In the press kit for the Lunar Lake launch, Intel has stated clearly which are the benefits of removing SMT, while in the previous presentations the numbers were somewhat ambiguous.


According to Intel, if the comparison is made between two cores having the same IPC as Lion Cove, one with SMT and the other without SMT, the core without SMT has a smaller area by 10%.

Moreover, when the powers consumed by the two cores are compared, when running a single thread (i.e. at equal performance), the core without SMT consumes a power smaller by 15% (because the switched electrical capacitance is diminished in this ratio).


Obviously, when running a multithreaded application, if the SMT gain for that application happens to be over 15%, the performance per watt would be better for a core with SMT and a SMT performance gain over 10% would be enough to provide a better performance per area.

In another interview with an Intel SoC architect, he has corroborated the numbers that I have seen in my own tests, i.e. he has said that in older CPUs there were higher SMT performance gains, up to 30%, while in modern CPUs (i.e. with better OoOE) the SMT performance gains are smaller, up to about 20%.

Therefore for servers SMT still improves the multithreaded performance per area and per power.

On the other hand, for a low-power hybrid CPU, the single-thread performance per area and power is better without SMT, while the multithreaded performance can be improved much more than by adding SMT when, at equal area, one P-core is replaced by 3 E-cores.


I believe that the most complete version of the Intel Press Kit for the Lunar Lake launch is the one at:

https://wccftech.com/intel-core-ultra-200v-lunar-lake-launch-cpu-gpu-ai-leadership-over-qualcomm-amd-much-better-battery-life/

The numbers for SMT are in the appendix, at claim 22.