Snapdragon 820 & Exynos 8890

[lopri]

IMO, no. Mali GPUs have not been known to be a versatile performer; they often have great specs and high theoretical maximums but when it comes to actual stuff they never seem to stretch their full potentials. Samsung is probably trying to match Adreno 530 there. I would welcome 4K if it did not negatively affect performance and battery, but we now know that is not the case. It is unlikely Samsung will risk there.

But yes, it looks real beefy. T880 will have 50% more ALUs, which will bring 40% or so increase in performance v. T760, and the 8890 will have 12 "cores," up from 8 in the 7420's T760MP8. Depending on clock speed, 70~80% performance jump seems realistic.

 

[dawheat]

IMO, no. Mali GPUs have not been known to be a versatile performer; they often have great specs and high theoretical maximums but when it comes to actual stuff they never seem to stretch their full potentials. Samsung is probably trying to match Adreno 530 there. I would welcome 4K if it did not negatively affect performance and battery, but we now know that is not the case. It is unlikely Samsung will risk there.

But yes, it looks real beefy. T880 will have 50% more ALUs, which will bring 40% or so increase in performance v. T760, and the 8890 will have 12 "cores," up from 8 in the 7420's T760MP8. Depending on clock speed, 70~80% performance jump seems realistic.

The Exynos 7420 GPU seems to handle the S810 GPU pretty handily in off-screen results. Regardless, I do hope this points to Samsung going with lower clocks most of the time to maximize efficiency. Sure burst performance is important, but would be nice to see an Android SOC that can maintain good extended performance without a huge power or heat penalty.

 

[lopri]

I agree. Especially for the GPU, its sustained performance should be a priority. Leave overclocking to users, and follow the voltage curve to find a sweet spot where everything stays cool.

 

[lopri]

A new GB entry of the S820. (Edit: It was run in 32-bit mode, which might partially explain the lower scores)

https://browser.primatelabs.com/geekbench3/4374813

Comparison with previously uploaded result

https://browser.primatelabs.com/geekbench3/compare/4159755?baseline=4374813

Lower score which appears to be due to irregular floating-point test results. If the Geekbench tests floating-point performance after integer performance, a possible explanation of the lower floating-point score is.. throttling caused by overheating?

Memory score is even higher than the previous record, though.

 

[lopri]

AnandTech has a preview up. (S820)

http://www.anandtech.com/show/9837/snapdragon-820-preview

I did not have a chance to read it yet but just giving heads up for those who do not frequent the front page. (There is also Huawei Mate S review by Mr. Frumusanu that I did not know about!) I will try to read them and post my thoughts here later.

I am surprised that Qualcomm was eager to unveil the chip's albeit preliminary performance characteristics this early, when the devices running on S820 are still months away.

 

[Commodus]

AnandTech has a preview up. (S820)

http://www.anandtech.com/show/9837/snapdragon-820-preview

I did not have a chance to read it yet but just giving heads up for those who do not frequent the front page. (There is also Huawei Mate S review by Mr. Frumusanu that I did not know about!) I will try to read them and post my thoughts here later.

I am surprised that Qualcomm was eager to unveil the chip's albeit preliminary performance characteristics this early, when the devices running on S820 are still months away.

It's an attempt to please investors, phone makers and enthusiasts. Qualcomm wants to show that it has better hardware in the pipeline and line up customers that might have been on the fence.

My concern is that this could backfire. After all, the AT benches show the 820 trailing behind Apple's A9 in most respects. That's not a ringing endorsement for a chip that's supposed to carry many major Android vendors through to early 2017. And what happens if the Exynos 8890 cleans the 820's clock, or when Apple's A10 inevitably does?

 

[shabby]

Qualcomm wants to show that it has better hardware in the pipeline and line up customers that might have been on the fence.

What other choice does moto/htc/lg have, even if its slower than the exynos they'll still pick the sd820. Don't think the mediatek/huawei chip will be making any dents in the american market.

 

[dawheat]

The CPU performance is a bit underwhelming though the GPU is quite impressive. Efficiency is the biggest factor that there's no data on yet.

It's a good enough chip that you can say more or less is competitive to the A9 (slower CPU, faster GPU) with Apple having a 6 month lead on everyone else right now. Honestly that gap doesn't seem that big a deal - same if it was vice versa. It's hardly an AMD vs. Intel type of situation.

