The Samsung Galaxy S20+, S20 Ultra Exynos & Snapdragon Review: Megalomania Devices
by Andrei Frumusanu on April 3, 2020 9:30 AM ESTMachine Learning Inference Performance
The new SoC generations also bring with them new AI capabilities, however things are quite different in terms of their capabilities. We saw the Snapdragon 865 add to the table a whole lot of new Tensor core performance which should accelerate ML workloads, but the software still plays a big role in being able to extract that capability out of the hardware.
Samsung’s Exynos 990 is quite odd here in this regard, the company quoted the SoC’s NPU and DSP being able to deliver a 10TOPs but it’s not clear how this figure is broken down. SLSI has also been able to take advantage of the new Mali-G77 GPU and its ML abilities, exposing them through NNAPI.
We’re skipping AIMark for today’s test as the benchmark couldn’t support hardware acceleration for either device, lacking updated support for neither Qualcomm’s or SLSI’s ML SDK’s. We thus fall back to AIBenchmark 3, which uses NNAPI acceleration.
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 initializing 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.
In the purely CPU accelerated workloads, we’re seeing both phones performing very well, but the Snapdragon 865’s A77 cores here are evidently in the lead by a good margin. It’s to be noted that the scores are also updated for the S10 phones – I noted a big performance boost with the Android 10 updates and the newer NNAPI versions of the test.
AIBenchmark 3 - NNAPI INT8
Integer ML workloads on both phones is good, but because the Snapdragon 865 leverages the Hexagon DSP cores for such workload types, it’s much in lead ahead of the Exynos 990 S20. This latter variant however also showcases some very big performance improvements compared to its predecessor. I still think that Samsung here is only exposing the GPU of the SoC for NNAPI, but because of the new microarchitecture being able to accelerate ML workloads, we’re seeing a big performance improvement compared to the Exynos 9820.
AIBenchmark 3 - NNAPI FP16
In FP16 workloads, the Exynos 990’s GPU actually manages to more often outperform the Snapdragon 865’s Adreno unit. In workloads that allow it, HiSilicon’s NPU still is far in the lead in workloads as it support FP16 acceleration which isn’t present on either the Snapdragon or Exynos SoCs – both falling back to their GPUs.
AIBenchmark 3 - NNAPI FP32
Finally, FP32 also again uses the GPU of each SoC, and again the Exynos 990 presents quite a large performance lead ahead of the Snapdragon 865 unit.
It’s certainly encouraging to see the Samsung SoC keep up with the Snapdragon variant of the S20, pointing out that other vendors now finally are paying better attention to their ML capabilities. We don’t know much at all about the DSP or the NPU of the Exynos 990 as Samsung’s EDEN AI SDK is still not public – I hope that they finally open up more and allow third-party developers to take advantage of the available hardware.
Facebook
Twitter
RSS
137 Comments
View All Comments
iSeptimus - Sunday, April 12, 2020 - link
They really don't. 2 years max on updates and the security patches only come out when it is critical after that.My Galaxy Fold only just got One UI 2.1 and Android 10. The most expensive phone they do and they are already slowing down updates for it. Samsung suck at software.
MAGAover9000 - Tuesday, April 7, 2020 - link
Sounds like a positive TBH. I hate some of the updates.Omega215D - Friday, April 3, 2020 - link
That's for the maxed out S20 Ultra, what S10 are you comparing it to? There are the following additions to the S20 lineup: camera tech, 5G, battery capacities, mobile chipsets, display tech, etc. Now not everyone may find those attractive enough to jump on board that doesn't mean they have to buy the damn thing.Threska - Friday, April 3, 2020 - link
5G is probably the biggest push for cost, since 5G phones usually run in the $1,000 range.Andrei Frumusanu - Friday, April 3, 2020 - link
The Galaxy S20 and S20+ 4G variants have the same launch prices as S10 and S10+ - with the only caveat being that they're only available in certain markets, with the US missing from that list.So in that sense, the only thing you're paying extra for is the 5G.
FunBunny2 - Friday, April 3, 2020 - link
"5G phones usually run in the $1,000 range"I still don't get why anyone would pay for "5G". mmWave is barely extant. Verizon, for one, adverts pump up use in football stadiums (if we ever have such), 'so it must be great for you' kind of thing. while sub6 is barely perceptibly different LTE. Why? If there's ever a true build out of mmWave, it'll be in suburban subdivisions with plenty of street lights, no longer the norm. In cities, you'll need some sort of '5G' wifi inside any kind of building with a (massive?) dish on the roof. Ah cannaw change da laws of physic, capn.
shabby - Friday, April 3, 2020 - link
We don't have a choice, Qualcomm is forcing it onto us.s.yu - Friday, April 3, 2020 - link
Qualcomm's going with the flow of the industry, there's no way but forward ;)But anyway, the Luneburg lenses China Mobile showcased, if they're compatible with mmWave, and if they're even willing to sell them(the issue is cost and somehow they're able to make them much cheaper, cheap enough for civilian use), then they could drastically increase coverage at essentially the same power levels.
PeachNCream - Friday, April 3, 2020 - link
Yup, that's a lot to pay for something that you carry around on a daily basis and expose to the outdoor world including weather, impact, and theft. I know the TELCOs are hiding the cost inside monthly payments over the life of a contract so most buyers will not take note of the price over say an $800 dollar handset, but wow people are forking over a lot of their earnings just to carry around a status symbol that other people will stop caring about after a couple of months.FunBunny2 - Friday, April 3, 2020 - link
"wow people are forking over a lot of their earnings just to carry around a status symbol"why does a Ferrari cost 3 times a Ford (not the GT, of course)? because the capital investment to make either is nearly the same, but spread out over a fraction of units. same with "5G" phones. the capital to make all those specialized bits and pieces, whether Samsung made or bought it, is about the same, but separate, from the capital to make a moto G6, etc. but far fewer units. unlike the march to LTE, "5G" is more likely to be a toddler's waddle. with sky high prices as an added benefit.
there was a headline recently that the Swiss watch industry is on a ventilator, few have any need for such bling. can the bling smartphone be far behind? at least your scion can use that Patek Philippe in 50 years.