Y-Cruncher and Far Cry 6
Let’s start with a bang, in the Y-Cruncher, here with preset 2.5b. The benchmark belongs to the most demanding ones at all regarding power consumption and thus also heat dissipation. While you might be able to run 5.3 GHz “stably” in Cinebench, it’s not uncommon to have to go back several hundred megahertz, say to 5.1 GHz, at the same voltage for Y-Cruncher. And the benchmark puts AVX-512 to use – so perfect to analyze the correlation between power consumption and compute power output.
In terms of power consumption, the first thing that stands out is that the P-cores alone draw more juice from the socket than when the E-cores drink along. So even if 24 threads are actually used by this benchmark, the performance is not better then, quite the opposite. “Too many cooks spoil the broth,” comes to mind. It looked similar in the test a few days ago in LinpackXtreme, where the small cores seemed to get in the way of the big ones. Alone and with AVX-512 enabled, the P-cores at least manage to almost return to the same power consumption as Stock.
Unlike LinpackXtreme, the E-cores at least help a bit here, so that the computing time is at least 10% shorter than when the P-cores do the work alone and without AVX-512. With AVX-512, however, the 8 P-cores alone are even more than a second faster again.
If you now put the computing time in relation to the power consumption, at the end of the whole action not much has happened in the first moment. The advantage of the 8 Sunny Cove P-cores with AVX-512 compared to the standard configuration of the Alder Lake CPU is vanishingly small and probably even in the range of the measurement tolerance. But I’ll explain in the conclusion why I think this should not be neglected.
Moving on to our game title, Far Cry 6. Here we measure with an RTX 3090 in 1080p resolution, as this is where the differences between CPUs were among the most apparent in our gaming tests in the past. GPU power and temperature limits are maximized for maximum focus on the CPU as well.
Again, the CPU consumes significantly more power without the E-cores than with, and more without AVX-512 than with. What is different from our previous measurements, however, is the difference between the power consumption measurements of software and hardware, which has shrunk a bit here. But that was actually to be expected as well, since the power consumption during gaming is lower overall than in the compute benchmarks and thus the “measurement error” is also smaller in the same proportion.
Even more interesting is that although the power consumption differs by up to 20 W, the FPS is almost the same between the configurations. I can’t really explain this yet, but it’s probably a peculiarity of the engine that benefits from both E-cores and AVX-512.
The efficiency for the P-cores without AVX-512 is then correspondingly poor, while the other two configurations are effectively on par.
49 Antworten
Kommentar
Lade neue Kommentare
1
Veteran
Veteran
Mitglied
Mitglied
1
1
Urgestein
Urgestein
1
Mitglied
Mitglied
Urgestein
Veteran
Veteran
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →