Presented by Solidigma
As artificial intelligence grows in popularity, data centers are facing a critical storage bottleneck, and at its center is established strenuous drives. Data that once sat idle in frosty archives is now often used to create more right models and provide better inference results. This transition from frosty data to toasty data requires low-latency, high-bandwidth storage that can support parallel computing. Demanding drives will remain the primary tool for low-cost frosty storage, but without rethinking their role, the high-capacity storage tier risks becoming the weakest link in the AI factory.
“Modern AI workloads combined with data center constraints have created new challenges for hard drives,” says Jeff Janukowicz, vice president of research at IDC. “While hard drive vendors are trying to increase storage capacity by offering larger drives, this often comes at the cost of lower performance. As a result, the concept of ‘Nearline SSDs’ is becoming an increasingly important topic of discussion in the industry.”
Today, AI operators must maximize GPU utilization, efficiently manage network-attached storage, and scale compute—all while lowering costs as power and space become increasingly insufficient. In an environment where every watt and every square inch counts, says Roger Corell, senior director of artificial intelligence and leadership marketing at Solidigma, success requires more than just a technical refresh. This requires deeper customization.
“This speaks to a tectonic shift in the value of data for AI,” Corell says. “This is where high-capacity SSDs come into play. In addition to capacity, they provide performance and efficiency – enabling exabyte-scale storage pipelines to keep pace with the relentless pace of increasing data set sizes. All of this consumes energy and space, so we need to do it as efficiently as possible to enable greater GPU scaling in this constrained environment.”
High-capacity SSDs don’t just displace strenuous drives – they remove one of the biggest bottlenecks on the AI factory floor. Providing massive increases in performance, efficiency, and density, SSDs free up the power and space needed to further scale your GPU. This is not so much a storage modernization as a structural change in the design of data infrastructure for the era of artificial intelligence.
Demanding drives vs. SDDs: More than just a hardware refresh
Demanding drives have an impressive mechanical design, but they are composed of many moving parts that consume more power, take up more space, and fail more quickly than solid-state drives on a immense scale. Reliance on spinning platters and mechanical read/write heads inherently limits input/output operations per second (IOPS), creating bottlenecks for AI workloads that require low latency, high concurrency, and consistent throughput.
Demanding drives can also handle latency-sensitive tasks because the physical act of searching for data introduces mechanical delays that are unsuitable for real-time AI inference and training. Moreover, their power and cooling requirements enhance significantly with repeated and intensive data access, reducing performance as data scales and heats up.
Meanwhile, VAST’s SSD-based storage solution reduces power consumption by ~$1 million per year, and in an AI environment where every watt counts, this is a huge advantage for SSDs. To demonstrate this, Solidigm and VAST Data conducted a study examining the economics of data storage at the exabyte – quadrillion byte, or billion gigabyte, scale, along with analyzing the energy consumption of storage compared to strenuous drives over a 10-year period.
As a starting point of reference, you’ll need four 30TB strenuous drives to get the capacity of a single 122TB Solidigm SSD. After taking into account VAST data reduction techniques, which are made possible by the excellent performance of SSDs, the exabyte solution includes 3,738 Solidigm SSDs compared to over 40,000 high-capacity HDDs. The study showed that SSD-based VAST consumes 77% less energy during storage.
Minimizing the data center space
“We ship 122 terabyte drives to the top OEMs and leading AI cloud service providers in the world,” says Corell. “When you compare a 122TB SSD configuration to a hybrid HDD + TLC SSD configuration, you get a nine-to-one savings in data center space. And yes, this is important in these huge data centers that build their own nuclear reactors and sign large power purchase agreements with renewable energy providers, but it becomes more and more important as we get to regional data centers, on-premises data centers and all the way to edge deployments where space can be at a premium.”
The nine-to-one savings go beyond space and power – they enable organizations to fit infrastructure into previously inaccessible spaces, scale up GPUs, or build smaller footprints.
“If you get X amount of land and Y amount of power, you’re going to employ it. You’re an AI,” Corell explains, “where every watt and square inch counts, so why not employ it in the most productive way possible? Get the most productive storage in the world and enable greater GPU scalability within the envelope you need to fit. On an ongoing basis, this will also save on operational costs. You have 90 percent fewer storage pockets to maintain and associated costs are gone.”
Another element that is often overlooked is the (much) larger physical size of data stored on mechanical strenuous drives, which results in greater consumption of building materials. Combined, concrete and steel production account for more than 15% of global greenhouse gas emissions. By reducing the physical footprint of storage, high-capacity SSDs can support reduce greenhouse gas emissions from concrete and steel by more than 80% compared to strenuous drives. However, in the last phase of the sustainability lifecycle, i.e. end of life of the drive, there will be 90% fewer drives to be disposed of. .
Changing the refrigerated and archival storage strategy
Switching to SDD is not just about upgrading memory; this is a fundamental shift in data infrastructure strategy in the AI era that is gaining momentum.
“Large hyperscale vendors want to squeeze the most out of their existing infrastructure by doing unnatural things, if you will, with hard drives, such as overproviding them to almost 90% to squeeze as much IOPS per terabyte as possible, but they’re starting to show up,” Corell says. “As they turn to modern high-capacity storage infrastructure, the entire industry will follow this trajectory. Additionally, we are starting to see that the lessons learned about the value of modern storage in AI also apply to other segments such as big data analytics, HPC and many others.”
Although all-flash solutions are almost universally used, there will always be room for strenuous drives, he adds. Demanding drives will continue to be used in applications such as archiving, frosty storage, and in scenarios where the cost per gigabyte alone outweighs the need for real-time access. However, as the token economy heats up and enterprises realize the value of data monetization, the toasty and warming data segments will continue to grow.
Solving future energy problems
Now in its fourth generation, with over 122 exabytes shipped, Solidigm’s Quad-Level Cell (QLC) technology leads the industry in balancing higher drive capacities with cost efficiency.
“We’re not thinking about storage just as storing bits and bytes. We’re looking at how we can develop these amazing drives that can deliver solution-level benefits,” says Corell. “The shining star is our recently launched E1.S, designed specifically for dense and efficient storage in direct-attached storage configurations for the next-generation fanless GPU server.”
Solidigm D7-PS1010 E1.S is a breakthrough, the industry’s first eSSD with single-sided direct liquid cooling technology. Solidigm worked with NVIDIA to address the dual challenges of thermal management and cost efficiency while ensuring the high performance required for demanding AI workloads.
“We are rapidly moving towards an environment where all critical IT components are directly on-chip, liquid-cooled,” he says. “I think the market needs to take a look at their approach to cooling because power constraints and power challenges are not going to go away, at least in my lifetime. They need to apply a neo-cloud mindset when designing the most efficient infrastructure.”
Increasingly intricate inference is pushing against the storage wall, making storage architecture a major design challenge rather than an afterthought. High-capacity SSDs combined with liquid cooling and productive design are becoming the only way to meet the growing demands of artificial intelligence. The task now is to build infrastructure not only for performance, but also for storage that can scale effectively as the amount of data grows. Organizations that adapt storage now will be able to scale AI tomorrow.
Sponsored articles are content created by a company that pays to publish or has a business relationship with VentureBeat and is always clearly marked. For more information, please contact us sales@venturebeat.com.