Snapdragon 835 press event previews potent performance


— 12:43 PM on March 22, 2017

Qualcomm's new Snapdragon 835 SoC is the company's next major revision of its flagship chip. The 835 is also the first high-performance SoC built on Samsung's 10nm process. Qualcomm recently invited a sprinkling of the tech press to its San Diego headquarters to take a look at a reference device built around its new design. The folks at AnandTech and PC Perspective both got some quality time with the new hardware.

The design of the Snapdragon 835 is a considerable departure from the 820 before it. While the Snapdragon 820 had four custom-designed Kryo CPU cores, the 835 has eight CPU cores arranged as a pair of four-core clusters—one for high performance and another with an eye towards power efficiency. According to AnandTech, the CPU in the Snapdragon 835 is the first design created under ARM's "Built on ARM Cortex Technology" license. The license lets vendors customize an existing ARM design rather than creating one from scratch. Qualcomm is tight-lipped about exactly which ARM cores it customized or what changes it made, but AnandTech's testing seems to indicate that the high-performance cores are based on the Cortex-A73.

Despite the fact that Qualcomm's latest CPU cores are only semi-custom affairs, the company is using its Kryo moniker for in-house designs. The Kryo 280 CPU in the Snapdragon 835 runs its high-performance cores at up to 2.45 GHz and its high-efficiency cores at up to 1.9 GHz, though only one of the clusters can be active at any given time. Qualcomm says that applications should spend as much as 80% of CPU time on the SoC's high-efficiency cores, making the 835 the company's most power-efficient design to date. However, the press didn't have enough time with the reference platform to quantify battery life or power efficiency.

There's plenty of hard performance numbers to go over, though. Here's the short version: the Snapdragon 835 is the fastest commodity SoC out there today. In PC Perspective's testing, Qualcomm's reference phone posted the highest single-threaded Geekbench score to date (outside of an Apple device), along with the highest multi-threaded score, period. In both metrics, the 835 is a fair bit ahead of the Kirin 960 processor in Huawei's Mate 9, so it seems that whatever customizations Qualcomm made have paid off. When we throw Apple's chips into the mix, even the relatively-aged A9 steps on the Android chips in single-threaded and browser tests, though that won't be a surprise to anyone familiar with this market.

Of course, a SoC is more than just a few CPU cores. Qualcomm says the Snapdragon 835's Adreno 540 GPU is 25% faster than the Adreno 530 in the SD821. That's a huge performance uplift given that the company admits the GPUs are fundamentally similar. Qualcomm says its engineers targeted specific bottlenecks in the Adreno 530's design to improve performance and efficiency. AnandTech writes that Qualcomm also raised the peak GPU clock speed to 710Mhz, a 14% boost over the SD820's 624 MHz. Whatever the source of the speedup, the Adreno 540 appears to be around 30% faster than the 530, by PC Perspective's reckoning. Given that the Snapdragon 821's GPU was already faster than everyone else's (including Apple's A10 Fusion), that means that the Snapdragon 835 is currently the ARM SoC with the fastest GPU. That's great news for folks who are into mobile gaming and VR.

If you believe the rumors, Samsung's Galaxy S8 phone is due for release in about a month and should be the first handset to use Qualcomm's new SoC. Looking at the performance data gleaned from the preview event, the prospects for the S8's performance are enticing. Still, it's important to keep in mind that we know nothing about the new SoC's efficiency, and also that the final implementations of the Snapdragon 835 chip may deviate from the reference platform in performance. Head on over to AnandTech or PC Perspective if you want to see all the pretty charts and numbers.

Tip: You can use the A/Z keys to walk threads.
View options

This discussion is now closed.