This time around, there not much point in running around, as the results are rather self-explanatory...
Absolutely mental stuff.
Here's a link to individual fixed-frequency "IPC" results, done one the updated suite of 36 workloads:
Here's the average of those workloads:
![Image]()
ER standing for "Extremities Removed" (i.e. absolute single high and low-results). For CML-S the absolute low being "Buddhabrot Fractal" and the high being "WinRAR".
Meanwhile for VMR, the removed low was "GMPBench" and the high "Vampire Numbers".
All of the test systems adhered to their maximum, officially supported memory frequency during the testing. For AMD (both MTS and VMR) that meant 3200MHz with 14-14-14 timings and 2933MHz with 13-13-13 timings for the Intel 10900K system.
I've got a truckload of other results for different scenarios however, I'm not sure if they're even relevant, as the performance of the AMD Ryzen 5000-series is so utterly crushing.
The reason why the performance in both GMPBench, or rather in libGMP and OpenSSL are lower than expected is being looked into.
To give you a quick idea of scaling, these CPUs are mostly temperature limited, due to the tiny CCD size of around 82mm², hence the extreme intensity.
Overall, the scaling is excellent.
![Image]()
Here's what a typical core only power dissipation vs. temperature will look like, more or less regardless of the cooling.
![Image]()
Absolutely mental stuff.
Here's a link to individual fixed-frequency "IPC" results, done one the updated suite of 36 workloads:
Here's the average of those workloads:
ER standing for "Extremities Removed" (i.e. absolute single high and low-results). For CML-S the absolute low being "Buddhabrot Fractal" and the high being "WinRAR".
Meanwhile for VMR, the removed low was "GMPBench" and the high "Vampire Numbers".
All of the test systems adhered to their maximum, officially supported memory frequency during the testing. For AMD (both MTS and VMR) that meant 3200MHz with 14-14-14 timings and 2933MHz with 13-13-13 timings for the Intel 10900K system.
I've got a truckload of other results for different scenarios however, I'm not sure if they're even relevant, as the performance of the AMD Ryzen 5000-series is so utterly crushing.
The reason why the performance in both GMPBench, or rather in libGMP and OpenSSL are lower than expected is being looked into.
To give you a quick idea of scaling, these CPUs are mostly temperature limited, due to the tiny CCD size of around 82mm², hence the extreme intensity.
Overall, the scaling is excellent.
Here's what a typical core only power dissipation vs. temperature will look like, more or less regardless of the cooling.
