Today, Solidigm announced the introduction of the world’s highest capacity PCIe solid-state drive (SSD), the 122TB (terabyte) Solidigm™ D5-P5336 data center SSD.
During the launch event, we attended a presentation on the new 122TB drives. However, the highlight for me was a panel featuring Solidum’s partners and customers. Throughout the event, I experienced a strong sense of déjà vu, that reminded me of my early days in the storage industry.
Origins of Data storage
If we want to go back to the very beginning, the history of data storage dates back to the 18th century with the invention of punch cards by Basile Bouchon in 1725 for use in textile looms.
But let’s skip ahead to the beginning of the hard drive era. IBM released the 350, also known as RAMAC (random access method of accounting and control), in 1956. For the first time, this machine made it possible for businesses to immediately access current business data.
The system weighed 1700 pounds and a capacity of 3.75MB. This video explains the engineering process.
Storage in the dotcom era
I joined EMC as a technical trainer right after I graduated college. They hired me because I knew vi. I helped mainframe SEs learn vi so they could attach UNIX servers to EMC storage arrays. Little did I know my career was taking off at the beginning of the end of the dotcom era.
When Dick Egan and Roger Marino founded EMC in 1979, and quickly pivoted from selling furniture to selling IBM . By1981 they were developing computer memory systems that could work with DEC and Wang Labs microcomputers. But their huge success came with EMC Symmetrix storage arrays.
The Symmetrix line was novel because the array could be attached to any host, no matter if it were a mainframe, a UNIX server, or a Windows server. The target on the array just showed up as a disk to the host. “A disk is a disk is a disk” is what we’d say to students.
The Symmetrix array provided extra capacity for mainframes but also provided storage capacity for agnostic server platforms. This was huge for the established businesses such as the airlines and the government but also for the up-and-coming online companies like Amazon, eBay, and Priceline.com that needed more capacity than what servers could provide at the time.
The Symmetrix 5500 family was the world’s first terabyte disk array, but that took 128 hard disk drives (HDDs).
Solidigm’s 122TB innovation impacts the future of AI
In 2024, we need a new type of architecture to host the data required to feed AI algorithms. For example, researchers trained the GPT-3 large language model (LLM) with 45TB of text data from different data sets. That’s 45 Symmetrix 3000 arrays, or 12,000,015 RAMAC units!
However, you could put 2.71sets of the GPT-3 data on one Solidigm 122TB hard drive, and it would fit in your hand.
That’s one reason the Solidigm announcement is so important. Artificial Intelligence is the workload of this new era. Training the AI models takes a lot of data, and the size requirements go up every year.
Currently, hard disk drives (HDDs) store most of that training data. Those disks take much more power to run than SSDs. They are also physically larger, so they take up much more space on the data center floor.
AI workloads need GPUs to get the work done quickly. Hard drives have to work harder than SSDs to support GPUs. They need lots of water to keep the systems cool.
Solidigm pioneered QLC (quad-level-cell) NAND technology. One benefit of QLC is that the drives takes much longer to wear out under typical workloads. In fact, these new 122TB drives can run 32KB or 4KB random writes 100% duty cycle 24/7 for five years and not wear out. That means they will not wear out while they are still under warranty. That’s a pretty bold claim!
The ecosystem is hungry for this innovation
At the event, Patrick Moorhead and Melody Brue interviewed a panel of Solidigm customers and partners. We heard from Arm, CoreWeave, Ocient, Peak AIO, and Vast. Each of these companies is actively building the data systems needed for the AI era, and they are all eager to put the new 122TB drives into their systems.
Wrapping up
We have entered the era of AI. We are beginning to see new technologies that can meet architectural requirements for this era. The new data platforms that will host AI workloads require small, dense, durable drives. What could you build with an array filled with 122TB SSDs?
There was a reason for my déjà vu, we are repeating a pattern. In the 1950s, the IBM RAMAC replaced the need to search through physical files for data. In the 1990s, EMC made storage reliably accessible to any host that fueled the dot com and internet era. And in our present day, we’re seeing the development of dense, durable SSD drives so that the data needed for AI can be available with the power and cooling that is currently available.
It’s exciting to watch developers create new technology to meet the needs of the AI era. It will also be interesting to see how the ecosystem will apply these storage innovations. It’s a pretty exciting time to be be part of the storage industry!
You can find out more about the Solidigm 122TB drives here or read the press release here.
You can read more industry product coverage by me here.
Disclaimer: Solidigm invited me to their launch event in NYC. They paid for my flight and hotel, and I attended a nice cocktail event and received a gift bag. However, they did not pay me for the content of this blog post or review it in advance of publication.
