

[french toast]

It actually saves power.

Saving area matters less than saving power

Yes I know big little in general saves power over a standard multi core setup..im only questioning the use in a smartphone of 8 cores...personally I think it might be overkill if almost all tasks can be run of a7s like some have put forward. .and one or two a15s can be used to load web pages and apps quick then race to sleep.

Not unlike samsungs/arms recent demo...shown here..
http://blog.gsmarena.com/samsung-pr...big-little-arm-architecture-of-exynos-5-octa/

 

[Jubei]

Right before I even start a reply to this nonsense I will say you are getting a little personal lately..If you have a gripe with me or my writing style..fine no problem with that...just PM me with it.
- If for nothing else that to saves everyone else wasting time reading it. (I find it mildy funny)

Right..nvidia prism..yes you are correct that is more nvidia marketing for their display technology. .and not the variable SMP core swapping set up I was referring to..although I dont take much stock in nvidias marketing...and besides im sure ive read sites that have also got that term mixed up...not suprising considering nvidia adds marketing terms to everything it can.

I never said tegra variable smt was the same thing as big little... I know fully well they are different technologies... (pointed out to me by exophase/metaphor last year)
However if you bothered to read the content of my hypothesis, I mearly put forward the idea of either 4 eagle cores and 1 a7 or vice versa as being more efficient than throwing 4 of each in....to my knowledge big little has that flexibility to be able to do such a scenario. ..
..
Quite a feasable and reasonable scenario put forward..yet you are more interested in childish ridicule before properly digesting the content.

You are not seriously telling me we need 4 cortex a15s @1.9 ghz just load web pages are you?? How much faster would that scenario be over 1 or 2 cores at around the same speed or higher?

Edit Apologies. Quote meant for nebuchanezza.

Your argument is all over the place though. You think 4 Cortex A15 + 1 A7 is a good idea then you turn around and question the need of 4 Cortex A15s?

As for why Samsung is going with 8 cores, it was explained a few pages ago that due to the software for the migration models not being ready yet it is recommended to have the same number of cores for both clusters

That means either 2 or 4 cores for each cluster. Now if Samsung had gone with 2 cores of each, that means either only 2 Cortex A7s or A15s can be active or 1 Cortex A7 + 1 Cortex A15 until MP model is ready to switch between any core or activate all of them

 

[french toast]

Your argument is all over the place though. You think 4 Cortex A15 + 1 A7 is a good idea then you turn around and question the need of 4 Cortex A15s?

As for why Samsung is going with 8 cores, it was explained a few pages ago that due to the software for the migration models not being ready yet it is recommended to have the same number of cores for both clusters

That means either 2 or 4 cores for each cluster. Now if Samsung had gone with 2 cores of each, that means either only 2 Cortex A7s or A15s can be active or 1 Cortex A7 + 1 Cortex A15 until MP model is ready to switch between any core or activate all of them

No I questioned the need for 8 cores...I missed recommendations from arm..its others who questioned the need for 4 a15s for anything other than loading web pages....I disagree.

 

[Ailuros]

I disagree with the continious waste of website bandwidth

 

[french toast]

I disagree with the continious waste of website bandwidth

Well stop then

 

[Nebuchadnezzar]

Right before I even start a reply to this nonsense I will say you are getting a little personal lately..If you have a gripe with me or my writing style..fine no problem with that...just PM me with it.
- If for nothing else that to saves everyone else wasting time reading it. (I find it mildy funny)

Sorry but an excruciating large amount of your posts are just nonsensical blabbering and you are incapable of composing a constructive and coherent thought on any kind of topic throughout this forum. And I'm not the only one of this opinion. It has nothing to do with your "writing style", it has to do with what you write.

Quite a feasable and reasonable scenario put forward..yet you are more interested in childish ridicule before properly digesting the content.

You're trying to take part in a discussion without the most basic and elementary knowledge on power management and you're telling me to digest content? I can't even reply to the rest of your post because I would have to get you through basic scheduling and threading before we can have a constructive discussion. The problem is, is that this has been explained in previous pages.

Something like this I linked..ironically a few pages back, practice what you preach
http://blog.gsmarena.com/samsung-pr...big-little-arm-architecture-of-exynos-5-octa/

What do you want to convey here again? It validates exactly what I said, webpage rendering is what loads the A15 cores to their maximum.

To the others: Sorry for my tone but these rantings are basically starting to sound like intentional trolling and bring the level and quality of discussions to the equivalent of spam. Moderators are welcome to kick in if I'm out of place.

 

[french toast]

I think your going way over the top to be honest...I do post alot..perhaps to much.if thats over the top then I will write less often.

However saying that all my posts are basically amount to trolling is absurd to say the least. .I have started and partaken in a number of popular threads. ..ones that you your self have wrote on and discussed with me..as well as constructive conversations with many other prominent members.

If people/moderators honestly agree with the personal crap your spewing and my contributions amount to nothing more than trolling then id be happy to leave the forum altogether.

Cheers.

 

[Exophase]

It validates exactly what I said, webpage rendering is what loads the A15 cores to their maximum.

Just some clarification for those who might be scratching their heads on this..

Games are bound by GPU and I/O. Even if the GPU can handle it it doesn't make sense to render beyond a certain number of frames per second because it won't make a difference to the user and the display won't be able to keep up, so it'll just waste power. GPU load on games is fairly negotiable, but CPU load nearly as much so, because a big chunk of it is stuff you agree on to make the game work (process internal events and input and stuff at a certain rate regardless of output frame rate). Since no one wants to make a game that requires top of the line new phone hardware just to work you don't get CPU requirements like this.

Web page loading is different because it's the kind of task that you want done ASAP, meaning you load the CPUs to the limit you can thread them to and you keep them there until the task is done. That task still end up taking under one second but that's forever compared to the amount of time a game dedicates to a frame.

Games probably do have periods of "get it done ASAP" too, like when loading a level, but that isn't their steady state.

 

[french toast]

Just some clarification for those who might be scratching their heads on this..

Games are bound by GPU and I/O. Even if the GPU can handle it it doesn't make sense to render beyond a certain number of frames per second because it won't make a difference to the user and the display won't be able to keep up, so it'll just waste power. GPU load on games is fairly negotiable, but CPU load nearly as much so, because a big chunk of it is stuff you agree on to make the game work (process internal events and input and stuff at a certain rate regardless of output frame rate). Since no one wants to make a game that requires top of the line new phone hardware just to work you don't get CPU requirements like this.

Web page loading is different because it's the kind of task that you want done ASAP, meaning you load the CPUs to the limit you can thread them to and you keep them there until the task is done. That task still end up taking under one second but that's forever compared to the amount of time a game dedicates to a frame.

Games probably do have periods of "get it done ASAP" too, like when loading a level, but that isn't their steady state.

Thanks exophase... its only the need for 4 full a15s in the setup that I question if a7s can cope with most tasks.
The link provided uses 2 a15s and 3 a7s...that actually looks like a nice ballance for a smartphone IMO.

 

[Ailuros]

Well stop then

If I stop posting on a forum like this I doubt anyone will miss me; now imagine how much anyone would bother about anyone else.

What is it exactly that you don't "get" here or don't want to get? Yes part of Samsung's decision for a QUAD A15 (it's only an "octacore" as a result of big.LITTLE) was/is marketing but not more or less than with any other current or upcoming competing SoC. Apart from that it's an overall well balanced SoC and it'll sell well and yes we can waste another 15-20 pages why it had to be 4+4 cores, 2+2 or why the GPU isn't a Borgo Sphinxter 6969 from hell instead.

It's not less a marketing decision than the quad A15 in NV's Tegra4 (whereby the companion core on that one is far more idiotic since it's bigger than the quad A7 on the Exynos 5410) or quad core Krait despite it having asynchronous frequencies. All of those alternatives have their advantages and disadvantages and guess what folks that design those things aren't complete idiots either.

 

[french toast]

If I stop posting on a forum like this I doubt anyone will miss me; now imagine how much anyone would bother about anyone else.

What is it exactly that you don't "get" here or don't want to get? Yes part of Samsung's decision for a QUAD A15 (it's only an "octacore" as a result of big.LITTLE) was/is marketing but not more or less than with any other current or upcoming competing SoC. Apart from that it's an overall well balanced SoC and it'll sell well and yes we can waste another 15-20 pages why it had to be 4+4 cores, 2+2 or why the GPU isn't a Borgo Sphinxter 6969 from hell instead.

It's not less a marketing decision than the quad A15 in NV's Tegra4 (whereby the companion core on that one is far more idiotic since it's bigger than the quad A7 on the Exynos 5410) or quad core Krait despite it having asynchronous frequencies. All of those alternatives have their advantages and disadvantages and guess what folks that design those things aren't complete idiots either.

I didnt call them idiots so please dont put words in my mouth.
I also never mentioned octa being marketing or that it woukdnt work. .again...

I only questioned the point of having 4 a15s with a7s...aside from the recommendations of using equal amounts of both cores. .due to Migration software not mature enough, I still think its over kill in a smartphone....tegra even more so because no a7s and no asynchronous mp.

I would have preferred the setup in the arm demo (running on android ics)...it loaded web pages instantly with only 2 a15s and no gpu..I fail to see what noticeable advantage another 2 eagle cores would achieve..other than sucking power.

Snapdragon is different because kraits consume less power than a15s, seem to be more optimised for power efficiency (L0 cache? Less execution units if I recall) and asynchronous mp.

 

[Jubei]

Given that they are power gated, what does it matter if the SoC has 4 or 2 A15s?

 

[Exophase]

If I stop posting on a forum like this I doubt anyone will miss me; now imagine how much anyone would bother about anyone else.

I'd miss you

The link provided uses 2 a15s and 3 a7s...that actually looks like a nice ballance for a smartphone IMO.

Your first mistake is thinking that this is a smartphone SoC and not considering what's good for other platforms. If you can only run those four Cortex-A15 cores at anything near 1.8GHz while in a tablet form factor (or even bigger, who knows what Samsung's aspirations are for this) then so be it. They're not Apple, they can't (yet) rationalize making a big phone and big tablet SoC separately and simultaneously. Those 2 A15 cores cost them ~6mm^2 which is far from trivial but not a total game changer. The 4 A7s on the other hand are a no-brainer, the cost over your hypothetical 1 A7 is under 3mm^2.

As has been told to you, going with a symmetric number of A15 and A7 cores is a good idea because otherwise you won't be able to reliably fall back on a simple mutual exclusive switch model. Otherwise I could see a strong argument for 2 A15 plus 4 A7. But overestimating OS support is a vital error (burned nVidia with the first wave of Tegra 2 tablets, that's for sure).

So if we want four hardware threads guaranteed by software they needed 4 A15s. Wanting four hardware threads is where the whole rest of the industry is at. AMD has it with most of the CPUs they sell and will be pushing it in tablets with Temash. Intel has it with their non-castrated IBs and it's moving into phone space with CloverTrail+. They do it with 2C/4T but there's no software problem there since it looks like 4 synchronous cores to the OS. Samsung wanting 4 hardware threads is nothing new, that started with Exynos 4.

nVidia's 4+1 decision with Tegra 4 makes no real sense. It's probably using the same hardware mechanism as big.LITTLE, ie two separate A15 clusters with a coherency link. They should let both run simultaneously under some condition but maybe there's some limitation with how they implement clocks or power that makes this impossible for them. And which would also make big.LITTLE impossible. But as usual they're trying to spin it as their own smart idea which has its own benefits.

I'm looking forward to some extensive real power numbers. nVidia saying that the companion core uses 40% less power @ 825MHz than a Tegra 3 A9 used at 1.6GHz doesn't impress me, because Tegra 3 @ 1.6GHz wasn't especially low power. This doesn't inspire great confidence that even a power optimized A15 layout can cut it for low power modes.

 

[Ailuros]

I didnt call them idiots so please dont put words in my mouth.

I never said you did; how about re-reading it again?

I also never mentioned octa being marketing or that it woukdnt work. .again...

Part of it is marketing and I know why I'm claiming it.

I only questioned the point of having 4 a15s with a7s...aside from the recommendations of using equal amounts of both cores. .due to Migration software not mature enough, I still think its over kill in a smartphone....tegra even more so because no a7s and no asynchronous mp.

I would have preferred the setup in the arm demo (running on android ics)...it loaded web pages instantly with only 2 a15s and no gpu..I fail to see what noticeable advantage another 2 eagle cores would achieve..other than sucking power.

Snapdragon is different because kraits consume less power than a15s.

Yes and we'll be moving in circles around the same part of the topic because you fail to read, comprehend and understand even the basics. Here's a video from ARM presenting CPU utilization on the Samsung octacore: http://www.computerbase.de/news/2013-02/arm-zeigt-samsung-exynos-5-octa-im-einsatz/

 

[french toast]

Given that they are power gated, what does it matter if the SoC has 4 or 2 A15s?

Because you have a7s... I know they can be power gated and down clocked via dvfs...but just use A7s...even 2+2 would have been more efficient.

Anyway as exophase has posted there are obvious reasons for this in software.
Im certainly not complaining about more performance.

 

[french toast]

I'd miss you



Your first mistake is thinking that this is a smartphone SoC and not considering what's good for other platforms. If you can only run those four Cortex-A15 cores at anything near 1.8GHz while in a tablet form factor (or even bigger, who knows what Samsung's aspirations are for this) then so be it. They're not Apple, they can't (yet) rationalize making a big phone and big tablet SoC separately and simultaneously. Those 2 A15 cores cost them ~6mm^2 which is far from trivial but not a total game changer. The 4 A7s on the other hand are a no-brainer, the cost over your hypothetical 1 A7 is under 3mm^2.

As has been told to you, going with a symmetric number of A15 and A7 cores is a good idea because otherwise you won't be able to reliably fall back on a simple mutual exclusive switch model. Otherwise I could see a strong argument for 2 A15 plus 4 A7. But overestimating OS support is a vital error (burned nVidia with the first wave of Tegra 2 tablets, that's for sure).

So if we want four hardware threads guaranteed by software they needed 4 A15s. Wanting four hardware threads is where the whole rest of the industry is at. AMD has it with most of the CPUs they sell and will be pushing it in tablets with Temash. Intel has it with their non-castrated IBs and it's moving into phone space with CloverTrail+. They do it with 2C/4T but there's no software problem there since it looks like 4 synchronous cores to the OS. Samsung wanting 4 hardware threads is nothing new, that started with Exynos 4.

nVidia's 4+1 decision with Tegra 4 makes no real sense. It's probably using the same hardware mechanism as big.LITTLE, ie two separate A15 clusters with a coherency link. They should let both run simultaneously under some condition but maybe there's some limitation with how they implement clocks or power that makes this impossible for them. And which would also make big.LITTLE impossible. But as usual they're trying to spin it as their own smart idea which has its own benefits.

I'm looking forward to some extensive real power numbers. nVidia saying that the companion core uses 40% less power @ 825MHz than a Tegra 3 A9 used at 1.6GHz doesn't impress me, because Tegra 3 @ 1.6GHz wasn't especially low power. This doesn't inspire great confidence that even a power optimized A15 layout can cut it for low power modes.

Yes obviously software has to play nice with it..I accept that and also 5410 is not necessarily a smartphone soc.

As a tablet soc it makes complete sense...although I hadnt realised just how small those a7s were. ..almost 'free'.
There seems to have been two points discussed...1) implementation of big.LITTLE.
2) what kind of performance a smartphone could use.

The soc design choices make complete sense to me now. ..the latter point I think makes octa more than enough performance for a smartphone.

I never understood nvidias shadow core philosophy. ..it looked intriguing when a9s were being used..less so with a15s.

 

[Turbotab]

Just some clarification for those who might be scratching their heads on this..

Games are bound by GPU and I/O. Even if the GPU can handle it it doesn't make sense to render beyond a certain number of frames per second because it won't make a difference to the user and the display won't be able to keep up, so it'll just waste power. GPU load on games is fairly negotiable, but CPU load nearly as much so, because a big chunk of it is stuff you agree on to make the game work (process internal events and input and stuff at a certain rate regardless of output frame rate). Since no one wants to make a game that requires top of the line new phone hardware just to work you don't get CPU requirements like this.

Web page loading is different because it's the kind of task that you want done ASAP, meaning you load the CPUs to the limit you can thread them to and you keep them there until the task is done. That task still end up taking under one second but that's forever compared to the amount of time a game dedicates to a frame.

Games probably do have periods of "get it done ASAP" too, like when loading a level, but that isn't their steady state.

I found a cool video which demonstrates your last point.

It shows the CPU utilisation during Sky Castle 3D demo, on a ST-Ericsson Novathor L8580. During in-game scenes, the CPU is often pegged at just 400 to 800 MHz.

Best to watch at 720p fullscreen, from 1.38 onwards.
http://www.youtube.com/watch?v=lM1PqGpFF0c