HPC Technique Propels Deep Learning at Scale

By Tiffany Trader

February 21, 2017

Researchers from Baidu’s Silicon Valley AI Lab (SVAIL) have adapted a well-known HPC communication technique to boost the speed and scale of their neural network training and now they are sharing their implementation with the larger deep learning community.

The technique, a modified version of the OpenMPI algorithm “ring all-reduce,” is being used at Baidu to parallelize the training of their speech recognition model, Deep Speech 2, across many GPU nodes. The two pieces of software Baidu is announcing today are the baidu-allreduce C library, as well as a patch for TensorFlow, which allows people who have already modeled in TensorFlow to compile this new version and use it for parallelizing across many devices. The codes are available on GitHub.

Ring all-reduce – all GPUs send data simultaneously

Baidu’s SVAIL team developed the approach about two years ago for their internal deep learning framework, named Gene and Majel (in tribute to the famous Star Trek creator and the actress who voiced the onboard computer interfaces for the series). The technique is commonplace in HPC circles, but underused within artificial intelligence and deep learning, according to Baidu.

Many of the researchers in the SVAIL group had come from the high performance computing space and recognized the competitive edge it offered.

“The algorithm is actually part of OpenMPI, but the OpenMPI implementation is not as fast,” comments Baidu Research Scientist Shubho Sengupta. “So the way we stumbled upon it was we started using OpenMPI for doing training and we realized it was not scaling to the extent that we want it to scale. I started digging through the OpenMPI source, found the algorithm, saw that it’s not very efficient, and reimplemented it.”

The SVAIL researchers wrote their own implementation of the ring algorithm for higher performance and better stability. The key distinction from the OpenMPI version is that the SVAIL implementation avoids extraneous copies between the CPU and GPU.

Explains Sengupta, “Once OpenMPI does the communication of these matrices, if the matrices are in GPU memory, it actually copies to CPU memory to do the reduction part of it – that’s actually quite wasteful. You don’t really need to do a copy, you could just write a small kernel that does the reduction in GPU memory space itself. And this especially helps when you are doing all-reduce within a node and all the GPUs are within a PCI root complex, then it doesn’t do any of the copies actually – it can just do everything in GPU memory space. This very simple idea of eliminating this copy resulted in this speedup in scaling over OpenMPI’s own implementation.”

Employing this algorithm along with SVAIL’s focus on fast networking (InfiniBand) and careful hardware-software codesign has enabled the team to get linear GPU scaling up to 128 GPUs, an achievement that was detailed in their December 2015 paper, “Deep Speech 2: End-to-End Speech Recognition in English and Mandarin.”

With their internal implementation of ring all-reduce, the team achieves between a 2.3-21.4X speedup over OpenMPI (version 1.8.5) depending on the number of GPUs.

Sengupta notes that their implementation is fastest for a small number of GPUs. “At 8 GPUs it’s about 20x faster, then as you increase the number of GPUs, it drops because now you actually have to copy data to the CPU to send across the network. But for the internal framework, we can scale all the way up to 128 GPUs and get linear scaling.”

Comparison of two different all-reduce implementations. All times are in seconds. Performance gain is the ratio of OpenMPI all-reduce time to SVAIL’s all-reduce time. (Source: Deep Speech 2 paper)

Sengupta’s teammate Baidu Research Scientist Andrew Gibiansky says similar benefits can now be seen with TensorFlow: “In terms of the TensorFlow implementation, we get the same linear scaling path past eight. In terms of a comparison with running on a single GPU, it ends up being about 31x faster at 40 GPUs.”

After the Deep Speech 2 paper was published, the SVAIL team began getting requests from the community who wanted to know more about the implementation. Given that the algorithm is pretty tightly coupled to SVAIL’s proprietary deep learning framework, they needed to come up with a different way to release it, so they created two new implementations, one specifically for TensorFlow and one that is more general.

Gibiansky, who led the work on the TensorFlow patch, describes their multi-pronged approach to disseminating the information. “You can read the blog post [for a thorough technical explanation] and figure it out. If you’re using TensorFlow, you can use our modification to train your own models with this. And if you’re a deep learning author, you can look at our C library and integrate that. The goal is really to take this idea we’ve found to be really successful internally and try to start spreading it so that other people can also take advantage of it.”

Sengupta shares an interesting perspective on the opportunities to be mined for deep learning within HPC.

“With MPI – people [in deep learning] think that it is this old technology, that it is not relevant, but I think because of our work we have shown that you can build very fast collectives using MPI and that allows you to do synchronous gradient descent which converges faster, gives you deterministic results and you don’t need to do asynchronous gradient descent with parameter servers which was the dominant way of doing this when we first started,” says Sengupta.

As for the reduced-copy approach propagating back to MPI, Gibiansky notes that if you look at some of the other MPI implementations, they’re slowly moving their collectives to GPU versions. “MVPICH recently introduced an all-gather that doesn’t end up copying to CPU – so OpenMPI will probably get there, it just might take a while. Potentially giving this a little more visibility, we can spur that on.”

“There’s a lot of interest now in collectives and one thing we also realized is the all-reduce operation used in traditional HPC setups, it actually transfers data that’s actually not very large,” Sengupta adds. “What it usually does, when I talk to HPC people, it’s trying to figure out the status of something across a bunch of machines – while in deep learning we are transferring these large matrices – like 2048×2048, essentially 4 million 32-bit floating points. For the traditional HPC community, this is a very atypical input for all-reduce. The traditional HPC community does not actually use all-reduce with really large data sizes. I think with deep learning, more and more people are realizing that collective operations for really large matrices is also very important.”

