HPC in Life Sciences Part 2: Penetrating AI’s Hype and the Cloud’s Haze

By John Russell

February 25, 2019

Weary of the constant din of AI hype. So is Ari Berman, vice president and general manager of consulting services for BioTeam, a research computing consultancy specializing in life sciences.

“Every vendor is selling AI. I think it has become the gluten-free tag of life sciences because it is everywhere. Putting gluten free on bacon packages; why would you do that of course bacon is gluten free. I sort of see it like that. People have used it as a marketing gimmick to catch your eye,” said Berman in a recent interview with HPCwire to review state of HPC (and AI) in the life sciences that’s become an annual exercise for HPCwire and BioTeam.

Exasperation vented, Berman quickly zeroed in on two biomedical applications that are already being transformed by deep learning and ticked off several others with great promise. One gets the sense from Berman that if we could tune out the hype and let the serious work proceed, AI (or whatever we end up calling it) may take hold faster than we anticipate.

As noted in Part One of this two-article set (HPC in Life Sciences Part 1: CPU Choices, Rise of Data Lakes, Networking Challenges, and More) this year’s discussion also included Chris Dagdigian, one of BioTeam’s founders and senior director of infrastructure, and Aaron Gardner, director of technology. Part one focused on core compute issues writ large. Part Two, presented here, examines the state of AI in life sciences broadly as well as trends in using the cloud.

HPCwire: Let’s jump into AI, warts and all.

Ari Berman, BioTeam

Ari Berman: In my opinion it is still so overhyped that it is unbelievable. I can’t browse Google, see my FB thread, or twitter or most periodical publications without seeing the word AI in the title somewhere. Every vendor is selling AI. I think it has become the gluten-free tag of life sciences because it is everywhere. Putting gluten free on bacon packages; why would you do that of course bacon is gluten free. I sort of see it like that. People have used it as a marketing gimmick to catch your eye. So, we would caution people to think that it’s the magic bullet for Big Data, because it isn’t.

That being said, as we said last year, that amount of focus has had the benefit of causing a whole lot more interest in AI and has driven some research in AI forward that wouldn’t have happened otherwise. People still come to us and everyone is talking about AI. “We need to do it. Can you help us?” We say why? And they say ‘Well I don’t know.”

I would say deep learning in general, which is really what we’re talking about when you say AI and machine learning, mostly refers to recursive or convolutional neural networks. String a few of these networks together to feed forward to each other for a good inference model. There are areas where deep learning is being used effectively and in production in life science and everything else is sort of going from the theory phase to the research phase or late research phase.

The area where it is really a benefit is in image processing, and there are two specific places where it’s a big deal. One is in Cryo-EM (cryo-electron microscopy) and includes automated region of interest selection in images and the creation of 3d molecular models that come out of Cryo-EM imaging. [That’s] very much deep learning based. It works very well in production. Uses GPUs. The whole thing. But it’s encapsulated in a couple of professional software packages. People don’t even necessarily know they are using it.

The other place where it is used very successfully (and is revolutionary) is in the area of pathology diagnostics. Pathologists typically stare at biopsies or xrays on screens and highlight areas that they think are highly relevant. That gets very subjective over time. So, using deep learning to take old datasets that pathologists have used as a training set, and using them to create an inference model works really well because the images tend to be very well annotated. They have done wonders to reducing error rates and amount of time it takes deal with complex biopsy samples. Those are two of the really profound changes that have happened due to deep learning in the life sciences.

There are a few others worth mentioning. There’s a bunch of natural language processing that’s being used in a research phase. Microsoft is working on a project to develop a machine called Hanover (Project Hanover). The goal is to use natural language processing to memorize all the papers necessary for cancer and help predict which combinations of drugs are likely to be most effective for each patient. Also, medicine and prescription management to reduce multiple drug interactions is an area that’s very much emerging. Another area of research is the use of deep learning to create inference models for drug development by reducing the number of small molecules that could have activity with a drug target. Another area that is in research is applying deep learning to genomics to infer important variants of genomes that have disease causing potential. The areas where it’s specifically being used are autism, cancer, and vaccine development. Also, CDC is actually starting this in a really big way to come up with flu targets every year. Again it’s very early research being done on these things.

