Soul Machines Stokes Next-generation AI Engines
The developer of the human OSTM platform uses high-performance, memory, and virtualization capabilities of the HP Z8 Workstation to advance brand experiences
As Corporations Seek to Connect With Customers in Meaningful Ways...
Share on
Soul Machines forges a new frontier in personalized service with its Human OS platform. The startup aims to help democratize the service industry by creating Digital People™, that run on AI-based simulation, providing seamless interactions with end users. To train machine learning and AI models, Soul Machines requires powerful machines with fast, efficient graphics capabilities that are not constrained by memory and compute capacity. The company used the HP Z8 G4 workstation with Nvidia Rapids software suite to set the course for large models such as an in-house version of Facebook’s BlenderBot, a machine-learning chatbot on one workstation—key steps toward revolutionizing the customer experience.
Industry
Technology
Objective
Run large, memory-intensive artificial intelligence training models and virtualizations on the HP Z8 G4
Approach
Train complex machine learning models such as an
in-house version of Facebook’s BlenderBot with 9 billion parameters
The goodness of human and machine collaboration
Soul Machines has pioneered the creation of Digital People™, combining hyper-realistic computer-generated imagery and autonomous animation resulting in human-like interactions for powerful brand experiences across industries worldwide. “Soul Machines’ ambition is to create Digital People to help democratize the service industry so we can help with education, medicine and other fields where a personalized service is a key difference,” says Shane Blackett, Vice President of Software and Technology at Soul Machines.
Based in San Francisco with research and development operations in Auckland, New Zealand, Soul Machines works with visionary clients who want their end user to have a delightful customer experience.
“Having a computer that could just facilitate it with out having to go to that extra level of engineering was critical. Otherwise, we probably just wouldn't have done it”
Shane Blackett
Vice President of Software and Technology,
Soul Machines
“We want to go further than that to where the digital person and the user and the content on the webpage, for example, co-exist in a three-way interaction where the digital person might suggest something, the user might do something or they might work together,” Blackett says. “We’re trying to provide a scalable platform for creating those Digital People and we’re running an AI-based simulation to provide the fidelity and realism.”
Soul Machines has developed a Human OS Platform, featuring a patented Digital Brain which helps deliver human and machine collaboration. The company works on a variety of models such as AI, machine and associative learning, and other complex solutions. Soul Machines’ customers cross industry sectors, from financial institutions to consumer goods. Case in point: a digital person for a leading global CPG in Japan sells cosmetics online in Japanese and English.
“In terms of the machines we have available, the Z8 is probably the most powerful single machine.”
Aiireza Nejati
Senior R&D Software Engineer, Soul Machines
Under the digital persona run models taught to recognize cognitive processes from learning and sensing to behavior. To train and run various AI-based and machine learning models and render these images, however, requires tremendous compute power, video memory, and graphics. Enter the HP Z8 Workstation.
“We have lots of training models that are memory constrained,” notes Alireza Nejati, Senior R&D Software Engineer at Soul Machines, whose team members work on the HP Z8. “The big motivating factor for us was the amount of Graphics (GPU), memory available on a single machine.”
IT Matters
• Harness compute power of the HP Z8
• Utilize large GPU memory capacity
• Train significant machine learning and AI models on a single machine
• Use cluster management software to assign tasks
• Expand virtualization competencies
Business Matters
• Explore new, available technologies to aid AI development
• Realize speed and efficiency
• Reduce need for manual intervention
• Make networks learn in real time
• Leverage a high-performing workstation for future projects
About Soul
Machines
Soul Machines is focused on humanizing artificial intelligence to create Digital People™. Its Human OS™ platform leverages the capabilities of human and machine collaboration in a responsive, relatable, and unprecedented way. Soul Machines is the only company with a patented Digital Brain which results in lifelike, interactive Digital People. It works with some of the leading brands in the world to change the customer experience through AI.
Wielding memory and power
When Soul Machines made a decision to use the HP Z8 in early 2020, New Zealand had implemented its COVID-19 lockdown. The company moved most of its IT operations off premises and Nejati’s team had to wait, and finally was able use the workstation in June of that year.
“In terms of the machines we have available, the Z8 is probably the most powerful single machine,” Nejati says. “It allowed us to run a lot of our models which wouldn’t be possible with just any machine because it takes a lot of memory.”
He points to the Z8’s configurability, Dual Xeon Gold processors, up to 3TB DDR5 ECC memory, 4 NVMe M.2 slots for high-speed data storage and up to 96 GB of GPU memory using two RTX 8000 graphic cards . The Z8’s Intel® Optane™ DC Persistent Memory is also a benefit when it comes to working with very large, complex data sets.
Initially, Soul Machines deployed the Z8 as a server, running it manually. The company then turned to the Nvidia Rapids suite of software libraries which was bundled with the z8 to manage resources more efficiently. The largest model Soul Machines had previously trained on the HP Z8 was an in-house version of Facebook’s BlenderBot, an open domain chatbot trained in online dialog from multiple sources with curated datasets. While BlenderBot isn’t as large as some other machine-learning models, it has 9 billion parameters.
“It’s quite challenging to train it all on one machine and you really need all 96 gigabytes of video memory to train that model,” Nejati says. “If it was 50 gigabytes, you won’t be able to do it.”
