Expanding the World of Software Defined: Software Defined Silicon
Living in a software defined world, CPUs are next to experience the transition from hardware to software defined.
Intel has been developing its new Software Defined Silicon (SDSi) mechanism for the past few years and recently announced it will be officially supported in Linux 5.18, which is expected to be released this spring.
As a fan of developing other software defined things such as software defined radios, the idea of software defined CPUs peaked my curiousity. Software Defined Silicon can mean a few different things from the research I’ve done into different variations of it.
Sometimes it does entail more of an FPGA-like reprogrammable approach, however the most common approach currently seems to be putting certain features of the CPU behind a paywall that is unlocked via connection to a vender’s server when the appropriate licenses are purchased. The latter seems to be the case with Intel’s latest release.
Not everyone I saw was particularly keen on Intel’s new SDSi, pointing out that the company has unsuccessfully attempted to put hardware features behind a paywall in the past, forcing Intel to kill the project. In 2010, the Sandy Bridge family of processors had higher clock speeds and increased memory cache for the Intel Core i3-2xxx series paywalled.
However, I do find the original intent of this method from Intel to be a sound idea: allowing a customer to purchase the silicon and only pay for the features they need at that point in time, but have the option to grow into the given chip as needed. This means costs of board respins and hardware swaps can be mitigated by users.
Of course this is all assuming that this only-pay-for-what-you-need-now doesn’t turn into users having to pay monthly subscriptions to utilize their CPUs…
Regardless, the kernel design and device tree overlay configurations that will surely spawn from SDSi‘s as a concept will undoubtedly be interesting for us Linux enthusiasts to see come about.