Chinese tech giant Baidu recently released its own open source online benchmarking service that can determine how fast processors can train neural networks which can be used for machine learning. Dubbed as DeepBench, the launch of the service was accompanied by the posting of the first results with Nvidia and Intel processors running the benchmark.

Like Us on Facebook

Among the benchmark tests provided by DeepBench include mostly low-level operations like convolutions, matrix multiplication, handing of recurrent layers and the time it takes to share date with all processors within a cluster.

Tech giants such as Google, Facebook and Baidu have all invested significant amounts of resources and workload towards machine learning. Among the services they provide are speech, video and object recognition, as well as automatic language translation. The initial DeepBench results used an Nvidia Titan, Titan X with Pascal architecture, M40 graphics processing units and Intel Xeon Phi processors.

In a statement acquired by EE Times, Baidu senior researcher Greg Diamos said, "Even for the same type of operations like matrix multiplies, depending on the sizes of the models and the ways they are used, performance varies even on the same processor."

Diamos added that the test undertaken were not focused on minor differences of the processors, instead they are pushing both processors to be 10 times faster.

Further, Baidu is hoping that other semiconductor manufacturers will take notice of its newly-established platform, the Deepbench, and contribute to expanding its benchmark. Baidu is also looking forward to data center operators to use its benchmark platform to test their systems.

TagsBaidu, AI, Artificial Intelligence, DeepBench, AI benchmark, Baidu AI Benchmark Platform Deepbench