Google wins MLPerf benchmark contest with fastest ML training supercomputer | Google Cloud Blog

Google wins MLPerf benchmark contest with fastest ML training supercomputer | Google Cloud Blog

Google set performance records in six out of the eight MLPerf benchmarks at the latest MLPerf benchmark contest

113 mentions: @JeffDean@jekbradbury@fchollet@kazunori_279@jaguring1@uhoelzle@hichaelmart@gneubig
Date:

Referring Tweets

@JeffDean
@JeffDean Very excited to see the MLPerf 0.7 results released today, where Google TPUs set records in six of the eight benchmarks! We need bigger benchmarks, because we can now train the ResNet-50, BERT, Transformer, & SSD benchmarks each in under 30 seconds. t.co/J6vbZ8srs1
@jekbradbury
@jekbradbury In 2016, when I was working on machine translation, it took me more than a week on a multi-GPU machine to train a competitive system on WMT English-German. Today, JAX on a TPU v3 supercomputer can train a better model on the same data in 16 seconds! t.co/fRK6vwZHDP t.co/Rof5Ba9saZ
@fchollet
@fchollet TensorFlow 2 is fast -- significantly faster than alternatives on 6 key benchmarks from MLPerf. t.co/glkVf0vqf3 t.co/FWpFIBZycB
@kazunori_279
@kazunori_279 Google、4096個(8192コア)のTPU v3を搭載したAIスパコンで430PFlopsを達成、MLPerfの世界記録を更新。また単体性能が2.7倍のTPU v4も新たに登場。 #gcpja t.co/14nLZ3i9vI
@jaguring1
@jaguring1 グーグルが世界最速の機械学習用スパコンを業界標準のベンチマーク「MLPerf」に投入。 TPU v3よりも性能が平均2.7倍向上したTPU v4も紹介してる。 ResNet-50、BERT、Transformer、SSDを30秒以内で学習できる。この5年でほぼ5桁(10,000倍)早く学習できるようになったそう t.co/F9Do1ruN1R
@uhoelzle
@uhoelzle Google breaks AI performance records in MLPerf with world's fastest training supercomputer. The headline says it all, but beyond that the overall industry progress is astounding: The fastest results improved by an average of 2.7x (!) @googlecloud t.co/mmOXopbE4v
@hichaelmart
@hichaelmart Damnnn, training BERT in under 25 secs 😲 t.co/7c0zvm55S6
@gneubig
@gneubig Large model/hardware trivia: Google's new TPU supercomputer (t.co/52qv2Gdn42) could potentially train GPT-3 (t.co/oRjMCdwvfM) in about 7.5 days. Actually a bit longer than I expected. (GPT-3 175B model requires 3.14E+23 flops, Google cluster does 480PFLOPs/s)
@danijarh
@danijarh No need to wait for your experiments to finish anymore! t.co/gSD7n6T7FO t.co/q9CgLXN0eq

Bookmark Comments

id:misshiki
id:Nyoho
id:koba789

Related Entries

Create a Model Card with Scikit-Learn | Google Cloud Blog
Read more Create a Model Card with Scikit-Learn | Google Cloud Blog
0 users, 21 mentions 2020/10/21 08:21
Google Cloud launches machine learning engineer certification | Google Cloud Blog
Read more Google Cloud launches machine learning engineer certification | Google Cloud Blog
1 users, 27 mentions 2020/11/11 23:21
BigQuery Omni for multi-cloud data analytics | Google Cloud Blog
Read more BigQuery Omni for multi-cloud data analytics | Google Cloud Blog
0 users, 3 mentions 2021/01/14 15:51
Fuel your custom models with Vertex AI | Google Cloud Blog
Read more Fuel your custom models with Vertex AI | Google Cloud Blog
2 users, 11 mentions 2021/08/12 21:09
GitHub - kwchurch/ACL2022_deepnets_tutorial: Materials for ACL-2022 tutorial: A Gentle Introduction ...
Read more GitHub - kwchurch/ACL2022_deepnets_tutorial: Materials for ACL-2022 tutorial: A Gentle Introduction ...
0 users, 1 mentions 2022/04/28 18:09

ML-Newsについて

機械学習の技術に関する情報は流速も早いし、分野も多様でキャッチアップが大変です。Twitterで機械学習用のリストを作っても、普段は機械学習以外の話題が多く流れており、効率的に情報収集するのは困難です。

ML-NewsはSNSを情報源とした機械学習に特化したニュースサイトです。機械学習に関する論文ブログライブラリコンペティション発表資料勉強会などの最新の情報を効率的に収集できます。

機械学習を応用した自然言語処理、画像認識、情報検索などの分野の情報や機械学習で必要になるデータ基盤やMLOpsの話題もカバーしています。
安定したサイト運営のためにGitHub sponsorを募集しています。

お知らせ

  • 2021/12/31: デザインを刷新しました
  • 2021/04/08: 日本語Kaggleのカテゴリを新設しました