Today, Samsung announced the launch of its newest mobile processor, the Exynos 980. This system-on-chip, which should enter mass production later this year, includes an integrated 5G modem, NPU (neural processing unit), and Mali G76 GPU.
How much should we care about 5G?
5G itself is probably not going to be a very big deal for most people. Aside from how relatively few and scattered 5G networks are for now, the really big speed increases for 5G happen on 5G FR2, the millimeter-wave band. The problem is, RF at millimeter wavelength has near-zero penetration—signals are easily blocked by walls, panes of glass, or even human bodies.
This makes 5G FR2 a potentially great delivery mechanism for site-to-site networking—such as an external antenna feeding the wired or Wi-Fi network for a large building—but much less so for an individual phone, whose connection would be interrupted by entering a building, a car, or even the user turning around and putting their body in between the phone’s antenna and the tower.
5G FR1, the sub-6GHz band, should be much more usable. It’s difficult to tell how much benefit real-world users will get from 5G FR1 connections. Current real-world speed tests show 5G FR1 speeds of 400Mbps to 600Mbps. This looks great compared to real-world speed tests of 4G at typically 50Mbps—but there’s a reason the 4G tests are so much slower, and that reason is congestion. 4G LTE is already capable of 1Gbps connections to stationary users, but there’s not enough bandwidth to support anything like that much speed delivered to all of the connected consumers.
Once 5G becomes a typical consumer use case rather than an uncommon and special one, we do expect to see some improvement over current 4G LTE—but don’t expect 500Mbps to become the new normal.
GPU and NPU
The Exynos 980’s GPU, Mali G76, is a significant improvement over earlier versions—roughly twice as fast as the Mali G71 used in earlier Samsung devices such as the Galaxy S8. It’s still not quite as fast in most benchmarks as the Qualcomm Adreno 630 used in Google’s Pixel 3 or Samsung’s own Galaxy S9+, though, so while consumers should expect snappy game and app performance, the GPU isn’t really a market game changer.
The real-world impact of the NPU, or Neural Processing Unit, is still a bit of a question mark. Samsung claims that the Exynos 980’s NPU is nearly three times as fast as the NPU used in earlier Exynos systems. We know that Samsung uses the NPU to enhance performance in its camera and augmented reality applications, but that’s likely it—for now. We expect that utilization of on-board NPUs in smartphones and tablets will increase greatly in the near future, as software vendors catch on to the possibilities of real-time, on-device machine learning.
A tale of shrinking process size
This won’t be Samsung’s first foray into 5G mobile devices; the difference here is not 5G connectivity itself but the fact that the modem is built in to the SoC (System on Chip) rather than being an external device. Building the 5G modem directly on-die with the CPU means less physical space needed for the components inside the phone, as well as lower power consumption and heat generation. All of this is really made possible by the shrink to an 8nm process—by comparison, the Snapdragon 845 used by Google’s flagship Pixel 3 phones is built on a 10nm process.
This is a story we’re seeing played out on all fronts this year. AMD shrank its CPU die process to 7nm earlier this year, allowing it to outperform Intel on both desktop and server CPUs. Intel itself gave us an even better example when it released both Ice Lake and Comet Lake notebook CPUs. Ice Lake, which is built on a 10nm process, was able to devote on-die space to a GPU three times as fast as the one in Comet Lake, which is still on a 14nm process.