MLPerf Results Validate CPUs for Deep Learning Training

I have worked on optimizing and benchmarking computer performance for more than two decades, on platforms ranging from supercomputers and database servers to mobile devices. It is always fun to highlight performance results for the product you are building and compare them with others in the industry. SPEC*, LINPACK*, and TPC* have become familiar names to many of us. Now, MLPerf* is filling in the void of benchmarking for Machine Learning.

I am excited to see the Intel® Xeon® Scalable processor MLPerf results submitted by our team because we work on both the user side and the computer system development side of deep learning. These results show that Intel® Xeon® Scalable processors have surpassed a performance threshold where they can be an effective option for data scientists looking to run multiple workloads on their infrastructure without investing in dedicated hardware.1 2 3

Back in 2015, I had a team working on mobile devices. We had to hire testers to manually play mobile games. It was fun initially for the testers, then it became boring and costly. One tester we hired quit on the same day. Our team created a robot to test mobile games and adopted deep learning. Our game testing robot played games automatically and found more bugs than human testers. We wanted to train neural networks on the machines we already had in the lab, but they were not fast enough. I had to allocate budget for the team to buy a GPU, an older version than the MLPerf reference GPU.4

Today CPUs are capable of deep learning training as well as inference. Our MLPerf Intel® Xeon® Scalable processor results compare well with the MLPerf reference GPU4 on a variety of MLPerf deep learning training workloads.1 2 3 For example, the single-system two-socket Intel® Xeon® Scalable processor results submitted by Intel achieved a score of 0.85 on the MLPerf Image Classification benchmark (Resnet-50)1; 1.6 on the Recommendation benchmark (Neural Collaborative Filtering NCF)2; and 6.3 on Reinforcement Learning benchmark (mini GO).3 In all these scores, 1.0 is defined as the score of the reference implementation on the reference GPU.4 For all the preceding results, we use FP32, the common numerical precision used in today’s market. From these MLPerf results, we can see that our game testing robot could easily train on Intel® Xeon® Scalable processors today.

The deep learning and machine learning world continues to evolve from image processing using Convolutional Neural Networks (CNN) and natural language processing using Recurrent Neural Networks (RNN) to recommendation systems using MLP layers and general matrix multiply, reinforcement learning (mixing CNN and simulation) and hybrid models mixing deep learning and classical machine learning. A general purpose CPU is very adaptable to this dynamically changing environment, in addition to running existing non-DL workloads.

Enterprises have adopted CPUs for deep learning training. For example, today, Datatonic* published a blog showing up to 11x cost and 57 percent performance improvement when running a neural network recommender system used in production by a top-5 UK retailer on a Google Cloud* VM powered by Intel® Xeon® Scalable processors.5 CPUs can also accommodate the large memory models required in many domains. The pharmaceutical company Novartis used Intel® Xeon® Scalable processors to accelerate training for a multiscale convolutional neural network (M-CNN) for 10,000 high-content cellular microscopic images, which are much larger in size than the typical ImageNet* images, reducing time to train from 11 hours to 31 minutes.6

High performance computing (HPC) customers use Intel® Xeon® processors for distributed training, as showcased at Supercomputing 2018. For instance, GENCI/CINES/INRIA trained a plant classification model for 300K species on a 1.5TByte dataset of 12 million images using 128 2S Intel® Xeon® processor-based systems.7 DELL EMC* and SURFSara used Intel® Xeon® processors to reduce training time to 11 minutes for a DenseNet-121 model.8 CERN* showcased distributed training using 128 nodes of the TACC Stampede 2 cluster (Intel® Xeon® Platinum 8160 processor, Intel® OPA) with a 3D Generative Adversarial Network (3D GAN) achieving 94% scaling efficiency.9 Additional examples can be found at https://software.intel.com/en-us/articles/intel-processors-for-deep-learning-training.

