dreams do come true, arm big.little [4x4] exynos 5 octo


samsung just unveiled the exynos 5, 8 core big.little soc...developing
http://www.theverge.com/2013/1/9/3855478/samsung-announces-8-core-exynos-5-octa-mobile-processor'>www.theverge.com/2013/1/9/3855478/samsung-announces-8-core-exynos-5-octa-mobile-processor
While it is beautiful, LTE chips will have to be seperate still and therefore consume more power. Hopefully Samsung releases an LTE version of the Octa.
 
Sounds amazing.

But as some people say heat  and battery life might be a problem here.
 
Sounds amazing.

But as some people say heat  and battery life might be a problem here.
While peak consumption goes up, this gives more resources for lowering consumption of current workloads, and is on an intrinsically lower power process. So it's progress, at least compared to the current Exynos 5250.


Peak is not the limiter here, if peak is too high you limit it. They give a relatively high performance threshold to support a wider array of devices. Phones and even tablets won't necessarily be allowed to stress that. This is nothing new.
 
Sounds amazing.

But as some people say heat  and battery life might be a problem here.
While peak consumption goes up, this gives more resources for lowering consumption of current workloads, and is on an intrinsically lower power process. So it's progress, at least compared to the current Exynos 5250.


Peak is not the limiter here, if peak is too high you limit it. They give a relatively high performance threshold to support a wider array of devices. Phones and even tablets won't necessarily be allowed to stress that. This is nothing new.
I want to use the full power of what I've got :D .

Don't know about the Exynos 5250. If you say it's progress, then it is.

Phones might not  use the full power but if Pandora 2 would do so, would it still be usable concerning heat and battery life?

If not, it might make more sense to use something that's maybe not as powerful but we can fully use.
 
Sexy... this kind of stuff has me excited for the Pandora 2 due to competition.

Because even if we don't get the best SOC.. the fourth best is still going to be pretty damn awesome.

There is always going to be the next best thing. but the R&D for the mobile ARM units is going to lead to great things.
 
Last edited by a moderator:
I want to use the full power of what I've got :D .
When you buy a laptop with an Ivy Bridge it's the same chip as you can find in the highest end desktops. But it's not allowed to sustain the same performance under heavy load because it has a tighter thermal envelope.

It's no different here, except more responsibility may be on the end device manufacturer than the SoC maker to set the limits.
 
I want to use the full power of what I've got :D .
When you buy a laptop with an Ivy Bridge it's the same chip as you can find in the highest end desktops. But it's not allowed to sustain the same performance under heavy load because it has a tighter thermal envelope.

It's no different here, except more responsibility may be on the end device manufacturer than the SoC maker to set the limits.
Didn't know this before.

Thanks for reducing my ignorance.

:)
 
samsung just unveiled the exynos 5, 8 core big.little soc...developing

www.theverge.com/2013/1/9/3855478/samsung-announces-8-core-exynos-5-octa-mobile-processor
Absolutely not, this would be horrible for the Pandora 2.

Dual core is a maximum, an octa-core would be so inefficient that there's almost no hope of being able to overclock. High core counts help multi-tasking, not emulation performance. For emulation, you want your core count to be low and the core clock high. It doesn't matter how many cores you have because emulators are only able to take advantage of one CPU core in most cases (some emulators for PS2 and GameCube are written to utilize two cores, which is why I said dual-core). Core clock is absolutely vital to the performance of emulators, and with that many cores crammed into one SoC there's no way you'd be able to clock it to a reasonable speed without melting the device.

Not only would emulation be laughably slow on that thing, it would also be an extreme power hog and probably wouldn't overclock at all. It'd also require some sort of active cooling mechanism, because that thing will run HOT under load. I will cry if the devs put anything more than a dual-core in the Pandora 2, let alone that monstrosity.
 
Last edited by a moderator:
Absolutely not, this would be horrible for the Pandora 2.

Dual core is a maximum, an octa-core would be so inefficient that there's almost no hope of being able to overclock. High core counts help multi-tasking, not emulation performance. For emulation, you want your core count to be low and the core clock high. It doesn't matter how many cores you have because emulators are only able to take advantage of one CPU core in most cases (some emulators for PS2 and GameCube are written to utilize two cores, which is why I said dual-core). Core clock is absolutely vital to the performance of emulators, and with that many cores crammed into one SoC there's no way you'd be able to clock it to a reasonable speed without melting the device.

