Snapdragon 800 Benchmarks


interesting numbers...


so about the power consumption? hmm...seems that this will be relatively power hungry. Any opinions?
As far as I know, the Krait (CPU of the Snapdragon) uses less power than the A15.
even so its gotta be clocked higher than a15 -nv t4- to outperform it...right? the t4 should be clocked at 1.9GHz  to compensate for the lower ipc from krait(?) so the clock would have to be bumped, maybe the 2.3GHz figure floating around is its clock speed.

so the power savings might be negated.
 
Last edited by a moderator:
interesting numbers...


so about the power consumption? hmm...seems that this will be relatively power hungry. Any opinions?


As far as I know, the Krait (CPU of the Snapdragon) uses less power than the A15.


even so its gotta be clocked higher than a15 -nv t4- to outperform it...right? the t4 should be clocked at 1.9GHz  to compensate for the lower ipc from krait(?) so the clock would have to be bumped, maybe the 2.3GHz figure floating around is its clock speed.


so the power savings might be negated.
Yes, the Krait seems to be a bit slower but uses a lot less power, as far as I know.
 
Note that the Exynos 5 Octa has both a Cortex-A15 and a Cortex-A7 (actually 4 of each). While the A15 is quite power-hungry, a Cortex-A7 at 28nm should be very low-power, and it could be just about powerful enough to run most Pandora 1 stuff.

In terms of power consumption, big.LITTLE seems like a really good idea. Too bad only Exynos 5 Octa and HiSilicon K3V3 are doing this, although OMAP4/5 and Tegra3/4 are doing something similar.

On the other hand, what is nice about the Snapdragon is that 1) it has integrated wireless, which means low power wifi (because it's 28nm too), probably less potential issues with dodgy wifi, and saving some PCB space, and 2) its GPU is Adreno, which has a FOSS driver (freedreno) currently available.
 
The remaining unknown will be the Baytrail SOCs coming out 3rd quarter. Should be interesting to see their benchmarks
 
Last edited by a moderator:
Baytrail sounds promising indeed - x86-64 in a low-power 22nm SOC!

I would be very careful when comparing benchmark results across architectures though - some stuff may be much better optimized for x86 than for ARM, for instance - e.g. SunSpider is just as much a benchmark of the JavaScript implementation you're using as it is an actual hardware benchmark.

Raw power is not my main concern - any of these new SOCs will be very powerful.

I'm more interested in some numbers or information about:

- power consumption (idle / low load / high load)

- availability in relatively low quantities

- price

- FOSS support
 
In terms of power consumption, big.LITTLE seems like a really good idea. Too bad only Exynos 5 Octa and HiSilicon K3V3 are doing this, although OMAP4/5 and Tegra3/4 are doing something similar.

big.LITTLE seems like it has a lot of potential. Sadly, Exynos 5410's implementation is probably broken, preventing you from running the A15 and A7 clusters at the same time. Since it's the only commercial solution out now I'm worried that this will hurt big.LITTLE's adoption and progress. Renesas is coming out with an SoC too but I doubt it'll be featured in mobile products.


OMAP4/5 aren't actually doing anything comparable.. they have a couple Cortex-M3/M4s on the SoC, but they're just there to help offload some special embedded tasks like coordinating video decode. The OS can't schedule code freely on them because they're not cache coherent with the main cores nor probably even in the same address space (they may not even have direct access to main memory) and they wouldn't support the same instruction set anyway. It's really no different from what TI has done for a long time, OMAP3/DM37xx/etc actually have an ARM9 in the video block that most people aren't aware of.


nVidia is doing something closer with its companion core, but it's very limited compared to what big.LITTLE is supposed to be capable of.


Qualcomm's advantage, power-wise, is that all of the CPUs run on independent clock and voltage plains. So when you're using one heavily and another lightly you don't have to run an entire cluster at the higher clock and voltage. It does have the downside of keeping the L2 on a different power/clock domain (higher latency). Still, I think that this or big.LITTLE will be great for the common use case where you have one very hungry thread and a few weaker threads. I think this could fit some emulators well.

_wb_ said:
Baytrail sounds promising indeed - x86-64 in a low-power 22nm SOC!
I don't think I've seen confirmation, but early leaks showed that Baytrail (and probably Merrifield by extension) will have Silvermont's x86-64 support fused off - just like Clovertrail/Clovertrail+ and Medfield do today. I think Intel may be doing this to save power but going in this direction for 2014 mobile platforms is jarring.
 
Last edited by a moderator:
Thanks for the clarification on OMAP4/5 and Tegra's "companion" cores.

Exactly how broken is the Exynos 5410? Can you run 2 A15's and 2 A7's at the same time, but just not all 8 cores at the same time - e.g. you can activate at most 4 cores and you can choose for each whether it's an A7 or an A15? Or can you not even do that - only 1 to 4 A15's or 1 to 4 A7's, but no mixing?

Anyway, I guess the Snapdragon should be at least as good in terms of power consumption as the Exynos 5410.

Bay Trail already sounds less promising if they stick to 32-bit.

Any idea of how prices of those SOCs will compare?
 
The flaw with Exynos 5410 is that cache coherency is (probably) broken between the two active clusters. So normal OS scheduling won't work if the two are running simultaneously. It also means that migrating has to flush everything to main memory instead of going through the interconnect for some period of time.

This means you can't run any A7s and A15s simultaneously. There are rumors that this was a big screw up for Samsung and that the mobile division wants to drop Exynos 5410 from phones entirely. Hence why most Galaxy S4s are using Snapdragon S600.

Sadly I have no idea whatsoever what pricing is like. Almost none of these SoC makers sell to low volume public distributors. You probably get varying prices with each order.
 
Well, right now I wonder what the future of the A15 is...?

It's been announced a long time ago, and as far as I remember it was expected that most phones end of 2012 will have an A15 as core.

Right now, only the Samsung S4 is available, but only some use the A15, others use the Snapdragon.

There are a few tablets or the Chromebook, but they have high-capacity batteries (way more than could fit in a normal smartphone) and battery life is just normal, so I wonder whether the A15 just uses a bit too much power for a phone..?

And the S4 is a pretty huge phone, so it probably has a pretty high-capacity battery as well.

It seems only two SoC manufacturers actually finished the A15 for mobile usage. Samsung and NVidia.

All others seem to have switched and only release it for industrial usage.

Of course people always want the most powerful phone - but does the speed in daily usage compared to the reduced battery life justify this?

I'm not saying this is a reason nor will I say these are facts (I'm not that much into that stuff), but the A15 sure has a slow start right now, especially when comparing it to the start the A9 had.
 
Well, right now I wonder what the future of the A15 is...?

It's been announced a long time ago, and as far as I remember it was expected that most phones end of 2012 will have an A15 as core.

Right now, only the Samsung S4 is available, but only some use the A15, others use the Snapdragon.

There are a few tablets or the Chromebook, but they have high-capacity batteries (way more than could fit in a normal smartphone) and battery life is just normal, so I wonder whether the A15 just uses a bit too much power for a phone..?

And the S4 is a pretty huge phone, so it probably has a pretty high-capacity battery as well.

It seems only two SoC manufacturers actually finished the A15 for mobile usage. Samsung and NVidia.

All others seem to have switched and only release it for industrial usage.

Of course people always want the most powerful phone - but does the speed in daily usage compared to the reduced battery life justify this?

I'm not saying this is a reason nor will I say these are facts (I'm not that much into that stuff), but the A15 sure has a slow start right now, especially when comparing it to the start the A9 had.
isnt a15 outdated already, its getting superseded by a12, a53 and a57. Also in the quest for higher performance arm soc makers have been bumping cores and clock speed faster than new processes are available. exynos 5 octa(high core) and 5 dual(relatively high clock) are good examples.
 
