Machine Learning Inference Performance

AIMark 3

AIMark makes use of various vendor SDKs to implement the benchmarks. This means that the end-results really aren’t a proper apples-to-apples comparison, however it represents an approach that actually will be used by some vendors in their in-house applications or even some rare third-party app.

鲁大师 / Master Lu - AIMark 3 - InceptionV3 鲁大师 / Master Lu - AIMark 3 - ResNet34 鲁大师 / Master Lu - AIMark 3 - MobileNet-SSD 鲁大师 / Master Lu - AIMark 3 - DeepLabV3

Unfortunately for the ROG Phone II, it’s another device that’s lacking the proper Qualcomm proprietary libraries that AI Mark makes use of, ending up not being in the charts anywhere as the application couldn’t start.

AIBenchmark 3

AIBenchmark takes a different approach to benchmarking. Here the test uses the hardware agnostic NNAPI in order to accelerate inferencing, meaning it doesn’t use any proprietary aspects of a given hardware except for the drivers that actually enable the abstraction between software and hardware. This approach is more apples-to-apples, but also means that we can’t do cross-platform comparisons, like testing iPhones.

We’re publishing one-shot inference times. The difference here to sustained performance inference times is that these figures have more timing overhead on the part of the software stack from initialising the test to actually executing the computation.

AIBenchmark 3 - NNAPI CPU

We’re segregating the AIBenchmark scores by execution block, starting off with the regular CPU workloads that simply use TensorFlow libraries and do not attempt to run on specialized hardware blocks.

AIBenchmark 3 - 1 - The Life - CPU/FP AIBenchmark 3 - 2 - Zoo - CPU/FP AIBenchmark 3 - 3 - Pioneers - CPU/INT AIBenchmark 3 - 4 - Let's Play - CPU/FP AIBenchmark 3 - 7 - Ms. Universe - CPU/FP AIBenchmark 3 - 7 - Ms. Universe - CPU/INT AIBenchmark 3 - 8 - Blur iT! - CPU/FP

We performed the AI Benchmark tests in the default mode of the phone, and here the ROG Phone II lands relatively in average to better than average amongst S855 devices.

AIBenchmark 3 - NNAPI INT8

AIBenchmark 3 - 1 - The Life - INT8 AIBenchmark 3 - 2 - Zoo - Int8 AIBenchmark 3 - 3 - Pioneers - INT8 AIBenchmark 3 - 5 - Masterpiece - INT8 AIBenchmark 3 - 6 - Cartoons - INT8

AIBenchmark 3 - NNAPI FP16

AIBenchmark 3 - 1 - The Life - FP16 AIBenchmark 3 - 2 - Zoo - FP16 AIBenchmark 3 - 3 - Pioneers - FP16 AIBenchmark 3 - 5 - Masterpiece - FP16 AIBenchmark 3 - 6 - Cartoons - FP16 AIBenchmark 3 - 9 - Berlin Driving - FP16 AIBenchmark 3 - 10 - WESPE-dn - FP16

AIBenchmark 3 - NNAPI FP32

AIBenchmark 3 - 10 - WESPE-dn - FP32

The same conclusion can be had of the INT8, FP16 and FP32 scores. It’s odd that the FP32 scores aren’t better than that of the competition as the workload is accelerated by the GPU, and we had expected the S855+’s higher GPU frequency to have a larger impact on this score.

We also have to note that if it weren’t for a very recent firmware update from ASUS we wouldn’t be able to publish any AI benchmarks at all, as it previously lacked the proper NNAPI drivers. As such, it seems that this aspect was pretty much just an afterthought for the phone rather than something that was on a critical to-do list.

System Performance GPU Performance - Top Performance
Comments Locked

75 Comments

View All Comments

  • Doc Rob - Monday, September 30, 2019 - link

    the only reason I do not buy one of the ROG phones is simply I NEED wifi calling on tmobile.. I do no want to try and use other apps etc.. allow the use of voLTE and WIFI calling and it would open the market up for many more consumers.
  • Kishoreshack - Tuesday, October 1, 2019 - link

    Volte is enabled & WiFi Calling too
    you might have got wrong memo
  • TheinsanegamerN - Thursday, December 26, 2019 - link

    Not in the USA it isnt
  • PeachNCream - Tuesday, October 1, 2019 - link

    Alternatively, if you use Skype you can pay a fairly minimal amount of money to get a phone number and that will work both over WiFi and cellular data as a VoIP phone. It may fill your needs and mitigate the need for native WiFi calling support on the device.
  • Kishoreshack - Monday, September 30, 2019 - link

    @iancutress Where is the Display Analysis?
    Very sad you skipped on it
  • s.yu - Tuesday, October 1, 2019 - link

    There's also a ton of mistakes in the spec sheet.
    It's 128/8GB or 1TB/12GB, there is no 256GB variant but there seems to be a 512GB one.
    It's UFS3.0 not 2.1.
    There are two C ports, which is highly unique but not noted.
    Possibly other mistakes.
  • Death666Angel - Tuesday, October 1, 2019 - link

    Probably some copy-paste issues in the table. And the second USB C port is mentioned on the first page below the picture of the connector.
  • s.yu - Wednesday, October 2, 2019 - link

    Yeah it was most certainly noted elsewhere, but the spec sheet should be comprehensive.
  • Kishoreshack - Monday, September 30, 2019 - link

    Why there is no display analysis
    Never seen Anandtech such an important aspect
  • Kishoreshack - Monday, September 30, 2019 - link

    Ian Cutress you should have let Andrei do the display analysis or the review itself
    half review is never expected from Anandtech

Log in

Don't have an account? Sign up now