Intel teases scalable ‘software-defined silcon’ • The Register


Intel teased a new technology it calls “Software Defined Silicon” (SDSi) but said next to nothing about it – and said The register it may not be worth anything.

SDSi appeared about three weeks ago in an article on the Linux Kernel mailing list, in which an Intel Linux software engineer named David Box described it as “a post-manufacturing mechanism to enable additional silicon functionality.” .

“Features are activated through a license activation process,” he wrote. “The SDSi driver provides one per socket, ioctl An interface that allows applications to perform three main provisioning functions. These provisioning functions are:

  1. Provision Authentication Key Certificate (AKC) – a key written to internal NVRAM that is used to authenticate a capacity-specific activation payload.
  2. Provision a Capacity Activation Payload (CAP) – a token authenticated using the AKC and applied to the processor configuration to activate a new feature.
  3. Read the SDSi Status Certificate – containing the processor configuration status.

Box’s post also pointed to a GitHub page that includes the following explanation:

Between this GitHub mention and the three functions added to the Linux kernel, it seems clear that Intel could ship Xeons with latent features that you could enable by sending money to them.

Intel offered a few other valuable details. The GitHub page includes a document explaining how to use SDSi-equipped silicon to enable dormant features, but without details of new features that could be enabled with this technology.

The register asked Intel to explain their posting to the Linux kernel mailing list. Chipzilla offered us the following non-binding response:

Yes indeed. Intel went to great lengths to find a way to obtain highly configurable Xeons licenses, but didn’t decide if it would become a product, and threw the technology into the Linux kernel anyway.

If you believe this, The register has a bridge that we would like to sell to you.

So let’s think about what Intel might be here – starting with why Intel wants to allow processor functionality.

Today, Intel is selling a processor and more often than not seeing any more money from its customers until their next purchase, which could take years. Licensing processor features would potentially give Intel more revenue, more often, perhaps even allowing it to create the kind of subscription services that investors love because they increase revenue and make its arrival more predictable. .

Intel is going to need predictable cash flow to fund its plans to spend tens of billions on new factories.

These factories are notoriously complex and Intel works hard on them, in part because they make so many variations of their products. If Intel could create fewer variations and instead bundle all of its technology into a smaller number of SKUs that can be reconfigured in software, the production savings could be substantial. Customers would still pay extra for a high-end kit, which would be software-enabled rather than created as discrete products.

We also know that Intel plans to make its products even more complex, with the “Alder Lake” architecture which mixes and associates different types of cores on the same die.

Configurable processors could delight customers, allowing them to purchase a processor with advanced capabilities such as Intel’s AVX-512 Extensions – a technology designed to accelerate machine learning – but pay to use these expansions only by when needed, rather than paying an upfront cost. Or buyers could acquire servers knowing they have additional overhead to activate as their needs increase.

This kind of flexibility is not overdone. In fact, Intel already offers something similar in its Speed ​​Select Technology (SST) – an offering that allows users to configure processors in configurations to suit different workloads. SST also allows you to define virtual CPUs with different characteristics than the physical CPU.

Another current option emphasizing composability and flexibility comes from HPE, which offers silicon on demand that allows customers in its GreenLake ITaaS environment to vary the number of active cores on Intel servers and pay only for those cores, rather than having to rent an entire server.

Also, don’t forget the mid-2010s craze for composable infrastructure – the idea that a set of connected components could be put together into servers that meet the demands of the day. This kind of concept almost always takes a few years to go from clever idea to practical adoption.

Intel has ended its response to The register‘s about SDSi by stating, “We are constantly innovating to ensure we provide flexible solutions to meet the unique demands of our customers and partners and to be at the forefront of the industry in terms of product capabilities and features. . “

Which means absolutely nothing. Still, going to the trouble of inserting SDSi into the Linux kernel surely indicates that something substantial is brewing. And if that something is configurable and / or composable processors, Intel may have something more substantial to say. ®


Comments are closed.