MLPerf v0.5 training results

By: David Kanter (dkanter.delete@this.realworldtech.com), December 12, 2018 3:03 pm
Room: Moderated Discussions
As many of you know, machine learning is one of the hottest areas today...but it is such a rapidly developing field that we do not have standard workloads or benchmarks. This means that the industry does not have robust ways to meaningfully compare today's existing hardware and software, let alone research and the systems of tomorrow.

For the last year, I have been working with the MLPerf community to help create a set of industry-standard machine learning benchmarks. MLPerf is a fantastic organization with participation from over 50 leaders in industry (e.g., Google, Nvidia, Baidu, Intel) and 7 major universities (e.g., Harvard and Stanford).

I'm thrilled to let you know that our first round of training benchmarks are now public. The v0.5 suite include 7 different benchmarks including areas such as object recognition, automatic language translation, and recommendation engines. The v0.5 training benchmarks include about 30 different submissions from the excellent teams at Google, Intel, and Nvidia.

You can see the results at https://mlperf.org/results/. I'd like to give a huge round of thanks to the whole MLPerf organization and the submitting companies. This was a tremendous effort from dozens of people and will be a huge boon the industry.

David
TopicPosted ByDate
MLPerf v0.5 training resultsDavid Kanter2018/12/12 03:03 PM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell green?