The new Qualcomm Snapdragon 888 processor just got its first benchmarks, and they’re a mixed bag. The chip scores consistently better than last year’s Snapdragon 865 series on a range of tests, but it’s unclear how it performs vis-à-vis Apple’s latest A14 chip.
Qualcomm usually lets journalists run their own benchmarks on its Snapdragon reference phones at its annual event in Hawaii, but this is 2020, so the company just sent out a spreadsheet and video of its device running the benchmarks.
The Snapdragon 888 has a bunch of innovative features. On the CPU side, it’s the first Snapdragon to feature a three-tier set of differently structured cores, capped by a new ARM Cortex-X1 “prime core.” Qualcomm also said it has a 25% faster GPU and a much faster AI processing unit.
Some of its key features can’t be tested by benchmarks, though. The 888 integrates the X60 modem into the chipset for lower heat and better battery life, and the X60 modem itself is capable of combining 5G bands in ways that could significantly improve performance on US and Canadian networks.
The chip gets 16-17% better Geekbench CPU scores than the Snapdragon 865+ in the Samsung Galaxy Note 20 Ultra, and 33-57% better results on two cherry-picked GFXBench tests, and 25% better scores on Antutu (which has CPU and GPU components).
But the score-settling geek crowd will mostly be interested in how it competes against Apple’s latest A14 chip. And there, it’s hard to draw any conclusions, I suspect because of how the benchmarks are programmed differently for Apple’s and Google’s OSes.
In the rest of the world, Snapdragon has several competitors. Samsung’s Exynos and Huawei’s Kirin both power flagship Android smartphones outside the US, but they don’t power any flagships sold in the US. MediaTek’s chipsets give Qualcomm’s midrange and high-midrange chips a run for their money. The MediaTek Dimensity 1000 in T-Mobile’s LG Velvet is superior to the Snapdragon 765 in AT&T’s version of the phone. But so far MediaTek hasn’t notched true flagship wins with Samsung, OnePlus, Motorola, LG, or other top-of-the-line US phone makers.
That leaves the top of the US market a battle between Qualcomm and Apple.
For at least a year now, Apple chips have run considerably ahead of Qualcomm chips on Geekbench, while the reverse has been true on Antutu. This difference goes down to the sub-scores; isolate the CPU score alone, and the A14 (and even the A13) do better on Geekbench than the 888, while the 888 crushes the Apple chips on Antutu.
Similarly, the chips split the difference on the two GFXBench tests that Qualcomm released. The Snapdragon 888 came out the same as our A13 device, and behind the A14, on the Aztec Ruins test, while it came out ahead of both devices on the Manhattan test.
To me, this just ends up coming down to how each benchmark is programmed and what APIs it uses. Unfortunately, I just don’t know enough about the nuts and bolts of how the benchmarks are programmed to be able to say more; I suspect the folks at AnandTech will.
There’s one interesting factor: the one Geekbench sub-score where the Snapdragon 888 did better than the A14 was on the multi-core Crypto benchmark, which may use the same kind of math you find in an AI processor. The Snapdragon 888’s AIMark scores ran way ahead of the A14 or any other competing chip.
I Don’t Like Benchmarks Anymore
In my mind, mobile benchmarks are a dying field, and have been increasingly useless for the past several years. We still do them, because they tell us a little bit, and they aren’t a heavy lift. But they’re inherently misleading because none of them try to simulate actual applications anymore. One of the reasons we stick with the aging, non-updated PCMark on Android phones is because it’s one of the few benchmarks I’ve found that actually tries to simulate app use.
Qualcomm points out, correctly, that mobile chipsets have complex workloads and components that don’t mirror the old PC world, where when your CPU ran at a 25% better clock speed, that meant your spreadsheets calculated 25% faster. DSPs, image signal processors, touch controllers, sensors, and especially manufacturer Android software alterations all figure in to mobile performance in ways they don’t on PCs. And Geekbench tests none of these things.
The result is that we’re stuck watching YouTubers with stopwatches to try to measure the responsiveness differences between Android and iOS phones, and I’m stuck wandering the streets of New York City running speed tests to figure out the difference between radios. This keeps me and the YouTubers in business, to be sure, but it’s less scientific and less easily repeatable than I think we all would like.
The first Snapdragon 888 phones are likely to be the Samsung Galaxy S21 line, which we anticipate will be announced in January. Only then will we see what real-world Snapdragon 888 performance will be like.