Previously, when models were smaller, he says it was possible to fit a model on a single GPU for training.
“Modern models are so big that even just one instance of the model needs to be spread across multiple GPUs, and this is infinitely easier when all the GPUs are on the same machine,” Nejati says. “When the GPUs are on different machines, it becomes another software engineering challenge altogether. You need to get it to properly distribute the whole model across multiple GPUs because you’re not running it in parallel, it’s running sequentially.”
Customer At a Glance
Application
Training machine-learning and AI-based models to render a personalized customer experience
Hardware
HP Z8 G4 Workstation
The ability to do more
The HP Z8 also gave Soul Machines the opportunity to use resources more effectively. The BlenderBot project, for example, was an effort to learn how to push that technology in conjunction with Soul Machine’s software. The power of the HP Z8 enabled a small company like Soul Machines to get it done.
“Having a computer that could just facilitate it without having to go to that extra level of engineering was critical,” Blackett says. “Otherwise, we probably wouldn’t have done it.”
Aside from training models, Soul Machines has used the HP Z8 to do more with virtualization.
“One big challenge in developing these machine-learning models is that you’ve got to be able to run other people’s code and build on it,” Nejati says.
Running open-source code developed on various machines natively can be arduous, with an environment ripe for errors. Virtualization, Nejati says, offers a big benefit.
“It really took a more powerful computer like the Z8 to make the virtualization work well. There were certain things that I thought were really impossible and virtualization made them possible.”
Aiireza Nejati
Senior R&D Software Engineer, Soul Machines
“You can just set up a virtual machine on your computer with any specifications you want, and it will run that code,” he says. “When you’re running machine-learning models, you want all the speed you can get. The HP Z8 gives you a lot of power and it allows you to run virtual machines as if they were native machines. That’s actually saved us a lot of time.”
There were models that Soul Machines tried to run locally on other workstations, but they did not work without virtualization.
“It really took a more powerful computer like the Z8 to make the virtualization work well,” Nejati says. “There were certain things that I thought were really impossible and virtualization made them possible.”
Making rapid and real-time machine learning— where networks are trained to be more learnable with only a few training samples—feasible has become necessary to create a seamless user experience. Having a system that serves immediate requests quickly is a big plus. The Soul Machines team learned that sending complicated machine-learning tasks to the HP Z8 was much faster than doing those jobs locally.
“Building that system on the hardware we were previously using was just not possible,” Blackett says.
As the world moves toward more sophisticated intelligence and data, a workstation like the HP Z8 opens many doors for Soul Machines. Having the ability to build on models and investigate new possibilities is essential as the company grows.
“I’m sure there will be other challenges that we’ll come across and we’ll say we need some way to investigate, evaluate how that works with our business,” Blackett says. “And if we’re shut out of that because we just don’t have gigantic computing power, then that doesn’t help us as a startup. Knowing that we have tools like the Z8, to go into some of these really big models, really matters.”
Z by HP for Data Scientists & Analysts
Get rapid results from your most demanding datasets, train models and create visualizations with Z by HP data science laptop and desktop workstations.
Exceptional performance
with Intel® Xeon®
and Intel® Core™ i9 processors.
You May Also Like
• Case Study
How is NASA Monitoring The Surface of The Sun?
Through AI and data analysis, NASA’s scientist teams have reached a deeper understanding of solar flares and plasma flows.
• Case Study
Cyark Discovers Extreme Power in HP Z8 for Digital Archiving
Here’s why this Z by HP equipment set the stage for the heavy-duty data analysis a comprehensive record of historical sites requires.
• Case Study
Z by HP and Red Rocks Community College Enable Next-Generation Classrooms
Learn how RRCC instructors boost data literacy through innovative machine learning platforms for the development of their student body.
Meet the Products
Have a Question?
Contact Sales Support.
Follow Z by HP on Social Media
Monday - Friday
7:00am - 7:30pm (CST)
Enterprise Sales Support
1-866-625-0242
Small Business Sales Support
1-866-625-0761
Monday - Friday
7:00am - 7:00pm (CST)
Government Sales Support
Federal
1-800-727-5472
State and local
1-800-727-5472
Monday - Friday
7:00am - 7:00pm (CST)
Education Sales Support
K-12 Education
1-800-727-5472
Higher Education
1-800-727-5472
Monday - Sunday
9:00am - 11:00pm (CST)
Chat with a Z by HP Live Expert
Click on the Chat to Start
Need Support for Your Z Workstation?
Disclaimers
4AA7-9918ENW, August 2021
Product may differ from images depicted.
The information contained herein is subject to change without notice. The only warranties for HP products and services are set forth in the express warranty statements accompanying such products and services. Nothing herein should be construed as constituting an additional warranty. HP shall not be liable for technical or editorial errors or omissions contained herein.
Intel, the Intel logo, Core and Xeon are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States and other countries. Microsoft and Windows are either registered trademarks or trademarks of Microsoft Corporation in the United States and/or other countries. NVIDIA and Quadro are trademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and other countries. Linux® is the registered trademark of Linus Torvalds in the U.S. and other countries. AMD is a trademark of Advanced Micro Devices, Inc.