Intel teases upgradable ‘software-defined silcon’ • The Register
Intel is annoyed by a new technology called “Software-defined silicon” (SDSi), but says almost nothing about it – and said The register it cannot be equal to anything.
SDSi appeared about three weeks ago in a post to the Linux kernel mailing list, in which an Intel software engineer named David Box describes it as a “post-production mechanism for activating additional silicone features.”
“Features are activated through a license activation process,” he wrote. “The SDSi driver provides a socket,
ioctl an interface for applications to perform three main delivery functions. “These delivery features are:
- Provision of an Authentication Key Certificate (AKC) – a key stored in the internal NVRAM that is used to authenticate a capability-specific activation payload.
- Capability Activation Payload (CAP) – A token certified by AKC and attached to the CPU configuration to activate a new feature.
- Read the SDSi state certificate – containing the state of the processor configuration.
The Box post also pointed to a GitHub page which includes the following explanation:
Between this mention of GitHub and the three features added to the Linux kernel, it seems clear that Intel can supply Xeons with latent features that you can activate by sending it some money.
Intel offered several other valuable details. The GitHub page includes a document describes in detail how to use silicon equipped with SDSi to activate sleep functions, but without details on what new functions can be activated with this technology.
The register asked Intel to explain its publication in the Linux kernel mailing list. Chipzilla offered us the following unrelated answer:
Yes correct. Intel has tried to come up with a way to license highly configurable Xeons, but has not decided whether to become a product and throw the technology into the Linux kernel anyway.
If you believe that, The register there is a bridge we would like to sell you.
So let’s think about what Intel can do here – starting with why Intel wants to license CPU features.
Today, Intel sells a processor and how often it doesn’t see more cash from its customers until their next purchase – which could happen years from now. Licensing CPU features would potentially give Intel more revenue, more often, perhaps even allow it to create the kind of subscription services that investors adore because it increases revenue – and will make its arrival more predictable.
Intel will need predictable cash flow to fund its spending plans tens of billions of new factories.
These factories are outrageously complex, and Intel is working hard – in part because it makes many variations of its products. If Intel could make fewer options and instead pack all of its technology into fewer SKUs that could be reconfigured into software, the production savings could be significant. Customers will still pay a premium for a high-end kit that will be included by software rather than created as separate products.
We also know that Intel plans to make its products even more sophisticated, with “Elhovo Lake” architecture that mixes and combines cores of different types on the same matrix.
Configured processors could delight customers by allowing them to purchase an advanced processor such as Intel AVX-512 technology extensions aimed at accelerating machine learning — but pay for the use of these extensions only when needed, instead of wearing advance price. Or buyers could acquire servers knowing they have some additional overhead to include as their needs increase.
This kind of flexibility is not excessive. In fact, Intel already offers something similar in its Speed selection technology (SST) – a proposal that allows users to set processors in configurations tailored to different workloads. SST also allows the definition of virtual processors with characteristics that differ from the physical processor.
Another current option, emphasizing composability and flexibility, comes from HPE, which offers silicon on request which allows customers in its GreenLake ITaaS environment to change the number of cores that are active on Intel-based servers and pay only for those cores instead of having to rent an entire server.
Remember also that mid-2010 is a fad for compostable infrastructure – the idea that a collection of connected components can be assembled into servers that meet today’s requirements. Such a concept almost always takes several years to move from a fine idea to a practical perception.
Intel terminated its response to The registerAsks questions about SDSi, stating: “We are constantly innovating to ensure that we enable flexible solutions that meet the unique requirements of our customers and partners and guide the industry in product capabilities and functions.”
Which means absolutely nothing. Still, trying to insert SDSi into the Linux kernel certainly signals something significant. And if that thing is configurable and / or composite processors, Intel may have something more substantial to say. ®