Page 2: Loudness and power consumption

Contents

One of the major criticisms of the Arc-A series was the unseasonably high power consumption in idle mode, i.e. when the card only had to display the Windows desktop. Although there were a few tricks, such as switching on native PCIe power management via ASPM in the BIOS and setting the "Connection state power management" for PCI Express to "Maximum power savings" in the Windows power saving plans, this did not work in all cases. This was also the case on our test system and various other PCs.

Intel's Arc B580 can now do this better. Without these tricks, the idle power consumption was still a high 37 to 40 watts, regardless of whether a Full HD display with 60 Hz was connected, a 4K144 screen or a mixed triple. However, with the above settings, the Arc B580 only needs just under 6 watts for an FHD60 display, 8 watts for two and 17 watts for three. 4K screens were at 21 (1 screen) to 23 watts (2 monitors), with 120 Hertz at 26 watts. With 144 Hertz in 4K or in mixed mode, the energy-saving tricks were virtually ineffective.

They also had no effect on the card's power consumption under load. This was up to 185 watts in 3D scenes and thus just below the total board power of 190 watts set by Intel; individual peaks in the millisecond range just exceeded the 200-watt mark. In the full load test with Furmark, we measured 207 watts with peaks of 225 watts, which corresponds exactly to the official limit of the PEG slot and the card's single eight-pin power connector.

The clock rate also showed that the card was working at its limit. In games and the 3D tests, it was otherwise stuck at 2850 MHz, i.e. 180 MHz above the specified "Game Clock" as the average boost in games. In Furmark, however, the card had to drop to 2700 MHz, but was still above Intel's promise.

Quiet reference card

The two-slot cooling system with its two 85-millimeter fans is not working at its limit. As usual, the fans mostly stand still in idle mode and do not produce any noise. As they start to rotate at 49 °C with a 30 percent PWM signal, they sporadically started up briefly at an inaudible 600 rpm during the test. This increased to up to 1500 rpm under full load, but even then the card was only faintly audible from the casing. Our measuring device in a low-noise room showed a very good 0.5 sone, even the control measurement with the side panel open showed only 0.1 sone more – very nice.

The OC functions for the graphics chip in the driver caused us a few crashes, which also deactivated the "rBar" function, which is important for optimum performance, each time after the subsequent reboot, so that we had to reactivate it in the BIOS setup. We cannot describe this function as really stable at the moment. We were able to overclock the graphics memory more reliably, albeit with moderate success: Our test sample only managed a measly 19.8 GT/s instead of the factory default 19.5 GT/s. We experienced regular reboots above 20 GT/s.