I do think the Exynos 8890 is going to edge the S820 and be a more head to head competitor to the A9. Sure 6 months later the A10 will jump ahead again, but then invariably the next gen QC/Samsung stuff will catch up 6 months later.

 

[Commodus]

What other choice does moto/htc/lg have, even if its slower than the exynos they'll still pick the sd820. Don't think the mediatek/huawei chip will be making any dents in the american market.

That's the rough part. The OEMs that don't make their own processors are boats at sea without sails or paddles. They're at the mercy of wherever Qualcomm and Mediatek go, and if their preferred chipmaker messes up, they're stuck. Apple, Huawei and Samsung have the luxury of charting their own courses.

 

[StrangerGuy]

I dunno, but comparing synthetic benchmarks is a purely academic exercise considering how even $150 Android phones now have SoCs that are already massively overkill for the general user and tech specs isn't even the appeal for most who buy Apple.

 

[ChronoReverse]

The Exynos 7420 GPU seems to handle the S810 GPU pretty handily in off-screen results. Regardless, I do hope this points to Samsung going with lower clocks most of the time to maximize efficiency. Sure burst performance is important, but would be nice to see an Android SOC that can maintain good extended performance without a huge power or heat penalty.

The SD810 beats up the Note5 in graphics despite the CPU handicap. Check out the Nexus 6P review.

 

[stlc8tr]

How reliable is ExtremeTech with their rumor reporting?

http://www.extremetech.com/mobile/2...eal-to-put-snapdragon-820-in-samsung-hardware

Last year, Samsung’s Galaxy S6 and S6 Edge won accolades and set performance records, thanks to their use of cutting-edge 14nm process technology and Samsung’s own Exynos 7420 SoC. This was a significant blow to Qualcomm’s business, since that company has often provided both the SoC and modem for many of Samsung’s flagship designs.

With Samsung using its own silicon and modem for the Galaxy S6, it was an open question whether Qualcomm could win back the Korean company’s flagship business. Now, rumors are indicating it did, thanks to an exclusivity offer that would put Qualcomm’s Snapdragon 820 in Samsung’s upcoming Galaxy hardware (likely the Galaxy S7) until at least April of 2016.

 

[ranjith]

what about the width of the data bus for these,heard that though 7420 is using 64 bit instructions,the data bus is only 32?,so processer gets waitingfor the instructions

 

[lopri]

^ Instruction sets do not dictate data paths, and vice versa. You can have a 32-bit instruction set and a 128-bit data width, and a 64-bit instruction set with a 32-bit data path, etc. Most modern high-end smartphones utilize dual-channel 32-bit memory.

Then there is an interconnect that transfers data from various parts of SOC (memory, GPU, ISP, I/O) to the logic units (A57, A53), as well as between the logic units themselves. (e.g. from A53 cluster to A57 clusters) In big.LITTLE configuration that job is handled by CCI (Cache Coherent Interconnect), and an article below explains it in depth.

http://www.anandtech.com/show/9330/exynos-7420-deep-dive/2

I have not fully understood the explanation, but my guess is that the CCI-400 (current incarnation of CCI that is used in A57+A53 big.LITTLE) is a significant bottleneck when it comes to memory throughput. Presumably Qualcomm's custom implementation is different and it does not suffer from this bottleneck, if the S820's Geekbench scores are to be believed.

Speaking of Geekbench, I see a whole bunch of runs have been uploaded since I last checked the S820's database. It seems like Qualcomm found and settled at a frequency target. Only remaining question is its power characteristics.

https://browser.primatelabs.com/geekbench3/search?dir=desc&q=msm8996&sort=multicore_score

P.S. I have read somewhere the following is Samsung's Exynos 8890, but the numbers look just like those of typical A57+A53, so I am not sure what to make out of it.

Samsung Project Lucky

 

[ranjith]

^ so that could be the reason behind the superior memory benchmarks of snapdragon 820.

 

[CLARiiON]

New 'leaks' from China saying Exynos 8870 also in development. Lower clocked version, probably around 2000 single core geekbench numbers.

 

[lopri]

Helio X20 with its odd configuration (i.e. big.Medium.LITTLE) seems to be tailor-made for benchmarks.