Taking everything into account that has been said above, I think Snapdragon is (at the moment) the best candidate SoC for a Pandora successor.

I'm going to take the liberty to move this thread to the Pandora 2 Spam subforum, since I don't really consider this topic to belong in "Offtopic General talk".

Automatic message


This topic has been moved from "Off topicOfftopic General talk" to "PandoraPandora 2 spam".
 
Last edited by a moderator:
Taking everything into account that has been said above, I think Snapdragon is (at the moment) the best candidate SoC for a Pandora successor.

I'm going to take the liberty to move this thread to the Pandora 2 Spam subforum, since I don't really consider this topic to belong in "Offtopic General talk".

Automatic message


This topic has been moved from "Off topicOfftopic General talk" to "PandoraPandora 2 spam".
to your point, on a logistics pov, will qcomm sell low volume?
 
Last edited by a moderator:
Well, right now I wonder what the future of the A15 is...?

It's been announced a long time ago, and as far as I remember it was expected that most phones end of 2012 will have an A15 as core.

Right now, only the Samsung S4 is available, but only some use the A15, others use the Snapdragon.

There are a few tablets or the Chromebook, but they have high-capacity batteries (way more than could fit in a normal smartphone) and battery life is just normal, so I wonder whether the A15 just uses a bit too much power for a phone..?

And the S4 is a pretty huge phone, so it probably has a pretty high-capacity battery as well.

It seems only two SoC manufacturers actually finished the A15 for mobile usage. Samsung and NVidia.

All others seem to have switched and only release it for industrial usage.

Of course people always want the most powerful phone - but does the speed in daily usage compared to the reduced battery life justify this?

I'm not saying this is a reason nor will I say these are facts (I'm not that much into that stuff), but the A15 sure has a slow start right now, especially when comparing it to the start the A9 had.
isnt a15 outdated already, its getting superseded by a12, a53 and a57. Also in the quest for higher performance arm soc makers have been bumping cores and clock speed faster than new processes are available. exynos 5 octa(high core) and 5 dual(relatively high clock) are good examples.
No, it isn't.

The A12, A53 and A57 are not available yet.

The A12 has less performance than the A15, btw, just a better battery time. It's better in my opinion, but it doesn't outdate the A15.

It's for a different usecase.

The first available devices with one of those will probably be mid to late 2014.

The A15 had been announced April 2012, now it's mid 2013 and there's not much available.

There are barely A15 devices available, and you think it's outdated already.

What's outdated for you - if something exists in design specs yet, but has not been produced so far?

to your point, on a logistics pov, will qcomm sell low volume?
This actually could be quite an issue.

I made a quick check using google, and it seems Qualcomm does not even list a distributors page or a page to get in contact when you want to buy it or samples.

This implies it's nigh impossible to get in low volumes.

Not  impossible, but very unlikely.
 
^ publicly announced is one thing, its seems like you are assuming they get access to the schematics -or other ee jargon- the same time as we get the announcement. its possible they could have gotten the newer cores a considerable amount of time before making a press release.

a15 kinda feels like a flawed product, maybe it just fell short in the design phase or an issue with sammys 32nm implementation but its just not suited for battery operated devices within arms perview [or maybe that was the point?]
 
Last edited by a moderator:
Well, right now I wonder what the future of the A15 is...?

It's been announced a long time ago, and as far as I remember it was expected that most phones end of 2012 will have an A15 as core.
Every processor ARM has made in recent memory has been announced a good 2-3 years before it has been implemented in anything. That's just how their license model operates. Cortex-A8 was announced in October 2005 and only shipped in devices in late 2008. Cortex-A9 in December 2007, only shipped August 2010. Cortex-A15 in October 2010, shipped in October 2012 (so actually it was the fastest turnaround yet, at least ignoring older CPUs that I don't know much about who shipped first with)

And when it was announced they explicitly said that it'll show up in devices in 2013. It was later revised to late 2012, which is exactly what happened.

