Snapdragon 820 GeekBench

[Hans de Vries]

Good progress for Kryo, especially if indeed combined with a
significantly lower power consumption as promised. (30%)

Snapdragon 810 vs Snapdragon 820

http://browser.primatelabs.com/geekbench3/compare/4209632?baseline=4159755

Apple's Twister is the new king of the hill CPU however,
extending Apple's lead even further:

Snapdragon 820 vs Apple A9

http://browser.primatelabs.com/geekbench3/compare/4159755?baseline=4198667

I like the memory performance of the 820. It seems that
we have 64 bit at 3.733 GHz here. Faster as the A9 in
the iPhone 6s. Will be good for Graphics.

Hans

 

[Arachnotronic]

Good progress for Kryo, especially if indeed combined with a
significantly lower power consumption as promised. (30%)

Snapdragon 810 vs Snapdragon 820

http://browser.primatelabs.com/geekbench3/compare/4209632?baseline=4159755

Apple's Twister is the new king of the hill CPU however,
extending Apple's lead even further:

Snapdragon 820 vs Apple A9

http://browser.primatelabs.com/geekbench3/compare/4159755?baseline=4198667

I like the memory performance of the 820. It seems that
we have 64 bit at 3.733 GHz here. Faster as the A9 in
the iPhone 6s. Will be good for Graphics.

Hans

Twister indeed looks to be ahead of the pack, and in some of the more interesting sub-tests (Ray Tracing and Lua, N-Body, etc.) it's very ahead.

 

[Thanatosis]

Why is Kyro's memory subsystem 2-3x faster than Twister? They must be using dual channel 128bit DDR4 or HBM to get that kind of speed. Surprising to see Apple so far behind here.



Also curious why Kyro is putting up better multicore numbers despite being effectively dual core. It's slower in single core, so I guess the smaller A53s are pitching in? That doesn't sound like big.LITTLE

 

[el etro]

Still a win for Qualcomm, even if not outperform the monster A9. Good job. Android phones fans(like me!) will win at the end. Ready to trade my Krait phone for a Kryo phone in 2016.

 

[Arachnotronic]

Why is Kyro's memory subsystem 2-3x faster than Twister? They must be using dual channel 128bit DDR4 or HBM to get that kind of speed. Surprising to see Apple so far behind here.

It's not 2-3x faster check the numbers again. S820 is using faster clocked LPDDR4 than A9 uses, that's pretty much it. I am sure with A10 Apple will adopt LPDDR4-4266 to get a nice bandwidth boost.

Also curious why Kyro is putting up better multicore numbers despite being effectively dual core. It's slower in single core, so I guess the smaller A53s are pitching in? That doesn't sound like big.LITTLE

There are four Kryo cores in HMP configuration; two high clocked Kryos (2.2GHz) and two lower clocked Kryos (1.6-1.7GHz). It is not "effectively dual core"

Also we should note that A9 has shipped in the tens of millions at this point and probably entered into production late last year/early this year. Apple's "last gen" CPU (A10 is probably at the same stage S820 is now) is ahead of a part that won't show up in devices until March/April.

Apple is a freaking beast.

 

[Face2Face]

The single core performance looks much better vs. the 810. I'm interested to see how well it holds up from thermal point of view.

 

[monstercameron]

So that means the 620 snapdragon could be faster?
Iirc 4*a72 +4*a53. Please correct me if I'm wrong.

 

[kpkp]

So that means the 620 snapdragon could be faster?
Iirc 4*a72 +4*a53. Please correct me if I'm wrong.

All being equal (or no thermal limits) it could be faster in MT, but 620 will be build on 28nm.

 

[krumme]

820 will probably have excellent dsp and lte as history shows.
Imo perf is plenty and more single thread perf is just stupid marketing. Its all about perf w and cost (mm2) then. There is zero meaning - user benefit - in boosting single thread perf further than this using loads of expensive mm2 for cache as apple or super wide or deep cpu arch. Stop this single thread monopoly marketing race.
Old anandtech style. Please Andrei kill it. Lets get leaner cores better coding and more fixed function hardware.

The best hopes for 820 beeing a new 800 !

 

[Andrei.]

Memory performance in Geekbench has nothing to do with the actual memory configuration used. If the CPU arch can't take advantage of the theoretical bandwidth, it's all wasted or goes to other SoC blocks. According to the benchmarks it seems Kryo has a massively vamped up memory architecture, so that's good.

Please Andrei kill it.

The best hopes for 820 beeing a new 800 !

The S800 was defined by not having any competition at its efficiency level, I don't think that'll happen this time around. In any case I'm prepped with S801/S805/S808/S810v2/S810v2.1 numbers and just waiting on getting S820 hardware.