NVMe is the silent IT revolution – what is it, and why you should adopt it?

#KingstonCognate introduces Rafael Bloom

Rafael has spent his career within senior Technology Product, Marketing Communications and Business Development roles. His advisory practice focuses on the new organisational, product, and communications challenges of technological and regulatory changes, with a central aim of building and preserving revenue streams.

This highly diverse work involves subject matter expertise on information governance and compliance by design, data privacy, and emerging technologies such as AdTech, Mobile & 5G, Artificial Intelligence and Machine Learning.

Computing: Power and speed

I have addressed before how storage and memory are key to the new digital experiences of the 2020s. This boils down to the fact that you can have the fastest processor imaginable yet be unable to tap into that power because of bottlenecks in the device’s internal architecture. NVMe drives – especially the newest drives that take fullest advantage of the newest generation standard – simply blow everything else out of the water.

There are some simple reasons why NVMe is different that I would like to outline, but first let us establish what typical use cases for choosing NVMe might be:

At the edge, ‘Power Users’ of high-performance workstations who deal with the most demanding computational tasks, such as 4K video post-production or rendering 3D graphics with ray tracing. To get smooth performance and speedy renders while editing and processing video images, you need to move massive amounts of uncompressed data around. These processes can be held up by many factors – insufficient RAM for example – but also these computational tasks are becoming too much for a spinning hard drive to handle, and with the exponential growth in data volumes, even a traditional Solid-State Drive (SSD) will struggle.

Essentially, the performance bottleneck is not necessarily the fact that the device is Solid-State – after all, NVMe drives are themselves SSDs – but the limits of the interface that sits between the SSD and the CPU. This goes back to how computers evolved – drives were plugged in via a cable, with a gradual evolution of the capability of an interface, from the old RS-232 interface, through SCSI and latterly, the SATA standard. NVMe takes a different approach. It might even be described as nitrous oxide for the CPU.

The new generation of NVMe drives uses the PCIe 4.0 standard to read, write and move data around. This standard has data throughput bandwidth that is several times than that of even the fastest SATA drives. SATA tops out at about 6Gb per second – pretty good, but NVMe can achieve up to 32Gb per second. This is because of many factors but the easiest to understand is that NVMe simply has more ‘lanes’ through which to move data. NVMe is also designed to integrate better with the networks that underpin today’s computing infrastructure.

NVMe SSDs are designed to more of an integrated part of the fabric of an IT network. NVMe functions with more direct integration into the computer’s CPU. This avoids data bottlenecks that happen even with the fastest SATA SSDs when pushed to the limit. The form factor of client NVMe device is crucial as well - they are about the size of a stick of chewing gum, meaning they can fit in small devices like laptops. You might not know it, but even the Sony PlayStation 5 runs NVMe, meaning loading times on the console can drop from 40-60 seconds for the PS4 down to around 3 or 4 seconds.

Data centers under duress

Secondly, we need only look at the massive upsurge in the global demand on data centers. Computation in the cloud powers much of our daily lives. This was happening at pace anyway, but COVID-19 proved to be the catalyst for the digitisation and virtualisation of offices and the absolute rule of the mobile device. What is more, the kinds of computational tasks taking place in data centers are ever more complex and demanding.

Artificial Intelligence, for example, is a term that covers a wide range of computational tasks, and many of these rely on machine learning. What AI computational modes have in common is that they all leverage massive amounts of data to function. Here again, NVMe is the technology that makes a significant difference when it comes to getting things done swiftly and seamlessly.

For example, let us consider what is going on in the healthcare industry vertical, not only the fact that medicine is now a digital domain. AI-assisted diagnostics and predictive algorithms need to deal with the analysis of millions of records – and each individual record might be something like a very high-resolution digital image that is gigabytes in size. Pushing all this data around involves computation that is unlike any cognitive process that a human might recognise – this is artificial intelligence, after all – and this demands massive data bandwidth.

At the industrial scale of a data center, NVMe can do the job – silently, at low temperatures and taking up minimum space.

What am I looking for when considering an upgrade to NVMe?

Here is where it can be less than simple – some non-NVMe SSDs can plug into the same slot as an NVMe drive, but still not deliver NVMe performance. This is because the connector – known as an ‘M.2’ connector, for those who still require more technical terminology and acronyms – is backwards compatible with SATA SSDs that have the same ‘chewing gum’ form factor.