CPU hardware and software performance for deep learning has increased by a few orders of magnitude in the past few years. Training that used to take days or even weeks can now be done in hours or even minutes. This level of performance improvement was achieved through a combination of hardware and software. For example, current-generation Intel® Xeon® Scalable processors added both the Intel® Advanced Vector Extensions 512 (Intel® AVX-512) instruction set (longer vector extensions) to allow a large number of operations to be done in parallel, and with a larger number of cores, essentially becoming a mini-supercomputer. The next-generation Intel® Xeon® Scalable processor adds Intel® Deep Learning Boost (Intel® DL Boost): higher throughput, lower numerical precision instructions to boost deep learning inference. On the software side, the performance difference between the baseline open source deep learning software, and the Intel-optimized software can be up to 275X10 on the same Intel® Xeon® Scalable processor (as illustrated in a demo I showed at the Intel Architecture Day forum yesterday).

Over the past few years, Intel has worked with DL framework developers to optimize many popular open source frameworks such as TensorFlow*, Caffe*, MXNet*, PyTorch*/Caffe2*, PaddlePaddle* and Chainer*, for Intel® processors. Intel has also designed a framework, BigDL for SPARK*, and the Intel® Deep Learning Deployment Toolkit (DLDT) for inference. Since the core computation is linear algebra, we have created a new math library, Intel® Math Kernel Library for Deep Neural Networks (Intel® MKL-DNN), specifically for deep learning, based on many years of experience with the Intel® Math Kernel Library (MKL) for high performance computing (HPC). The integration of Intel MKL-DNN into the frameworks, and the additional optimizations contributed to the frameworks to fully utilize the underlying hardware capabilities, are the key reason for the huge software performance improvement.

I’ve often been asked whether CPUs are faster or slower than accelerators. Of course, accelerators have certain advantages. For a specific domain, if an accelerator is not generally faster than a CPU, then it is not much of an accelerator. Even so, given the increasing variety of deep learning workloads, in some cases, a CPU may be as fast or faster while retaining that flexibility that is core to the CPU value proposition. Thus, the more pertinent question is whether CPUs can run deep learning well enough to be an effective option for customers that don’t wish to invest in accelerators. These initial MLPerf results1 2 3, as well as our customer examples, show that CPUs can indeed be effectively used for training. Intel’s strategy is to offer both general purpose CPUs and accelerators to meet the machine learning needs of a wide range of customers.

Looking forward, we are continuing to add new AI and deep learning features to our future generations of CPUs, like Intel® Deep Learning Boost (Intel® DL Boost), plus bfloat16 for training, as well as additional software optimizations. Please stay tuned. For more information on Intel® software optimizations, see ai.intel.com/framework-optimizations. For more information on Intel® Xeon® Scalable processors, see intel.co.id/xeonscalable.

Informasi Produk dan Performa

1

Skor 0,85 pada benchmark Klasifikasi Gambar MLPerf (Resnet-50) 0,85X lebih dari baseline MLPerf(+) yang menggunakan 2 chip prosesor Intel® Xeon® Platinum 8180. Divisi MLPerf v0.5 training Tertutup; sistem menerapkan Optimalisasi Intel® untuk Caffe* 1.1.2a dengan Intel® Math Kernel Library untuk Deep Neural Networks (Intel® MKL-DNN) v0.16 library. Diambil dari www.mlperf.org 12 Desember 2018, entri 0.5.6.1. Nama dan logo MLPerf adalah merek dagang. Lihat www.mlperf.org untuk informasi lebih lanjut.

2

Skor 1,6 pada benchmark Rekomendasi (Neural Collaborative Filtering NCF) 1,6X lebih dari baseline MLPerf(+) yang menggunakan 2 chip prosesor Intel® Xeon® Platinum 8180. Divisi MLPerf v0.5 training Tertutup; sistem menerapkan Framework BigDL 0.7.0. Diambil dari www.mlperf.org 12 Desember 2018, entri 0.5.9.6. Nama dan logo MLPerf adalah merek dagang. Lihat www.mlperf.org untuk informasi lebih lanjut.

3

Skor 6,3 pada benchmark Pembelajaran Penegasan (mini GO) 6,3X lebih dari baseline MLPerf(+) yang menggunakan 2 chip prosesor Intel® Xeon® Platinum 8180. Divisi MLPerf v0.5 training Tertutup; sistem menerapkan TensorFlow 1.10.1 dengan Intel® Math Kernel Library untuk Deep Neural Networks (Intel® MKL-DNN) v0.14 library. Diambil dari www.mlperf.org 12 Desember 2018, entri 0.5.10.7. Nama dan logo MLPerf adalah merek dagang. Lihat www.mlperf.org untuk informasi lebih lanjut.

Baseline (+) MLPerf (diadopsi dari Uraian Singkat Pers Komunitas MLPerf v0.5): MLPerf Training v0.5 adalah rangkaian benchmark untuk mengukur kecepatan sistem ML. Setiap benchmark MLPerf Training didefinisikan berdasarkan Kumpulan Data dan Target Kualitas. MLPerf Training juga memberikan implementasi referensi untuk setiap benchmark yang menggunakan model tertentu. Tabel berikut merangkum tujuh benchmark dalam versi v0.5 dari rangkaian.

Benchmark

Kumpulan Data

Target Kualitas

Model Implementasi Referensi

Klasifikasi gambar

ImageNet

74,90% klasifikasi

Resnet-50 v1.5

Deteksi objek (ringan)

COCO 2017

21,2% mAP

SSD (Resnet-34 backbone)

Deteksi objek (berat)

COCO 2017

0,377 Box min AP, 0,339 Mask min AP

Mask R-CNN

Terjemahan (pengulangan)

WMT Inggris-Jerman

21,8 BLEU

Terjemahan Mesin Neural

Terjemahan (nonpengulangan)

WMT Inggris-Jerman

25,0 BLEU

Transformator

Rekomendasi

MovieLens-20M

0,635 HR@10

Pemfilteran Kolaboratif Neural

Pembelajaran penegasan

Game profesional

40,00% prediksi gerakan

Mini Go


Aturan pelatihan MLPerf: https://github.com/mlperf/training_policies/blob/master/training_rules.adoc

4

Sistem referensi MLPerf*: Google Cloud Platform konfig: 16 vCPU, Intel Skylake atau lebih baru, RAM 60 GB (n1­standar­16), 1 NVIDIA* Tesla* P100 GPU, CUDA* 9.1 (9.0 untuk TensorFlow*), nvidia­docker2, Ubuntu* 16.04 LTS, Pra­emtibilitas: mati, Mulai ulang otomatis: mati, disk boot 30 GB + 1 disk persisten SSD 500 GB, docker* image: 9.1­cudnn7­runtime­ubuntu16.04 (9.0­cudnn7­devel­ubuntu16.04 untuk TensorFlow*).

6

Novartis: Diukur pada 25 Mei 2018. Berdasarkan kenaikan kecepatan untuk 8 node yang relatif terhadap satu node. Konfigurasi node: CPU: Prosesor Intel® Xeon® Gold 6148 @ 2,4 GHz, memori 192 GB, Hyper-threading: Aktif. NIC: Intel® Omni-Path Host Fabric Interface (Intel® OP HFI), TensorFlow: v1.7.0, Horovod: 0.12.1, OpenMPI: 3.0.0. OS: CentOS* 7.3, OpenMPU 23.0.0, Python 2.7.5. Waktu Pelatihan untuk memusatkan akurasi model hingga 99%. Sumber: https://newsroom.intel.com/news/using-deep-neural-network-acceleration-image-analysis-drug-discovery.

7

GENCI: Occigen: 3306 node x 2 prosesor Intel® Xeon® (12-14 core). Node Komputasi: 2 soket prosesor Intel® Xeon® dengan 12 core @ 2,70 GHz untuk total 24 core per node, 2 thread per core, DDR4 96 GB, Mellanox InfiniBand Fabric Interface, rel ganda. Perangkat Lunak: Intel® MPI Library 2017 Update 4 Intel® MPI Library 2019 Technical Preview OFI 1.5.0PSM2 dengan Multi-EP, Ethernet 10 Gbit, SSD lokal 200 GB, Red Hat* Enterprise Linux 6.7. Caffe*: Optimalisasi Intel® untuk Caffe*: https://github.com/intel/caffe Intel® Machine Learning Scaling Library (Intel® MLSL): https://github.com/intel/MLSL Kumpulan data: Pl@ntNet: CINES/GENCI Hasil Performa Kumpulan Data Internal didasarkan pada pengujian per 15/10/2018.

8

Kolaborasi Intel, Dell, dan Surfsara: Diukur tanggal 17/5/2018 pada 256x node 2 soket prosesor Intel® Xeon® Gold 6148. Node Komputasi: 2 soket prosesor Intel® Xeon® Gold 6148F dengan 20 core @ 2,40 GHz untuk total 40 core per node, 2 Thread per core, L1d 32K; L1i cache 32K; L2 cache 1024K; L3 cache 33792K, DDR4 96 GB, Intel® Omni-Path Host Fabric Interface (Intel® OP HFI), rel ganda. Perangkat Lunak: Intel® MPI Library 2017 Update 4 Intel® MPI Library 2019 Technical Preview OFI 1.5.0PSM2 dengan Multi-EP, Ethernet 10 Gbit, SSD lokal 200 GB, Red Hat* Enterprise Linux 6.7. TensorFlow* 1.6: Dibangun & Diinstal dari sumber: https://www.tensorflow.org/install/install_sources ResNet-50 Model: Spesifikasi topologi dari https://github.com/tensorflow/tpu/tree/master/models/official/resnet. DenseNet-121Model: Spesifikasi topologi dari https://github.com/liuzhuang13/DenseNet. Konvergensi & Performa Model: https://surfdrive.surf.nl/files/index.php/s/xrEFLPvo7IDRARs. Kumpulan data: ImageNet2012-1K: http://www.image-net.org/challenges/LSVRC/2012 /. ChexNet*: https://stanfordmlgroup.github.io/projects/chexnet/. Performa diukur dengan: OMP_NUM_THREADS=24 HOROVOD_FUSION_THRESHOLD=134217728 export I_MPI_FABRICS=tmi, export I_MPI_TMI_PROVIDER=psm2 \ mpirun -np 512 -ppn 2 python resnet_main.py –train_batch_size 8192 –train_steps 14075 –num_intra_threads 24 –num_inter_threads 2 — mkl=True –data_dir=/scratch/04611/valeriuc/tf-1.6/tpu_rec/train –model_dir model_batch_8k_90ep –use_tpu=False –kmp_blocktime 1. https://ai.intel.com/diagnosing-lung-disease-using-deep-learning/.

9

CERN: Diukur tanggal 17/5/2018 pada Stampede2/TACC: https://portal.tacc.utexas.edu/user-guides/stampede2. Node komputasi: 2 soket prosesor Intel® Xeon® Platinum 8160 dengan 24 core @ 2,10 GHz untuk total 48 core per node, 2 Thread per core, L1d 32K; L1i cache 32K; L2 cache 1024K; L3 cache 33792K, DDR4 96 GB, Intel® Omni-Path Host Fabric Interface (Intel® OP HFI), rel ganda. Perangkat Lunak: Intel® MPI Library 2017 Update 4 Intel® MPI Library 2019 Technical Preview OFI 1.5.0PSM2 dengan Multi-EP, Ethernet 10 Gbit, SSD lokal 200 GB, Red Hat* Enterprise Linux 6.7. TensorFlow* 1.6: Dibangun & Diinstal dari sumber: https://www.tensorflow.org/install/install_sources Model: CERN* 3D GANS dari https://github.com/sara-nl/3Dgan/tree/tf Kumpulan data: CERN* 3D GANS dari https://github.com/sara-nl/3Dgan/tree/tf Performa diukur pada 256 Node dengan: OMP_NUM_THREADS=24 HOROVOD_FUSION_THRESHOLD=134217728 export I_MPI_FABRICS=tmi, export I_MPI_TMI_PROVIDER=psm2 \ mpirun -np 512 -ppn 2 python resnet_main.py –train_batch_size 8 \ –num_intra_threads 24 –num_inter_threads 2 –mkl=True \ –data_dir=/path/to/gans_script.py –kmp_blocktime 1. https://www.rdmag.com/article/2018/11/imagining-unthinkable-simulations-without-classical-monte-carlo.