The last one that is really almost at a production stage is this robotic assisted surgery. Lasik surgery, for instance, has a huge deep learning and real-time inference model to be able to track your eye while it is shooting a laser at your lens. That’s very deep learning based. Yet another one is being used, at the research level, is to improve patient selection in clinical trials. There’s software called Deep6 AI which basically uses deep learning to read doctors’ notes, pathology reports, diagnoses, recommendation, and lifestyle data to try to match patients with clinical trials criteria. That would go a long way towards making clinical trials work better.

Aaron Gardner: I would add a couple points. Overall you get what you incentivize, and doing AI-related research into deep learning is so incentivized for researchers seeking grants as well as getting time on HPC systems right now that everyone is trying to shift their problems to incorporate “AI” solutions. That trend hasn’t changed in the last year and it takes several years to work through such a hype cycle.

Nvidia DGX-2

I would note a trend we’ve seen continue is the development of supporting platforms and infrastructure to do deep learning.  All the major cloud providers, whether it be through providing GPUs or creating APIs for imaging or NLP, continue to expand their AI-related capabilities. You look at Nvidia with their DGX offerings and GPU-Accelerated Cloud (NGC). We are also seeing every storage vendor wanting to figure out and push their target architecture for supporting deep learning. It really points to the gap that we spoke to—the gap between real solutions and solutions in search of problems. As vendors are doing this the next question we get asked a lot is “so what can we do besides MLperf or the Deep500 to actually prove our [concept] out”. The pace of actual real world applications of ML/DL moving into something production scale and useable across life sciences is going to take a while. We probably will be having a similar conversation even next year about AI.

Chris Dagdigian: I can be super quick from the enterprise perspective. I think it is one of those technologies that has the potential to be truly transformative but it is also currently in the ridiculous hype phase. Even the vendor pitches are just flat out bogus. While we see it’s transformative, people understand the transformative nature, they’re still very much in the tire kicking phase, and oddly enough this is one of the cloud adoption drivers because you don’t just need a GPU to do ML or AI, you are probably using one of the libraries or development kits from Google or Amazon at the same time. It [DL/ML] is experimental but it is recognized as being real, still wildly overhyped and oddly enough it is a cloud driver because the cloud companies are actually putting pretty good SDK around both training and DL.

HPCwire: On that note, let’s turn to the cloud. Ari has said that today, basically everyone uses the cloud although how much and for what varies.

Chris Dagdigian: My commercial bent here is in HPC only. If you’re main concern is economics or cost, the numbers still work out that for 24×7 compute load, particularly with lots of storage, financially you are still better off not using the cloud for HPC. The big driver is the month over month cost of storage. So storage at high speed at $1M one-time expense on premise becomes a $100K/month bill on cloud that goes on forever. The reason people are going into the cloud has less to do with cost and more to do with gaining new capabilities. Cost optimization is a secondary focus.

The number one capability is use of the cloud as a pressure release valve when your on premise IT can’t keep up with how fast the instruments are changing. If you need that new node type or you need that new GPU or you quickly need a machine that has 4 terabytes of RAM, you can get that in the cloud. The second big thing is collaboration and in partnerships. The future of pharmaceutical drug development is increasingly involving really complex partnerships with companies that might be fierce competitors in one area and friendly partners in another. The cloud is a neutral territory for collaboration there. A third driver is drug development programs being run by consortia where no one really wants to open their corporate firewalls to someone who could be a competitor in another space. The other big driver for the cloud is there are petabytes of freely available really valuable open access data and it is also much easier to exchange data with service providers if you are on the cloud.

Ari Berman: There are some other considerations too. For example, capital investments become a consideration when organizations go to the cloud versus building their own systems. If you are a small business, going to a co-lo and paying a monthly fee and building out your own infrastructure can take too long to get rolling. You can spin up an infrastructure really fast in the cloud and start working quickly while you figure out your on premise stuff.

