Area for HT/SMT components, future perspectives extended
Area for HT/SMT components, future perspectives extended
It's actually the opposite. Intel's Hyper Threading is based on SMT, which is the technical term. Marketing uses the name Hyper Threading. You can improve this by adding access checks, which might be as simple as a single bit indicating whether it's part A or B. For SMT 4 you need two bits, and when accessing something you verify if the bit fits your path; otherwise it's ignored. It just needs to be done properly. The implementation cost is low, but making it safe adds some expense. CMT isn't very effective either. So we stick with SMT. Do you have another way to boost CPU performance without increasing power usage? Everyone wants to know...
My focus wasn't on guaranteeing SMT, but on whether it was essential if core counts keep rising. In short: Core numbers will rise significantly, and most software may struggle to handle more threads. SMT adds extra overhead; if cores are plentiful, the benefits diminish beyond certain limits. Removing SMT could free up resources, making it easier to add more cores. Running a server with a million threads is doable with SMT, but I wasn't insisting on it. The goal was to enhance performance by potentially replacing SMT, which might involve shifting away from traditional silicon designs. That shift isn't likely to be cost-effective soon. I agree partially. I've often noted that for typical tasks, we're already near our limits. Real improvements will likely come from specialized hardware solutions. For instance, Ryzen excels in cryptography thanks to built-in support, while Intel is boosting performance via AVX extensions and FPU features.