10

Peningkatan kinerja throughput inferensi 275X dengan Optimalisasi Intel® untuk Caffe* dibandingkan dengan BVLC-Caffe*: Intel diukur pada 11/12/2018. 2S CPU prosesor Intel® Xeon® Platinum 8180 @ 2,50 GHz (28 core), HT AKTIF, turbo AKTIF, total memori 192 GB (12 slot * 16 GB, Micron 2666MHz), Intel® SSD SSDSC2KF5, Ubuntu 16.04 Kernel 4.15.0-42.generic; BIOS: SE5C620.86B.00.01.0009.101920170742 (mikrokode: 0x0200004d); Topologi: Resnet-50 Baseline: FP32, BVLC Caffe* (https://github.com/BVLC/caffe.git) commit 99bd99795dcdf0b1d3086a8d67ab1782a8a08383 Performa Saat Ini: INT8, Optimalisasi Intel® untuk Caffe* (https://github.com/Intel/caffe.git) commit: Caffe* commit: e94b3ff41012668ac77afea7eda89f07fa360adf, MKLDNN commit: 4e333787e0d66a1dca1218e99a891d493dbc8ef1.

Perangkat lunak dan beban kerja yang digunakan pada pengujian performa mungkin telah dioptimalkan hanya untuk performa pada mikroprosesor Intel. Uji perfoma, seperti SYSmark* dan MobileMark*, diukur menggunakan sistem, komponen, perangkat lunak, pengoperasian, dan fungsi komputer tertentu. Perubahan apa pun terhadap salah satu faktor tersebut bisa menyebabkan hasil yang bervariasi. Anda harus mempelajari informasi dan sejumlah uji performa lain agar benar-benar dapat mempertimbangkan pembelian dengan cermat, termasuk performa produk tersebut bila digabungkan dengan produk lainnya. Untuk informasi lebih lanjut, kunjungi www.intel.co.id/benchmarks.

Pemberitahuan Optimasi: Pengompilasi Intel bisa atau tidak bisa mengoptimalkan hingga tingkat yang sama bagi mikroprosesor non-Intel untuk optimsasi yang tidak unik bagi mikroprosesor Intel. Optimasi ini meliputi rangkaian instruksi SSE2, SSE3, dan SSSE3 dan optimasi lainnya. Intel tidak menjamin ketersediaan, fungsionalitas, atau keefektifan optimisasi pada mikroprosesor yang tidak diproduksi oleh Intel. Optimisasi yang tergantung pada mikroprosesor dalam produk ini dimaksudkan untuk digunakan dengan mikroprosesor Intel. Optimisasi tertentu yang tidak secara khusus untuk mikroarsitektur Intel disediakan bagi mikroprosesor Intel. Harap lihat Panduan Referensi dan Pengguna produk yang berlaku untuk informasi lebih lanjut mengenai serangkaian instruksi tertentu yang disediakan dalam pemberitahuan ini.

Hasil performa mungkin tidak mencerminkan semua pembaruan keamanan yang tersedia secara umum. Lihat pengungkapan konfigurasi untuk detailnya. Tidak ada produk yang sepenuhnya aman.

Intel, logo Intel, prosesor Xeon yang Dapat Diskalakan, Deep Learning Boost adalah merek dagang dari Intel Corporation atau anak perusahaannya di AS dan/atau negara lainnya.
*Nama dan merek lain adalah hak milik dari masing-masing pemiliknya.
© Intel Corporation.