Cloud has made interesting gains in life sciences in general in the last year and in the federal space specifically. That’s interesting from a lot of perspectives because the federal space has been really resistant to the cloud because of dogmatic beliefs [about its] lack of security. That totally changed with the very public announcement of STRIDES from NIH (Science and Technology Research Infrastructure for Discovery, Experimentation, and Sustainability) – NIH is working directly with cloud service providers to develop sort a give-and-take relationship with them. They get lower costs to the government for reusability of those public data sets in the cloud. This interesting change in cloud focus is really being driven forward by the model of data commons, which has been around for a while but really caught on like wildfire in 2018.

The idea of a Data Commons is that you take these very large publicly funded datasets and put them in a common environment in a cloud and use that to dynamically associate compute, workflow algorithms, and collaboration all in one space. This way, you don’t have to buy multi-PB storage systems and a 100Gb internet link to download 2PB datasets to work with for your research. You don’t move the data at all, you work with it in place in the cloud. This is really in the early stages of coming out. The cancer industry is furthest along. It’s been a really interesting shift away from non-centralized computation on public datasets.

I would also say that if you are looking for commodity services, you are not going to find a lot of differences between the clouds. Amazon, and Azure and Google may edge each out here and there in pricing schemes and such, but the advantages are in service offerings that are specific to their cloud. [Offerings such as] compute-specific type offerings, like GPUs and TPUs, and Google’s query engine, which is one of the fastest you can use, and similar types of services. I think there’s another aspect to consider here, which is that the array of cloud offerings is mindboggling and complex and really hard to navigate for the average person. Which do you use? Amazon releases about 30 new services a month and they don’t turn off the old ones necessarily. How do you know what the difference is between GCP and Amazon? I think on some levels, simplicity wins, but looking at the distinct capabilities that are needed and deciding which cloud provider does that thing better will help make the decision a lot easier.

Aaron Gardner: I’ll only add a couple of things. I would note Amazon has made some interesting changes in terms of offering services like ParallelCluster, and has spun out a filesystem service (FSx) to provide Lustre to underpin it. This gives Amazon Web Services official parallel HPC file system support that wasn’t there before from AWS. They have also focused on looking at how their networking needs to improve for HPC and released Elastic Fabric Adapter (EFA) to help with MPI workloads for example. Cloud providers have traditionally been really bad at a lot of traditional HPC needs, they are now starting to address and create services around that need.

Also, we are now in our second decade of cloud adoption and are starting to see cloud providers get more competitive in terms of cost for their services. We are starting to see those costs influence buying decisions, that who customers choose can be influenced by the economics of a particular deal they were able to strike. I think that is a natural maturation of the cloud. The last thing is the idea of hybrid cloud that has been hanging in the air pretty much from the beginning of cloud as we know it. Figuring out how to leverage resources well between on premise, a cloud provider, or a multi cloud environment—it’s a lot easier said than done. We’re starting to see software development catch up, with next gen file systems, storage systems, and resource managers as well as other key componentry that go into building a hybrid environment starting to become available.

HPCwire: Looping back to the cost question for a moment. If storage costs are such a barrier to adoption, why haven’t cloud providers figured out how to lowers costs?

Ari Berman: [The answer, generally,] is that data intensive science is not one of their biggest customers. In the realm of customers they are trying to serve, which includes all of compute needs in the world, changing their pricing schemes to help life sciences which is generating petabytes of data a month, is marginally interesting but it is not a big problem for them. I think the NIH negotiation with cloud had some significant arrangements around storage costs to exactly take care of that. There’s a possibility that in the future, whether it will be brokered by federal agencies or grants or something, that there will be some relief.

Aaron Gardner: Cloud storage, when compared with enterprise storage approaches that are still too often employed for storing life science data, those costs are not so far apart anymore. I do not want to vilify the cloud in terms of cost. The cloud just exemplifies that total package price that delivering storage as a service requires.  In life sciences we typically slice and dice and chop all that stuff down to see how close can we get to just the cost of the storage media and then ask “can I make it even cheaper than that?” What will shift things is advancements in storage technologies themselves. Affordable HAMR drives, NAND prices continuing to drop, better software algorithms for compression and things like that. What will help costs is those innovations being pieced together by storage companies and software vendors and then those technologies being adopted by the cloud providers.

