Organizations that decide to make a major push into AI must ensure that the AI has enough data to make a difference. Without a well-designed data platform, that’s not possible. Everpure is working hard to incorporate as much intelligence and efficiency as possible into the data platform it offers to customers.
Today, just a few months after outlining its overarching vision for what it calls “Data Primacy,” Everpure is already rolling out a series of updates. That’s a positive sign in itself, as it shows the entire Data Primacy initiative isn’t just for show. Not that we would have expected anything less from Everpure, by the way. The company is generally very good at doing exactly what it says it will do.
For a complete overview of what Everpure aims to achieve with Data Primacy and how underlying components such as Data Intelligence and Data Stream contribute to this, we’d like to refer you to the article we wrote on the subject a few months ago. Below, we’ll discuss some of what we consider to be the most interesting updates and new features in the data platform that Everpure is announcing today.
Get started quickly
Some of today’s updates relate to how organizations can get started with Data Intelligence. Data Intelligence searches for and finds the data organizations need to make optimal use of AI. At least, that’s the promise. In addition, Data Intelligence classifies the data and provides the proper context.
If you’re thinking this is probably quite a task to set up, Everpure has good news. Starting today, it offers the ability to roll it out directly from the Pure1 management platform. No expensive or complicated professional services are required. According to Fred Lherault, Field CTO EMEA at Everpure, this approach can yield the first results from Data Intelligence in as little as two hours.
Of course, speed isn’t everything. The use of Data Intelligence must also be secure and controlled. Everpure has now built this capability in without compromising privacy. You gain insight into who has access to which file share and how up-to-date it is, without Data Intelligence ever reading the contents of the files.
More intelligence and better performance
A second category of updates that Everpure is adding to the data platform today brings more intelligence and better performance. A good example of the former is Purity DeepReduce. Lherault promises that this new feature offers, on average, an additional 2:1 reduction on top of the compression you’re already using on your data in your FlashBlade//S and FlashBlade//E systems.
DeepReduce uses similarity-based compression. This is much more flexible than deduplication, Lherault explains, because it analyzes the data at a much finer granularity. If two blocks or pages are 99 percent the same, they are not identical. However, DeepReduce recognizes that the vast majority of the data is indeed identical. There is no need to store a “copy” for this.
To enable GPUs to start processing organizations’ data as quickly as possible, Everpure has added a feature it calls PureKVA. This stands for Key-Value Accelerator. Key-values are among the core concepts of AI. They ensure that an AI model understands how different pieces of data belong together.
With the announcement of PureKVA, FlashBlade now loads this context directly into GPU memory. According to the company, this results in significant performance gains. The first tokens are processed 20 times faster than without PureKVA. A key advantage of loading context directly into GPU memory is that it prevents multiple GPUs from performing the same task.
Lower costs and faster compliance
The last two updates from today that we’d like to highlight concern tokenomics and compliance. In our view, Intelligent Token Optimization is the most important addition Everpure is making to its own platform today. It is a reference architecture designed to ensure that customers can easily scale and switch between models.
Intelligent Token Optimization is an architecture that Everpure first used internally. According to Lherault, this architecture reduced costs by 50–75 percent, which is no small feat. In addition, Intelligent Token Optimization also gives organizations more control. Among other things, it makes it easy to ensure that sensitive data is always processed by a local model. Organizations that want AI but also want to operate independently can, in principle, set this up using this technology.
Finally, there’s the Pure1 Config Advisor. It may not be the most exciting feature, but it’s important because it allows you to tighten up data policies. Over the years, the number of configurations with specific policies has only increased. This has become very confusing, particularly when it comes to compliance. With Pure1 Config Advisor, you can significantly reduce the number of configurations and policies. Lherault shows an example where 270 configurations were reduced to 11. He assures us this was done without compromising compliance. We imagine auditors will be pleased with this as well.
All in all, Everpure is keeping the pedal to the metal in developing its own Data Primacy vision. Many of today’s updates may not be earth-shattering on their own. However, when a large number of them are released at once, it really makes a difference. The platform as a whole becomes significantly more powerful with each update.