May 29, 2024


My Anti-Drug Is Computer

Google Tensor G2 benchmarks feature image

Google says it’s ‘very comfortable’ with Tensor not winning benchmarks

Google Tensor G2 benchmarks feature image

Robert Triggs / Android Authority


  • Google is “very comfortable” with Tensor chips not winning benchmarks, a organization government reported.
  • The director said Google benchmarks program workloads that will actually run on Pixel telephones.

Google’s Pixel 6 and Pixel 7 telephones are powered by semi-custom made Tensor processors, but these SoCs really do not seriously take the combat to Apple, Qualcomm, and other chipmakers when it arrives to most benchmarks.

Now, senior director of product management for Google Silicon Monika Gupta has spoken out about the Tensor line’s benchmark efficiency in the official Designed By Google Podcast (h/t: 9to5Google).

Gupta advised that benchmarks are only aspect of the tale when it arrives to smartphone processors:

I assume classical benchmarks served a function at some second in time, but I believe the business has evolved given that then. And if you seem at what Google is attempting to do by pushing AI improvements into a smartphone — mainly because we feel like this is the method that will produce practical experiences like some of the ones I just talked about — classical benchmarks were authored in a time the place AI and telephones did not even exist. They may well notify some story, but we never truly feel like they inform the finish tale.

She additional that Google benchmarks the “actual software workloads” that will run on its chipsets rather than relying on benchmark applications with synthetic workloads. 

“We know specifically what we’re developing for, and if that indicates we’re not going to get on benchmarks or not seem as terrific on benchmarks, we’re flawlessly at ease with that, due to the fact the conclude-outcome speaks for by itself,” Gupta described.

“Like on Pixel 6 and Pixel 7, you can see all the awesome improvements that we have landed, and a ton of them have been the first on Pixel. So we’re extremely comfortable with that approach.”

The correct method for Tensor?

It is absolutely genuine that benchmarks aren’t totally agent of actual-entire world final results and workloads. For 1, several benchmark tests target on peak functionality that you will almost never see somewhat than sustained overall performance. In fact, our possess Tensor G2 testing reveals that the Pixel 7 chipset lagged guiding rivals when it arrived to peak results but basically beat some Snapdragon 8 Gen 1 telephones in sustained screening. And sustained general performance is a must for tasks like online games and pictures.

What do you consider of benchmark checks?

7 votes

Google is also pushing device learning in a major way for tasks like computational pictures, and benchmark apps like Geekbench and 3DMark really do not genuinely consider this components into account. There are some AI benchmarks out there, but the varying nature of AI components and workloads signifies that even these assessments don’t inform the entire tale. For what it’s truly worth, the AI-Benchmark rating shows that the Pixel 7’s Tensor G2 SoC slots in powering Snapdragon 8 Gen 1 and Dimensity 9000 telephones.

It will have to also be claimed that this isn’t a new approach for Google. The authentic Tensor chipset viewed in the Pixel 6 makes use of two Cortex-X1 CPU cores, and the business observed at the time that this solution was meant to produce optimum effectiveness at “medium” workloads. The company asserted that 1 Cortex-X CPU main (as witnessed in rival SoC types) was only practical for successful benchmarks.