Right now, only the Samsung S4 is available, but only some use the A15, others use the Snapdragon.
big.LITTLE and ARM7 were both announced well after Cortex-A15 and were expected to hit quite a bit later. Exynos 5410 came out earlier than I expected it to anyway, which may be why it's so broken.

There are a few tablets or the Chromebook, but they have high-capacity batteries (way more than could fit in a normal smartphone) and battery life is just normal, so I wonder whether the A15 just uses a bit too much power for a phone..?

And the S4 is a pretty huge phone, so it probably has a pretty high-capacity battery as well.
In battery life tests it doesn't do badly precisely because of the A7s. It'd do better if the SoC actually worked right and the OS support was more robust (Samsung isn't great at this). A working big.LITTLE SoC would not be bad for something like Pandora.

It seems only two SoC manufacturers actually finished the A15 for mobile usage. Samsung and NVidia.

All others seem to have switched and only release it for industrial usage.
Two other SoC manufacturers announced A15 SoCs a long time ago: TI and ST-Ericcson. The reason why these didn't come to market for mobile devices is not because A15 tanked for mobile devices but because these companies withdrew from mobile devices completely. Freescale has too. Only Samsung and nVidia are positioning Cortex-A15 for phones because they are literally the only two SoC manufacturers that use bleeding edge CPUs (ie, not cheap Chinese SoCs) that aren't their own in-house designs. That's not A15's fault; although some people like to say that it drove the other companies to make custom CPUs they actually made that decision long before that, some people don't realize Qualcomm has been using custom CPUs for several years. In a couple generations nVidia will be using custom too. So maybe there won't be a market for high end ARM CPUs in mobile platforms period (who knows what Samsung will do), and maybe ARM realizes this and is adjusting their designs to be more useful in other markets.

Of course people always want the most powerful phone - but does the speed in daily usage compared to the reduced battery life justify this?
If you only use it when you need it? Yes. Idle consumption on pure A15 SoCs is fine. Having A7s helps.

I'm not saying this is a reason nor will I say these are facts (I'm not that much into that stuff), but the A15 sure has a slow start right now, especially when comparing it to the start the A9 had.
Selective memory. TI announced OMAP4 would be in phones by the end of 2010, didn't hit until months later. And for quite a while it was just nVidia (who initially sucked because they didn't have NEON), TI, and Samsung. Freescale and ST-E took forever to release theirs, but it doesn't really matter since they're out of the game now. I'm sure Chinese SoCs will start using A15 eventually like they started using A9.
 
Last edited by a moderator:
_wb_ said:
Baytrail sounds promising indeed - x86-64 in a low-power 22nm SOC!
I don't think I've seen confirmation, but early leaks showed that Baytrail (and probably Merrifield by extension) will have Silvermont's x86-64 support fused off - just like Clovertrail/Clovertrail+ and Medfield do today. I think Intel may be doing this to save power but going in this direction for 2014 mobile platforms is jarring.
Mmm...  maybe to save power, maybe more likely to differentiate product lineup?

Use the same process to produce both parts.  One gets used in phones and tablets (32bit only).  The other gets put in embedded systems and high density server clusters (64bit).

Think about it - if a certain % of CPUs fail, then some are going to fail in a critical area (deal) others are going to fail in an area or core that can be 'shut off' and then still sell the chip as a lesser product.  In this hypothetical case, only the 'perfect' ones go into the server (Xeon?) lineup.

Now, the big question - would the chips put into the server (Xeon) lineup still retain their otherwise vestigial mobile device capabilities? It could be entertaining to have a handheld mobile device with a 64bit Xeon CPU and 32GB RAM and blah blah blah...  Of course the CPU part could very well cost $1000 or more at that point.

If the 32bit 22nm SOC is viable and competitive to same availabilty ARM from a perf/watt, cost and max thermal perspective - I don't see any reason to exclude it.
 
Back
Top