Live the Dream of Unlimited Compute Power

 

This startup aims to speed up your deep learning initiatives, simply by optimizing compute.

 

We hear a lot about the power of AI to potentially change our lives, but also about its challenges - cost, difficulty getting to production, complexity for companies to deploy, the dearth of data scientists. Add to that the things that are critical for building an effective AI datacenter: fast storage for managing huge datasets, efficient networking, and state-of-the-art hardware accelerators. Bottom line: if your storage isn’t vast and your network isn’t fast, your innovation and progress is going to be severely limited.

Solution: easier access to -- and control of -- hardware accelerators such as graphics processing units (GPUs). Run:AI’s reason for being is to greatly increase compute resource access to data scientists, essentially creating unlimited compute. As researchers access more compute, utilization across the AI datacenter can be increased as much as three times, from 25 percent to more than 75 percent.

Ultimately, Run:AI offers a deep-learning orchestration platform that helps organizations manage the resource allocation of GPUs while increasing cluster utilization.

As a result, IT departments can gain full control over GPU utilization across nodes, clusters and sites, while data scientists can speed up initiatives by accessing compute when and how they need it.  

If your company is new to the AI space, as many are, ask yourself these questions:

• How am I managing my company’s compute power?
• How do IT teams allocate GPU resources? How do data scientists share these resources easily and efficiently?

Many companies currently keep track of GPU usage the old fashioned way -- Excel spreadsheets, for instance. In the meantime, when it comes to a GPU’s limited access, it’s often every data scientist team for themselves, calling dibs on a GPU server before anyone else can claim it. Not an easy, fast way to team up and innovate.

“At some point, data scientists will need massive computing power in order to build and train AI models,” says Run:AI CTO and co-founder Ronen Dar. “Getting access to GPUs can be difficult. The company will need to build infrastructure management and resource allocation for those data scientists, whether on-premises or in the cloud.”

 

Run:AI works to reduce blind spots due to this static allocation of GPUs. The Run:AI graphic user interface (GUI) helps speed productivity by providing a more holistic view of: 

• GPU infrastructure utilization
• Usage patterns
• Workload wait times
• Related costs

These revelations often come as a surprise to Run:AI clients.

“Many companies think that their GPUs are being fully utilized,” says, Run:AI marketing VP Fara Hain, “but they have no idea there is low utilization. So we help by e optimizing their GPUs because they are  a very important resource for the company.” 

Run:AI pulls it all together with two coexisting strategies: utilization and speed:

• Optimizing utilization of AI clusters by enabling flexible pooling and the sharing of resources between users and teams. Run:AI’s software distributes workloads in an “elastic” way: changing the number of resources allocated to a job as needed. The result: data science teams can run more experiments on the same hardware.
• Maximum speed to run data experiments and training AI models (better utilization helps make this happen). Abstract AI workloads from compute power and then apply distributed computing principles for faster results -- essentially allowing a guaranteed quota of GPUs for each project.

A recent funding round is allowing the Israel-based company to expand its global reach and hire more engineers. The goal, always, is to help data scientists manage and expand their workloads and help them train the models that lead to amazing breakthroughs.”

“If you open up compute power to very smart people, they will do incredible things,” Ronen says. “We’re just trying to prep the platform so that they can do that.”

Find out more about our content partner, Run:AI, and to apply for a free trial, click here.