Not only would emulation be laughably slow on that thing, it would also be an extreme power hog and probably wouldn't overclock at all. It'd also require some sort of active cooling mechanism, because that thing will run HOT under load. I will cry if the devs put anything more than a dual-core in the Pandora 2, let alone that monstrosity.
More cores don't make the system less efficient just by existing. The cores you don't use can be turned off completely. When they're off they don't contribute to power consumption at all, nor do they impact the peak clock you can get.

The reason why there are 4 Cortex-A15 cores and 4 Cortex-A7 cores is not because Samsung expects them all to be running at the same time in any normal scenario, but to give you more scheduling flexibility which improves power consumption. For instance, emulators that run fine on CC Pandoras can use a single Cortex-A7 clocked at under 800MHz, with another Cortex-A7 core periodically active to take in background stuff to prevent anything from stealing process time from the emulator. More demanding ones can use one or two Cortex-A15 threads while delegating other stuff - like 3D driver coaxing, frame post-processing, audio handling etc on Cortex-A7 threads. Or anything that uses a software 3D render (DS emulation for instance) can divvy up the task among multiple Cortex-A7s. A single Cortex-A7 on Samsung's 28nm will tend to use significantly less power than the Cortex-A8 in the OMAP3 and even DM3730 Pandoras, while offering almost as good perf/MHz for a lot of tasks.

Let's look at a simplified scenario. Like you say dual core is good. Not just because high end emulators can heavily utilize both threads, but because often you'll have two strong threads, one strong thread + several weak threads, one string + one weak thread, one moderate + one weak thread, and one or two weak threads. In these cases you'd want Cortex-A15 + Cortex-A15, Cortex-A15 + Cortex-A7, or Cortex-A7 by itself (Cortex-A15 by itself generally only pays off if you save nothing at all in off-loading to an A7). You may never really need more than two cores active at a time. But you still need to have all four cores on the SoC, and you don't incur much added complexity in being able to run 3 or 4 of them simultaneously.

IMO a 2x Cortex-A15 + 4x Cortex-A7 solution makes a lot of sense and 4x Cortex-A15s seem like overkill, but if they have the die area to spend it doesn't hurt that much, unless it could have been better spent on L2 cache. The Cortex-A7s are comparatively tiny so having 4 of them barely makes a difference.
 
Last edited by a moderator:
Absolutely not, this would be horrible for the Pandora 2.

Dual core is a maximum, an octa-core would be so inefficient that there's almost no hope of being able to overclock. High core counts help multi-tasking, not emulation performance. For emulation, you want your core count to be low and the core clock high. It doesn't matter how many cores you have because emulators are only able to take advantage of one CPU core in most cases (some emulators for PS2 and GameCube are written to utilize two cores, which is why I said dual-core). Core clock is absolutely vital to the performance of emulators, and with that many cores crammed into one SoC there's no way you'd be able to clock it to a reasonable speed without melting the device.

Not only would emulation be laughably slow on that thing, it would also be an extreme power hog and probably wouldn't overclock at all. It'd also require some sort of active cooling mechanism, because that thing will run HOT under load. I will cry if the devs put anything more than a dual-core in the Pandora 2, let alone that monstrosity.
More cores don't make the system less efficient just by existing. The cores you don't use can be turned off completely. When they're off they don't contribute to power consumption at all, nor do they impact the peak clock you can get.

The reason why there are 4 Cortex-A15 cores and 4 Cortex-A7 cores is not because Samsung expects them all to be running at the same time in any normal scenario, but to give you more scheduling flexibility which improves power consumption. For instance, emulators that run fine on CC Pandoras can use a single Cortex-A7 clocked at under 800MHz, with another Cortex-A7 core periodically active to take in background stuff to prevent anything from stealing process time from the emulator. More demanding ones can use one or two Cortex-A15 threads while delegating other stuff - like 3D driver coaxing, frame post-processing, audio handling etc on Cortex-A7 threads. Or anything that uses a software 3D render (DS emulation for instance) can divvy up the task among multiple Cortex-A7s. A single Cortex-A7 on Samsung's 28nm will tend to use significantly less power than the Cortex-A8 in the OMAP3 and even DM3730 Pandoras, while offering almost as good perf/MHz for a lot of tasks.

IMO a 2x Cortex-A15 + 4x Cortex-A7 solution makes a lot of sense and 4x Cortex-A15s seem like overkill, but if they have the die area to spend it doesn't hurt that much, unless it could have been better spent on L2 cache. The Cortex-A7s are comparatively tiny so having 4 of them barely makes a difference.
I am aware of this, but for the best emulation performance there's no need to even have the extra cores in there. It'd also be cheaper to get a 2x A15 SoC that can clock up to 3 GHz than it would to get one with 8 cores that only clocks to 1.5-2 GHz. I'm not saying we couldn't make it work for the Pandora 2, but it's way more than what is necessary and will be quite expensive, and probably won't provide the maximum performance we could get from the system. The OMAP5430 looks like it will be the best for the application we're discussing. 
 
Last edited by a moderator:
I am aware of this, but for the best emulation performance there's no need to even have the extra cores in there. It'd also be cheaper to get a 2x A15 SoC that can clock up to 3 GHz than it would to get one with 8 cores that only clocks to 1.5-2 GHz. I'm not saying we couldn't make it work for the Pandora 2, but it's way more than what is necessary and will be quite expensive, and probably won't provide the maximum performance we could get from the system.
I don't think you are aware of it or you wouldn't be saying that more cores makes the SoC less efficient or capable of lower clock speeds because neither of those things are true.

There is no real trade-off where you can clock a single core higher by including fewer cores overall, so stop claiming that the maximum performance goes down. You're not going to get 3GHz Cortex-A15s on this process in power consumption a handheld can handle either way. And you don't actually know how much it'll cost.

As for emulation performance I disagree with you, there are at least some cases where more than two cores can help and there are definitely cases where a Cortex-A15 + Cortex-A7 mix is preferable to a pure Cortex-A15 mix. SSF and PCSX2 both can greatly benefit from more than two hardware threads, for instance. And like I said, software rasterization can easily scale to multiple cores.
 
Last edited by a moderator:
takato14,  Are you saying that the Next Pandora (if it ever happens) will only be used for emulation?.  Given the strong likely hood that it will also run Android , it needs the extra power to run the top of the line Android games.  Thinking of the next Pandora as a game emulation machine only is not seeing the whole picture.
 
Last edited by a moderator:
takato14,  Are you saying that the Next Pandora (if it ever happens) will only be used for emulation?.  Given the strong likely hood that it will also run Android , it needs the extra power to run the top of the line Android games.  Thinking of the next Pandora as a game emulation machine only is not seeing the whole picture.
The Pandora's main function is emulation. It can also do anything a computer does and run Android on the side. The Pandora 2 should be powerful enough to run Android but that should not be its primary function.

No Android game is going to be written to take advantage of 8 cores. Period. It's unnecessary to have that many cores in this device and it's just driving the cost up without reason.
 
I'd say: get the SOC with the best open-source driver support, after that the one with the best single-core performance, then one with the best power management, then with most cores. (in order of priority)
 
The Pandora's main function is emulation. It can also do anything a computer does and run Android on the side. The Pandora 2 should be powerful enough to run Android but that should not be its primary function.

No Android game is going to be written to take advantage of 8 cores. Period. It's unnecessary to have that many cores in this device and it's just driving the cost up without reason.
It. Does. Not. Matter.

You don't need to use all 8 cores at the same time to benefit from a 4 + 4 arrangement. You simply need to use a combination of some of one core and some of the other. If 4 of either are ever useful then you've justified 4 + 4. I don't know how you're not getting this. It's the start of what has long since been referred to as "dark silicon" - since transistor budgets are growing much more quickly than power budgets you end up with a chip that's optimized so that you only use parts of it at any given time. It's like the old figure that you only use some small percentage of your brain at any given time, yet that doesn't mean it isn't mostly there for a reason.

You keep going on about driving up costs but you don't know what the actual costs are. Consider the following: a Cortex-A15 core is about 4x the size of a Cortex-A7 core (http://chip-architect.com/news/2013_core_sizes_768.jpg). Tegra 4 has 4 + 1 Cortex-A15 cores. So the die investment is similar to what you'd get with 4 A15 + 4 A7. They also have a huge investment in GPU, basically scaling the Tegra 3 allotment by 6 times even though the transistor density didn't even improve by 2x. Yet the whole chip is around 80mm^2, about the same size as Tegra 3 and while not tiny not especially large either.

I don't know if Samsung's 28nm is as dense as TSMC's, but assuming it isn't off by a huge factor this chip shouldn't be considerably larger.
 
Last edited by a moderator:
Back
Top