Geekbench 4 released for Android and iOS

Geekbench 4 released for Android and iOS
Geekbench, one of the most popular benchmark applications for mobile devices, has just received a major update that brings a new set of tests and some visual improvements.

The source of many leaks, Geebench's online database of test results is quite impressive. With the latest update, the developer is trying to align the benchmarking tool with the latest requirements and technological advancements in the mobile industry.

The new Geekbench 4 includes updated CPU benchmarks, but also new GPU Compute tests for the latest graphics processing units launched on the market. In terms of visual improvements, the folks behind Geekbench announced the latest version comes with a Material Design interface.

If you've never used the benchmark application, then you should know Geekbench's tests have been especially customized to accurately measure mobile CPU performance. The test results can be compared between devices since Geekbench has a huge database.

The developers also confirmed Geekbench 4's tests are “multi-core aware,” so users will know precisely the full potential of their phone's processor.

Keep in mind that Geekbench 4 is now available for both the Android and iOS platforms, so you can either update it if you have a previous version, or download it via Google Play or App Store.

For those owning Android devices, Geekbench 4 should be compatible with phones and tablets powered by Android 5.0 and up, while Apple fans will need devices running iOS 9.0 or later.



1. Nopers unregistered

Wait for the complaints that it's biased towards iOS because the scores didn't change much whereas android scores all went down, by a large margin on the Exynos 8890 and Snapdragon 820.

15. Finalflash

Posts: 4063; Member since: Jul 23, 2013

Yea, the iGasms will be real at least for the first week or so. The SD810 and SD820 seem to be scoring lower than the SD801 and SD805 chips in Geekbench 4. This usually means there is something wrong with the benchmark rather than the processors. Here are the benchmarks from their own website: (single/multi averages) Z5 Compact (8 cores SD 810, 4 big 4 small) 1300/2200 Nexus 6P (8 core SD 810, 4 big 4 small) 1300/3200 (the multi scores range from 2200-4000 for the 6P) HTC 10 (4 cores SD 820, 2 big, 2 small) 1300/2500 (once again, variance on multicore is between 2300-3900) HTC M8 (8 cores, SD 805, 4 big, 4 small) 1100/2800 (variance on multi is about 2400-3000) LG Nexus 5 (4 cores SD800) 1100/2500 OnePlus One (4 cores, SD 801) 1050/2700 So yea, just from those alone, this is several degress of messed up. It is likely a problem with the benchmark rather than the chips and so they will likely fix it soon enough, with real benchmarks coming shortly after. If not, then likely the benchmark is worthless because apparently the Nexus 5 beats the Nexus 6P and other devices from last year in several instances consistently.

17. sarcastic_nerd unregistered

It just means that the HTC 10 is throttling hardcore. I ran the test 3 times on my OP3 and got 4050, 4060 and 4043 in multi core and 1710, 1723 and 1720 in single core. The scores and the benchmark are fine.

25. Finalflash

Posts: 4063; Member since: Jul 23, 2013

There isn't any throttling going on though and neither does it heat up. You can track the numbers and the processor frequency doesn't stutter at all. The numbers are wildly different for no real discernible reason across the same devices and different devices. Also, it seems as though the benchmark doesn't use all the cores in the SD chips. It only seems to be using the big clusters and not the little clusters it seems.

26. sarcastic_nerd unregistered

That's just the frequency CPU is rated at. 820 will report 1.59 GHz even before you running the test. And there's no way to get 1700 in single and 4000 on multi from. 2 cores.

7. sarcastic_nerd unregistered

The 8890 still scores much higher in multicore.

22. jove39

Posts: 2148; Member since: Oct 18, 2011

Incorrect. There hasn't been any credible concern for difference between iOS and Android version. Geekbench essentially compare SOC performance instead of just CPU. Why i say that - because several computations are offloaded to dedicated hardware instead of running them on CPU. Here Linus T has raised concern on why geekbench scores are not worh comparing across architectures -

2. siddharthayadav202

Posts: 286; Member since: Apr 23, 2016

If this happens, why not? I know different tests give different results but I would seriously doubt that Ax Chips are way ahead of Sd or Exynos in anything particular that is imp.

3. AmashAziz

Posts: 2934; Member since: Jun 30, 2014

You surely don't know the potential of the Ax series chip. Just upgrade ur knowledge by studying the tests and results all around the internet. Also compare a 12.9 inch ipad pro to a galaxy note 7 regarding performance, if u have access to both. Then you will know the true power of the A9x...... And are u an indian?

12. siddharthayadav202

Posts: 286; Member since: Apr 23, 2016

A9X is a complete different category. See A9. A9X is much bigger in size and needs more power and therby cannot get in a phone. And I am Indian. A9 though faster in single core, is still slower in multicore perf. Both are likely almost same in size as well.

13. siddharthayadav202

Posts: 286; Member since: Apr 23, 2016

I see reason for confusion. I meant A* like A7, A8 , A9. Sry for confusion.

24. AmashAziz

Posts: 2934; Member since: Jun 30, 2014

It's good that u cleared up the confusion. Otherwise I thought why is this retard talking about the Ax series and then moving onto A series. So now, yes, the A9 doesn't have a significant benefit in performance in comparison to sd 820 and exynos 8890. But its the optimizations of the iphone 6s which give it the iupper hand. By the way my country is ur neighbour.

4. zunaidahmed

Posts: 1185; Member since: Dec 24, 2011

I believe Ax chips are for tablets, which makes them considerably more powerful, not to mention tablets can sustain the performance way longer and the battery life is a less of a matter, we don't have 8890 and 820 tablets yet, so it's really hard to compare, for now we can only compare them to the a series chips

8. Finalflash

Posts: 4063; Member since: Jul 23, 2013

I think you misunderstood him, he mean Ax as in x=integer like 8 or 9.

14. siddharthayadav202

Posts: 286; Member since: Apr 23, 2016

Yeah. Thanks bro.

10. ebilcake

Posts: 1231; Member since: Jul 16, 2016

This new test really doesn't like Snapdragon 820, apart from the GPU Compute score which is quite high. I don't like to see large differences like this between how hardware compares between versions, doesn't inspire you with much confidence. If it wasn't accurate the first time whose to say it is now?

11. sarcastic_nerd unregistered

It was accurate the first time and it is accurate now. Don't know what you are talking about. They removed the memory bandwidth test, which reduced the score of the Snapdragon 820 considerably compared to other SoCs. Now it's a pure CPU bound test and the SD820 does pretty well here. I got about 1720 on single core and 4050 on multi core on my oneplus 3. It gets the same multi-core as the A9, which is pretty great and on par with what it was getting before in CPU tests. The 820 scored considerably higher in memory bandwidth which increased the overall score of the 820 compared to A9 before.

18. ebilcake

Posts: 1231; Member since: Jul 16, 2016

So you have no opinion about this at all? Memory plays quite a large role in overall performance so I'm not sure I would agree with that test being removed, especially considering how much it effects the end score.

20. sarcastic_nerd unregistered

This is a CPU benchmark. Memory should not be a part of this. They can make it as a seperate test. And memory bandwidth plays insignificant role in general performance. DDR3 to DDR4 doubled the bandwidth with less than 5% gain in performance. More RAM is more useful than less faster RAM.

28. vincelongman

Posts: 5747; Member since: Feb 10, 2013

Memory is still tested its just got a lower weight on the final scores The weightings are 5% - Cryptography 45% - Integer 30% - Floating Point 20% - Memory Which IMO are very fair for what's meant to mainly test the CPU of the SoC

16. sarcastic_nerd unregistered

Result highlights: Changes: The memory bandwidth tests were removed from the CPU tests(makes sense) which put the SD 820 and Exynos 8890 with their high frequency LPDDR4 modules on disadvantage compared to Apple's A9. Android highlights: Exynos 8890 is now the highest scoring single and multi core CPU at 1850 single and 5400 multi core in the android ecosystem now. Closely followed by SD 820 in single core and kirin 950/955 in multi core. GPU compute is largely useless and just shows how much compute power you have if you use your GPU for rendering. Since mobile OS can't run maya or blender or no one is using them as a workstation, this is absolutely useless. Radeon and adreno GPUs will usually lead in this due to high FLOP rating. However, it usually doesn't imply higher gaming performance, since there are other metrics like fill rate and tesselation to consider. iOS: Scores largely remain unchanged because the RAM in A series CPUs were nothing to write home about and weren't affecting CPU scores much. Desktop and laptop( intel): Intel CPUs gain roughly 25-40% in single core score due to removal of the memory bandwidth test, since memory scores were much less than CPU tests and were pulling down the CPU results. 6700k is still the single threaded king with 6300 on single core at stock CPU speed. Expect over 7300 with a average overclock of 4.8 GHz. 6850x is still the consumer multi-core king with over 35000 on multi-core. Xeons can score higher though.

27. vincelongman

Posts: 5747; Member since: Feb 10, 2013

Nice summary! BTW another thing I noticed is that they moved the AES tests to its own subsection and gave it a lower weighting on the final score

33. sarcastic_nerd unregistered

Didn't see that. I should have looked at the subtests properly. Earlier they took a geometric mean of the three tests. Now it's a better representation of real world tests. Also desktops and phones run the exact same script now.

31. ebilcake

Posts: 1231; Member since: Jul 16, 2016

Are you aware the last three tests in Geekbench 4 are all memory tests? Copy, Latency and Bandwidth ? What did they remove again?

32. sarcastic_nerd unregistered

Oh yeah, didn't see that. However, now they have lower rating compared to older geekbench. vince already told that though.

23. bucky

Posts: 3794; Member since: Sep 30, 2009

Who cares. If the phone runs the way you want it, then you're good.

29. ibend

Posts: 6747; Member since: Sep 30, 2014

on Geekbench4 SD652 multicore = 4012 SD820 multicore = 3458 something wrong with this new version :-/

30. sarcastic_nerd unregistered

652 gets 2400 and 820 gets 4000.

34. vincelongman

Posts: 5747; Member since: Feb 10, 2013

Nothing is wrong with that SD652 has 4 big A72s and 4 little A53s SD820 has 2 big Kryos and 2 little Kryos

Latest Stories

This copy is for your personal, non-commercial use only. You can order presentation-ready copies for distribution to your colleagues, clients or customers at or use the Reprints & Permissions tool that appears at the bottom of each web page. Visit for samples and additional information.