A detailed explanation of ring all-reduce and Baidu’s GPU implementation is covered in this technical blog post, published today by Baidu Research. A variant of the technique is also used to provide high-performance node-local scaling for PaddlePaddle, the company’s open source deep learning framework.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industy updates delivered to you every week!

Nvidia Touts Strong Results on Financial Services Inference Benchmark

February 3, 2023

The next-gen Hopper family may be on its way, but that isn’t stopping Nvidia’s popular A100 GPU from leading another benchmark on its way out. This time, it’s the STAC-ML inference benchmark, produced by the Securi Read more…

Quantum Computing Firm Rigetti Faces Delisting

February 3, 2023

Quantum computing companies are seeing their market caps crumble as investors patiently await out the winner-take-all approach to technology development. Quantum computing firms such as Rigetti Computing, IonQ and D-Wave went public through mergers with blank-check companies in the last two years, with valuations at the time of well over $1 billion. Now the market capitalization of these companies are less than half... Read more…

US and India Strengthen HPC, Quantum Ties Amid Tech Tension with China

February 2, 2023

Last May, the United States and India announced the “Initiative on Critical and Emerging Technology” (iCET), aimed at expanding the countries’ partnerships in strategic technologies and defense industries across th Read more…

Pittsburgh Supercomputing Enables Transparent Medicare Outcome AI

February 2, 2023

Medical applications of AI are replete with promise, but stymied by opacity: with lives on the line, concerns over AI models’ often-inscrutable reasoning – and as a result, possible biases embedded in those models Read more…

Europe’s LUMI Supercomputer Has Officially Been Accepted

February 1, 2023

“LUMI is officially here!” proclaimed the headline of a blog post written by Pekka Manninen, director of science and technology for CSC, Finland’s state-owned IT center. The EuroHPC-organized supercomputer’s most Read more…

AWS Solution Channel

Shutterstock 2069893598

Cost-effective and accurate genomics analysis with Sentieon on AWS

This blog post was contributed by Don Freed, Senior Bioinformatics Scientist, and Brendan Gallagher, Head of Business Development at Sentieon; and Olivia Choudhury, PhD, Senior Partner Solutions Architect, Sujaya Srinivasan, Genomics Solutions Architect, and Aniket Deshpande, Senior Specialist, HPC HCLS at AWS. Read more…

Microsoft/NVIDIA Solution Channel

Shutterstock 1453953692

Microsoft and NVIDIA Experts Talk AI Infrastructure

As AI emerges as a crucial tool in so many sectors, it’s clear that the need for optimized AI infrastructure is growing. Going beyond just GPU-based clusters, cloud infrastructure that provides low-latency, high-bandwidth interconnects and high-performance storage can help organizations handle AI workloads more efficiently and produce faster results. Read more…

Intel’s Gaudi3 AI Chip Survives Axe, Successor May Combine with GPUs

February 1, 2023

Intel's paring projects and products amid financial struggles, but AI products are taking on a major role as the company tweaks its chip roadmap to account for more computing specifically targeted at artificial intellige Read more…

Quantum Computing Firm Rigetti Faces Delisting

February 3, 2023

Quantum computing companies are seeing their market caps crumble as investors patiently await out the winner-take-all approach to technology development. Quantum computing firms such as Rigetti Computing, IonQ and D-Wave went public through mergers with blank-check companies in the last two years, with valuations at the time of well over $1 billion. Now the market capitalization of these companies are less than half... Read more…

US and India Strengthen HPC, Quantum Ties Amid Tech Tension with China

February 2, 2023

Last May, the United States and India announced the “Initiative on Critical and Emerging Technology” (iCET), aimed at expanding the countries’ partnership Read more…

Intel’s Gaudi3 AI Chip Survives Axe, Successor May Combine with GPUs

February 1, 2023

Intel's paring projects and products amid financial struggles, but AI products are taking on a major role as the company tweaks its chip roadmap to account for Read more…

Roadmap for Building a US National AI Research Resource Released

January 31, 2023

Last week the National AI Research Resource (NAIRR) Task Force released its final report and roadmap for building a national AI infrastructure to include comput Read more…

PFAS Regulations, 3M Exit to Impact Two-Phase Cooling in HPC

January 27, 2023

Per- and polyfluoroalkyl substances (PFAS), known as “forever chemicals,” pose a number of health risks to humans, with more suspected but not yet confirmed Read more…

Multiverse, Pasqal, and Crédit Agricole Tout Progress Using Quantum Computing in FS

January 26, 2023

Europe-based quantum computing pioneers Multiverse Computing and Pasqal, and global bank Crédit Agricole CIB today announced successful conclusion of a 1.5-yea Read more…

Critics Don’t Want Politicians Deciding the Future of Semiconductors

January 26, 2023

The future of the semiconductor industry was partially being decided last week by a mix of politicians, policy hawks and chip industry executives jockeying for Read more…

Riken Plans ‘Virtual Fugaku’ on AWS

January 26, 2023

The development of a national flagship supercomputer aimed at exascale computing continues to be a heated competition, especially in the United States, the Euro Read more…

Leading Solution Providers

Contributors

SC22 Booth Videos

AMD @ SC22
Altair @ SC22
AWS @ SC22
Ayar Labs @ SC22
CoolIT @ SC22
Cornelis Networks @ SC22
DDN @ SC22
Dell Technologies @ SC22
HPE @ SC22
Intel @ SC22
Intelligent Light @ SC22
Lancium @ SC22
Lenovo @ SC22
Microsoft and NVIDIA @ SC22
One Stop Systems @ SC22
Penguin Solutions @ SC22
QCT @ SC22
Supermicro @ SC22
Tuxera @ SC22
Tyan Computer @ SC22
  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire