APFS issue? Huge disk speed drop in 10.3 vs 10.2.1 (Antutu - iPhone SE)

[dikrek]

Hello all,

See attached a picture of the I/O performance in 10.3 vs 10.2.1

So in 10.2.1 the SE I/O speed was getting 13348, and in 10.3 it's 5875??

That's a huge drop in I/O performance.

Can anyone else run Antutu and confirm?

Thx

D

 

[keysofanxiety]

System might still be be indexing (or whatever the iOS equivalent is). Try a hard reset and re-run?

Also was that previous score benched on your phone or was it a different device?

 

[NT1440]

Perhaps that benchmark app needs to be updated due to conflicts with APFS and the way it reads to the app itself?

 

[dikrek]

System might still be be indexing (or whatever the iOS equivalent is). Try a hard reset and re-run?

Also was that previous score benched on your phone or was it a different device?

I tested it at various times, before and after reset, before and after reindexing.

The previous score was the reference benchmark in the app but I had tested it and I got a similar number (just didn't save the result since it was almost identical).
[doublepost=1490801319][/doublepost]

Perhaps that benchmark app needs to be updated due to conflicts with APFS and the way it reads to the app itself?

Possibly - then there's the slower 3D scores to think about too (look at the bottom of the pic)

 

[keysofanxiety]

I tested it at various times, before and after reset, before and after reindexing.

The previous score was the reference benchmark in the app but I had tested it and I got a similar number (just didn't save the result since it was almost identical).
[doublepost=1490801319][/doublepost]

Possibly - then there's the slower 3D scores to think about too (look at the bottom of the pic)

Is the reference benchmark of an iPhone the same storage size? Flash chips work in parallel so the bigger the storage, the faster the speeds.

Granted, you ran the app before and it gave similar results. But I'm just thinking maybe the app needs updating, or possibly the new APFS is skewing the scores?

With regards to real-world usage, have you noticed any slowdowns or does the system run a little quicker?

Would appreciate any insight you can provide -- you've got my curiosity.

 

[NT1440]

Possibly - then there's the slower 3D scores to think about too (look at the bottom of the pic)

To be honest I think it's a bit foolish to expect accurate results from a benchmarking app after an iOS .X update, let alone an entirely new filesystem before the app itself receives any updates.

As the poster above asked, have you seen any performance degradation or are we looking at numbers for numbers sake? If it's the latter, wait for the app to be updated before taking the results seriously.

 

[dikrek]

To be honest I think it's a bit foolish to expect accurate results from a benchmarking app after an iOS .X update, let alone an entirely new filesystem before the app itself receives any updates.

As the poster above asked, have you seen any performance degradation or are we looking at numbers for numbers sake? If it's the latter, wait for the app to be updated before taking the results seriously.

Real world usage seems fluid - no issues I can see, but then again the SE has an excess of speed so maybe I'm just not hitting the bottleneck.

So it could be that

1. There IS a degradation, and I simply don't feel the degradation since I don't do something that stresses the IO apart from the benchmark (I have the 64GB SE, even at the reduced speed it's faster in IO than many other models) OR
2. The benchmark needs to be updated as many of you said

Now, I do storage for a living, and if the Antutu benchmark follows the API rules, it should NOT have to be rewritten. If anything, the results show that doing the exact same system calls, in a 64-bit app, that's current as of 10.2.1, the new OS is half the speed for I/O.

D

 

[NT1440]

Real world usage seems fluid - no issues I can see, but then again the SE has an excess of speed so maybe I'm just not hitting the bottleneck.

So it could be that

1. There IS a degradation, and I simply don't feel the degradation since I don't do something that stresses the IO apart from the benchmark (I have the 64GB SE, even at the reduced speed it's faster in IO than many other models) OR
2. The benchmark needs to be updated as many of you said

Now, I do storage for a living, and if the Antutu benchmark follows the API rules, it should NOT have to be rewritten. If anything, the results show that doing the exact same system calls, in a 64-bit app, that's current as of 10.2.1, the new OS is half the speed for I/O.

