Qualcomm SoC & ARMv8 custom core discussions

Does anybody know why this SOC is having so many heat issues? Is this a repeat of the infamous 90nm Pentium 4 issue or something?
 
Wonder what happened... Qualcomm isn't exactly a slouch at making chips.

I saw a comment from someone who claimed to be a Qualcomm employee saying that the work was more outsourced than usual. Maybe so they could keep their main teams focused on Kryo.

It didn't really help that Qualcomm had no prior experience with implementing ARM's flagship cores. And their lower tier implementations (ARM11, A7, A53) may have used ARM hard macros.
 
I think it's that, plus an element of having to get it out of the door quickly regardless of whether it was designed entirely in-house or not.
 
I saw a comment from someone who claimed to be a Qualcomm employee saying that the work was more outsourced than usual.
That guy was complaining about sourcing/using ARM's architecture instead of in-house, which is totally besides the point and really makes me doubt he has any clue of the issues.
Qualcomm has been using ARM's cores for a very long time in the low and mid-range, so heat issues with the S615 are not really explained by "inexperience". They don't use ARM macros.

Rys is right that it was rushed / changed last-minute. The whole industry fucked up all due to the 64bit marketing nonsense and absolutely horrible management overriding engineering decisions. Besides Samsung's 7420 basically every other vendor basically had a wasted year in 2015 and things will only start getting back on track in the coming generation.
 
That guy was complaining about sourcing/using ARM's architecture instead of in-house, which is totally besides the point and really makes me doubt he has any clue of the issues.

That's not what I remember, are you sure we're talking about the same post?

Qualcomm has been using ARM's cores for a very long time in the low and mid-range, so heat issues with the S615 are not really explained by "inexperience". They don't use ARM macros.

I'm not convinced experience with the lower end cores directly translates to the higher end ones which are significantly larger and more complex. I could see this still being a blindspot for them.

Do you have a source that they didn't use hard macros for these parts?
 
That's not what I remember, are you sure we're talking about the same post?
No but you worded it exactly like one claimed employee was complaining about stuff. Might be two different people.
I'm not convinced experience with the lower end cores directly translates to the higher end ones which are significantly larger and more complex. I could see this still being a blindspot for them.
I don't think the size of the IP has any impact on the quality of the physical design, maybe a bigger core takes more time but that's about it. Again, I can be wrong here but that's what I understand from researching the matter over the last year or so.
Do you have a source that they didn't use hard macros for these parts?
Yes, pretty certain. Users of ARM's hard macros for seem to be limited to Chinese vendors. Any of the big vendors (QC, Samsung, MediaTek) have their own design teams.
 
Yes, pretty certain. Users of ARM's hard macros for seem to be limited to Chinese vendors. Any of the big vendors (QC, Samsung, MediaTek) have their own design teams.

I wonder given Qualcomm's unique position in this space; unlike Samsung, MediaTek or nVidia their ARM cores (until pretty recently) were used only in lower end offerings while their custom cores were used in the medium and high end. They haven't exactly pushed aggressive clock speeds with their Cortex-A5, A7, or even A53 offerings. While there's no doubt that they can do physical implementation I wonder if for these products they don't feel that ARM's implementations are good enough and that their design efforts are better spent focusing on their own cores.
 
After seeing this tweet I suspect that Qualcomm will finally release S820 and reveal all the details this Tuesday. This would coincide with rumored announcement of s820 running xiaomi device on 11th of November.
What do you think?
 
Was reading the anandtech nexus 5x review. Never knew the 808/810 could throttle so bad as to shutdown the A57s.
That's old news though, arstechnica did some article half a year ago showing the same thing: http://arstechnica.com/gadgets/2015/04/in-depth-with-the-snapdragon-810s-heat-problems/
Except of course the problem is worse with 4 A57 cores... Albeit it seems the power management is a bit different, might be driver differences or something else (like having slightly different allowed surface temperatures).
That's why I always considered the 810 pointless - the thing has just about the same sustained performance as the 808, the 810 has higher peak performance but the peaks are short enough (compared to the 808) that it probably doesn't really matter (though the 810, being the high-end option, has some other benefits, like a faster ISP).
 
Anandtech's article is up. 2+2 kryos differing in cache/mhz and l3 cache.

Kinda cool that ultrasonic finger print works thru aluminum.
 
Ah interesting no big.little (at least in any meaningful sense). I guess it's a similar concept to the X20 but without the need of a dedicated interconnect. I wonder if there's a different synthesis between the two pairs (I think the X20 was like this too). Regardless, the power curves of the 820 should be intriguing.

I'm eagerly awaiting your analysis Neb! :smile:
 
Back
Top