Qualcomm 615, an arm littleLittle?

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

lopri

Elite Member
Jul 27, 2002
13,314
690
126
This thread is hilarious! Reminds me of all things ______ about this sub-form.

IIRC A9 isn't much faster than updated A7 (now rev. 4?) while consuming more power. The hierarchy goes something like:

A57 > A15 > A53~A9 > A7 > A8

What might have been marketing-driven is not the core count but rather 64-bitness at this point in time, but seeing that its introduction overlaps with the 20nm transition (and the benefits accompanying it) I guess it is something that had to be done. In any case I do not remember OEMs or Qualcomm advertising "8-core monster" yet when it comes to S615. Maybe between themselves? Who knows, I have no idea. Oh, Motorola did it once with its "X8 Engine" in Moto X (2013) and it was kind of laughed at.

Average consumers could care less, though. :D
 

Exophase

Diamond Member
Apr 19, 2012
4,439
9
81
What I don't understand is why would you not just scale up the voltage or scale down the voltage with dynamic voltage frequency scaling if the chips are identical on a transistor level. You can always scale down the 1.7 ghz part to lower clock speeds such as 1 ghz or something like a few hundred mhz. The only way this would make sense to me is that it is faster to switch from one cluster to the other cluster and then turn off the high power cluster instead of switching the voltage and the frequency. I seriously doubt this is the case for you would have to copy all that data from one cache to the other and that would have latency.

With two clusters in this arrangement you can run simultaneously (for example) 2 cores on the faster cluster and 2 cores on the slower cluster. If this is matches your workload requirements (as is often the case - look at core loading and you'll see it's often far from uniform, although there can be dependencies that prevent you from actually maintaining it at different clocks) this will be more efficient than running all four cores on the faster cluster, where everything will have to be the same clock and voltage.

Qualcomm is an obvious believer in asynchronous DVFS, since both Krait and Scorpion employed it at a per-core level.

Or maybe you don't realize that you can run both clusters at the same time, with varying numbers of cores enabled? This isn't strictly about switching from one cluster to the other.
 

gdansk

Diamond Member
Feb 8, 2011
4,680
7,906
136
This thread is hilarious! Reminds me of all things ______ about this sub-form.

IIRC A9 isn't much faster than updated A7 (now rev. 4?) while consuming more power. The hierarchy goes something like:

A57 > A15 > A53~A9 > A7 > A8

What might have been marketing-driven is not the core count but rather 64-bitness at this point in time, but seeing that its introduction overlaps with the 20nm transition (and the benefits accompanying it) I guess it is something that had to be done. In any case I do not remember OEMs or Qualcomm advertising "8-core monster" yet when it comes to S615. Maybe between themselves? Who knows, I have no idea. Oh, Motorola did it once with its "X8 Engine" in Moto X (2013) and it was kind of laughed at.

Average consumers could care less, though. :D
8-core processors seemed all the rage (especially in China): Allwinner A80 & A83T, MediaTek MT6592 & MT6595, HiSilicon's KIRIN 920 & 925, and Samsung's Exynos 5410 & 5420 & 5422 & 5800 & 5430 & 5433/7410. So far though Apple's A8(X) & Nvidia's K1 prove the big-core supremacy.

Some of these a ridiculously stupid, such as the 8 core Cortex A7 MT6592 but they sell well so they shall be made. At least the Snapdragon 615 is an interesting 64-bit and asymmetric design.
 
Last edited:

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126
With two clusters in this arrangement you can run simultaneously (for example) 2 cores on the faster cluster and 2 cores on the slower cluster. If this is matches your workload requirements (as is often the case - look at core loading and you'll see it's often far from uniform, although there can be dependencies that prevent you from actually maintaining it at different clocks) this will be more efficient than running all four cores on the faster cluster, where everything will have to be the same clock and voltage.

Qualcomm is an obvious believer in asynchronous DVFS, since both Krait and Scorpion employed it at a per-core level.

Or maybe you don't realize that you can run both clusters at the same time, with varying numbers of cores enabled? This isn't strictly about switching from one cluster to the other.

Yes this is possible, but I do not know how beneficial this will be since to my understanding you can not adjust the frequencies or the voltage of those individual cores (you can do that on other designs such as krait). Thus you have to be running those 2 cpus at normal voltage on the higher mhz cluster.

I understand there may be technical things I do not understand for I am not an engineer designing chips. I just hope the design makes sense from a technical perspective and not just something done because it is easy to synthesize (or quick to synthesize for time to market standpoint), or done for marketing, or for some other reason that causes compromises to be made and you getting inferior performance. Silicon is very important and thus I just want my silicon to be the best it can be for the money we spend.
 

krumme

Diamond Member
Oct 9, 2009
5,956
1,596
136
Platform strategy is crazy important in product development and production. It makes tons of sense and its one of the ways to control creative engineers that always want to invent the perfect wheel for each new situation :)
 

Exophase

Diamond Member
Apr 19, 2012
4,439
9
81
Yes this is possible, but I do not know how beneficial this will be since to my understanding you can not adjust the frequencies or the voltage of those individual cores (you can do that on other designs such as krait). Thus you have to be running those 2 cpus at normal voltage on the higher mhz cluster.

So this is still better than having to run all cores at the same voltage and clock speed. It's a compromise between the two extremes. It sounds like you're saying if it's not fully asynchronous, why bother having it anything less than fully synchronous?