D

I sincerely doubt that APFS has caused I/O speed to actually drop, in fact from the WWDC demos (admittedly on a mac, but I can't imagine such a drastic deviation to iOS) performs many function at a breathtakingly faster speed.

Perhaps there is a bug in 10.3 itself (not related to APFS) that is causing the issue? I guess we'll have to see but this one is kind of hard to figure out without being able to peek under the hood you know?

 

[dikrek]

I sincerely doubt that APFS has caused I/O speed to actually drop, in fact from the WWDC demos (admittedly on a mac, but I can't imagine such a drastic deviation to iOS) performs many function at a breathtakingly faster speed.

Perhaps there is a bug in 10.3 itself (not related to APFS) that is causing the issue? I guess we'll have to see but this one is kind of hard to figure out without being able to peek under the hood you know?

It could be due to many things - even advanced filesystems often are slower in specific things than old-fashioned ones, but the overall benefit is greater.

For example, Antutu might be testing a specific IO pattern that's not friendly to APFS. Totally possible and I've seen this happen many times in my career in storage.

 

[Kalloud]

There definitely is a significant ux I/O performance and 3D performance drop on my iPhone 6s after the iOS 10.3 update .
What about you guys using other iPhone models because the 6s and the SE have basically the same soc?

 

[Ladybug]

I haven't updated my 6s to 10.3 yet. I think I will hold off for awhile longer and see how this shakes out. Please keep us updated. Thanks

 

[miragebg]

I still keep my 6S on 10.2 and will wait to see how this goes...
Do you think that 10.2 is still faster than 10.3, despite synthetic benchmarks..?

 

[dikrek]

I still keep my 6S on 10.2 and will wait to see how this goes...
Do you think that 10.2 is still faster than 10.3, despite synthetic benchmarks..?

10.3 feels fast and in synthetic benchmarks the CPU stuff scores faster. Even Geekbench improved a bit on my SE.

I was just wondering about the IO speeds reported by AnTuTu mostly.

 

[cswifx]

Regardless of I/O benchmarks, my SE feels a lot faster on 10.3. It might be the Antutu app.

 

[dikrek]

Regardless of I/O benchmarks, my SE feels a lot faster on 10.3. It might be the Antutu app.

I ran passmark and the disk score was 43123, the reference was 52552, so that's a relatively small difference.

Reference:

http://www.iphonebenchmark.net/diskmark_chart.html

They benchmark itself is from 2012 so not exactly new code. The disk part of it gets done in a second.
[doublepost=1490872947][/doublepost]Also ran 3Dmark slingshot extreme and got 2011, the SE is supposed to get 1952, so it was a bit faster than normal.

 

[Everbarve]

My experience with Antutu Benchmark under iOS 10.3 is that the score deviation from 10.2.1 is minimal with old hardware (iPhone 6, iPad Air 2) and around 6% lower with the new hardware (iPhone 6s, iPad Pro), e.g. my iPad Pro 9.7 dropped from approx. 174000 under 10.2.1 to 163000 under iOS 10.3; iPhone 6s dropped from 143000 to 134000

 

[iScholtz]

Hello all,

See attached a picture of the I/O performance in 10.3 vs 10.2.1

So in 10.2.1 the SE I/O speed was getting 13348, and in 10.3 it's 5875??

That's a huge drop in I/O performance.

Can anyone else run Antutu and confirm?

Thx

D

same here, also on a SE. And with beta 5 (at least, may have been since a beta before 5) the 3D Garden test results are also down.
[doublepost=1493595890][/doublepost]

Hello all,

See attached a picture of the I/O performance in 10.3 vs 10.2.1

So in 10.2.1 the SE I/O speed was getting 13348, and in 10.3 it's 5875??

That's a huge drop in I/O performance.

Can anyone else run Antutu and confirm?

Thx

D