Subscribe to HPCwire's Weekly Update!

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

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion XL — were added to the benchmark suite as MLPerf continues Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing power it brings to artificial intelligence.  Nvidia's DGX Read more…

Call for Participation in Workshop on Potential NSF CISE Quantum Initiative

March 26, 2024

Editor’s Note: Next month there will be a workshop to discuss what a quantum initiative led by NSF’s Computer, Information Science and Engineering (CISE) directorate could entail. The details are posted below in a Ca Read more…

Waseda U. Researchers Reports New Quantum Algorithm for Speeding Optimization

March 25, 2024

Optimization problems cover a wide range of applications and are often cited as good candidates for quantum computing. However, the execution time for constrained combinatorial optimization applications on quantum device Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at the network layer threatens to make bigger and brawnier pro Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HBM3E memory as well as the the ability to train 1 trillion pa Read more…

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HB Read more…

Nvidia Looks to Accelerate GenAI Adoption with NIM

March 19, 2024

Today at the GPU Technology Conference, Nvidia launched a new offering aimed at helping customers quickly deploy their generative AI applications in a secure, s Read more…

The Generative AI Future Is Now, Nvidia’s Huang Says

March 19, 2024

We are in the early days of a transformative shift in how business gets done thanks to the advent of generative AI, according to Nvidia CEO and cofounder Jensen Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. Wh Read more…

Alibaba Shuts Down its Quantum Computing Effort

November 30, 2023

In case you missed it, China’s e-commerce giant Alibaba has shut down its quantum computing research effort. It’s not entirely clear what drove the change. Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Shutterstock 1285747942

AMD’s Horsepower-packed MI300X GPU Beats Nvidia’s Upcoming H200

December 7, 2023

AMD and Nvidia are locked in an AI performance battle – much like the gaming GPU performance clash the companies have waged for decades. AMD has claimed it Read more…

DoD Takes a Long View of Quantum Computing

December 19, 2023

Given the large sums tied to expensive weapon systems – think $100-million-plus per F-35 fighter – it’s easy to forget the U.S. Department of Defense is a Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Intel’s Server and PC Chip Development Will Blur After 2025

January 15, 2024

Intel's dealing with much more than chip rivals breathing down its neck; it is simultaneously integrating a bevy of new technologies such as chiplets, artificia Read more…

Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move

January 5, 2024

Reuters reported this week that Baidu, China’s giant e-commerce and services provider, is exiting the quantum computing development arena. Reuters reported � Read more…

Leading Solution Providers

Contributors

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Shutterstock 1179408610

Google Addresses the Mysteries of Its Hypercomputer 

December 28, 2023

When Google launched its Hypercomputer earlier this month (December 2023), the first reaction was, "Say what?" It turns out that the Hypercomputer is Google's t Read more…

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

Shutterstock 1606064203

Meta’s Zuckerberg Puts Its AI Future in the Hands of 600,000 GPUs

January 25, 2024

In under two minutes, Meta's CEO, Mark Zuckerberg, laid out the company's AI plans, which included a plan to build an artificial intelligence system with the eq Read more…

Google Introduces ‘Hypercomputer’ to Its AI Infrastructure

December 11, 2023

Google ran out of monikers to describe its new AI system released on December 7. Supercomputer perhaps wasn't an apt description, so it settled on Hypercomputer Read more…

China Is All In on a RISC-V Future

January 8, 2024

The state of RISC-V in China was discussed in a recent report released by the Jamestown Foundation, a Washington, D.C.-based think tank. The report, entitled "E Read more…

Intel Won’t Have a Xeon Max Chip with New Emerald Rapids CPU

December 14, 2023

As expected, Intel officially announced its 5th generation Xeon server chips codenamed Emerald Rapids at an event in New York City, where the focus was really o Read more…

IBM Quantum Summit: Two New QPUs, Upgraded Qiskit, 10-year Roadmap and More

December 4, 2023

IBM kicks off its annual Quantum Summit today and will announce a broad range of advances including its much-anticipated 1121-qubit Condor QPU, a smaller 133-qu Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire