Intel patent teases software-defined 'Super Cores' to boost single-threaded performance on CPUs

Intel has been suffering against AMD for many years now in the CPU business, but the tide is slowly turning... while a newly-filed patent from Intel teases software-defined Super Cores, which would see smaller cores working together as a bigger Super Core to boost single-threaded performance.

2

Normally, we'd see larger cores used in a CPU but they've got their limits, with a single huge core having diminishing returns, with Intel's new patent "EP4579444A1" showing how Intel wants to overcome these issues through an SDC aka "Software Defined Super Cores".

The idea behind SDC is that Intel would use more cores than a single large core, fusing them together -- virtually -- when required. For example, two smaller cores could work together instead of a larger core dividing the workload, which would boost single-threaded performance considerably.

It wouldn't be easy, as splitting the tasks you'd run through your operating system (Windows or Linux) across multiple cores, all while keeping the program order, isn't an easy task. Intel's new patent says that SDC would maintain the instructions in their correct order, while to the software being run, it would still appear as a single larger core, executing a single thread.

This almost sounds like the return of Hyper-Threading, but it's done quite differently, as the goal behind Intel's proposed SDC is to enhance single-threaded performance. It would significantly boost IPC (instructions per clock) all without boosting voltages or CPU frequencies.

Through dynamic fusion, whenever a heavier single-threaded task needs to be executed, Intel's proposed new CPU would create a "Super Core" to get those jobs done (much) faster. SDC would be splitting the instructions, firstly the load across multiple smaller cores, and then the cores work together to maintain the ordering.

Intel would then use things like the Shadow Store Buffer, where it can make sure correct data transfers are happening between the cores. There are some major roadblocks in between, but it's good to see Intel thinking outside of the box in order to better compete against AMD, which has been kicking ass, and will continue to kick ass in CPU technologies in the years to come.

Not only that, but it also reminds me of rumors from over a year ago now from leaker Moore's Law is Dead. In his rumors about Rentable Units, Intel would see P-Cores split into two smaller cores in its Royal Cove 1.1 implementation, while Royal Cove 2.0 was rumored to feature up to 4 threads per P-Core.

This was in a story about Intel developing its next-gen Cobra Cove architecture on x86, rumored as the potential successor to Royal Core.

In further rumors the month after -- also from MLID -- we were told that we could expect Nova Lake's successor to be Griffin Cove with a "Unified Core" that would eliminate E-Cores from the roadmap. MLID did note in those rumors that he's been warned that everything after Nova Lake is NOT finalized, and that there would "likely to have political battles fought over their final design choices".

Cobra Cove was reportedly cancelled when Beast Lake was cancelled, with Cobra Cove to be "Royal Core 2.0" but now RIP to Cobra Cove. These new patents lean a little into those rumors, which I thought was interesting to see.