CHERI webinar banner
WP_Term Object
    [term_id] => 157
    [name] => EDA
    [slug] => eda
    [term_group] => 0
    [term_taxonomy_id] => 157
    [taxonomy] => category
    [description] => Electronic Design Automation
    [parent] => 0
    [count] => 3923
    [filter] => raw
    [cat_ID] => 157
    [category_count] => 3923
    [category_description] => Electronic Design Automation
    [cat_name] => EDA
    [category_nicename] => eda
    [category_parent] => 0

Will the Snapdragon 820 Overheat?

Will the Snapdragon 820 Overheat?
by Vinay Shivakumar on 10-30-2015 at 7:00 pm

 So, I happened to read an article today about overheating issues on the snapdragon 820 from rather flaky sources. To guess the answer to this question, one has to understand the snapdragon 810 “heat-gate” and the sequence of events that led to it.

1) Apple moved to ARMV8 ahead of everyone else. Suddenly 64 bit became a compulsory marketing sticker. I believe Qualcomm was caught wrong footed with their custom core development (because of this or otherwise) and had to stick with run of the mill A57s to get the 64bit sticker and have a product on time. The single thread performance of the Apple custom core left (and continues to leave) everyone else in the dust.

Anandtech did an excellent article detailing the performance differences between the a15 and A57. In summary the A57 boosts cpu performance by roughly 20-30%. The A15 vs Krait 400 are roughly in the same ballpark in integer performance (say 5%). The A57 increases power quite a bit, which makes on believe that it was never meant to be a quad core cluster for mobiles. ARM might have also rushed to the A57 to market , since they were surprisingly a little behind Apple ( the A72 is significantly more power efficient , even on the same process).

2) 20nm being a planar process was not all good news. To get a performance kick out of 20nm needs careful design planning. The newer process nodes do not necessarily give a large improvement in the high voltage speed of the devices. This meant Qualcomm might have had to move to use very high voltages to achieve high frequencies/ benchmark scores. Also 20nm being a planar process leaks a LOT more than 28nm.

These are the pin voltages for the snapdragon 810 (8994) v2.1 (the revision used in oneplus two) dug out from a random pastebin here:

[TABLE] cellspacing=”3″
| Cortex A53
| 384Mhz
| 815mV
| 1552Mhz
| 1.125V
| Cortex A57
| 384Mhz
| 830mV
| 1766Mhz
| 1.040V

These are the voltages for the snapdragon 801 (8974AC) (Dug out from here)

[TABLE] cellspacing=”3″
| Krait 400
| 300Mhz
| 0.75V
| 2457Mhz
| 0.95V – 1.05V
| Assuming bin PVS 6 or better

From the tables it is clear that the 810 A57 implementation hits much lower frequencies at much rather high voltages. These high voltages would have offset most of the decrease in dynamic power due to the process switch. These numbers make one believe that this might have been a quick and dirty implementation, rather than a planned one.

3) Also Qualcomm had to do the stupid 8 core thing, since Mediatek and Samsung were doing just that. More the merrier right ???

If Qualcomm had launched the snapdragon 810 without boosting the frequencies to insane numbers at insane voltages , the performance would have looked like this – ~20% uplift in total on single core performance compared to 801 and well behind the apple A8, not all that impressive…
Looking at the snapdragon 820, looks like Qualcomm should not have the same issues that were present with the snapdragon 810:

  • The 14nm finfet process will give a huge uplift in frequency at “sensible voltages” ie. ~0.9V and below. If Qualcomm chooses not to release the 820 with insane voltages, things should be peachy. However, the 3GHz boost gives me jitters. Qualcomm should be sensible on how long they choose to stay at those high frequencies.

  • Sensible, Quad cluster of custom Kryo cores. Not 16 cores. Thanks!! Very promising benchmark leaks.

So, the short answer is that it should not overheat, IF Qualcomm remains sensible with power management in non-benchmark scenarios. Like the time they launched the snapdragon 810, they are not stuck between a rock and a hard place. They don’t need to do the things they had to do with the 810.

Whether they choose to do that, however, is anyone’s guess…

Follow the adventures of SemiWiki on LinkedIn HERE!

Share this post via:


0 Replies to “Will the Snapdragon 820 Overheat?”

You must register or log in to view/post comments.