Well then, finally to the hard numbers. What is the point of all this fiddling around and is it worth it? Decide for yourself! I have measured some games with different settings. Here are my defined settings.
- UV: Equivalent to 2.4Ghz with reduced power consumption and about 940mv. I use the ARC tool for this.
- Floor: The map our of the box.
- Balanced: A moderate overclocking to 2.6 Ghz at a voltage of 1070 mv I use the ARC tool for this. The power consumption hereby mostly stays in a range between 190-230 watts ASIC.
- OC: Overclocking with the ACC to always get the maximum performance out of the power budget via V/F Curve. The card then clocks with 2.55-2.72 Ghz and usually runs at the limit.
In the benchmarks, I partly included the archive data of my PowerColor Hellhound 6600XT. The data is no longer dewy, but should be a useful orientation in the benchmarks shown.
First the synthetic benchmarks. You can see here that the OC setting is not necessarily faster than the balanced setting. The card can thus keep the 2.6Ghz stable and with less power consumption, which is an advantage especially in performance hogs where the card can load itself well, like here. The Powercolor is always inferior in the synthetic benchmarks, regardless of whether it is stock or overclocked. The UV setting is usually pretty close to the stock values, the Phantom has 210 watts of ASIC available and can thus, in contrast to Intel’s Limited Edition, actually maintain the 2.4 GHz almost throughout. However, we partly save up to 40 watts of ASIC power compared to the default settings. Even in TS Extreme, the card only needs about 180 watts.
The synthetic benchmarks show the performance potential of the card and the chip very well. For example, an RTX 3070 achieves comparable rates at around 15k in the TimeSpy benchmark.
Unfortunately, the world in games then looks somewhat different again. The drivers have matured considerably since the release and have certainly brought one or two performance improvements. However, there is still potential, as we will now see in comparison to the 6600XT. As an example, here are some titles in WQHD resolution (2560×1440).
The Hellhound can keep up well in the titles here, especially because of its considerable overclocking headroom that characterizes the whole RDNA 2 series. Nevertheless, the ARC 770 runs very well in the titles and brings the performance to the road. In some titles, the Hellhound is then also clearly inferior, like here in Red Dead Redemption 2.
The ARC can also convince with its performance in Cyberpunk 2077. With overclocking, up to 10% more performance is possible in some cases. In Cyberpunk and especially for resolutions above WQHD, an overclocking of the memory would really be desirable since it could certainly increase the performance considerably. It’s a pity that Intel doesn’t make anything possible here.
In conclusion, tuning on the ARC is a lot of fun, and some of the results are respectable. Currently, it is unfortunately quite cumbersome, and the card reacts to excessive settings, especially in the ACC with sometimes nasty crashes that can lead to forgetting RBar. This can all be fixed, nothing will break. But if the layman then sits in front of such a screen after a failed overclocking, he will surely quickly lose interest in it.
I advise moderate OC with the ACC based on the power consumption you are comfortable with. This is the easiest to handle and usually brings similar results as a stable overclock with the ARC tool in all applications. Do not overdo it with the “power gain” control. Values significantly above 50 lead to unpleasant crashes and are usually not stable everywhere. Use an application that is really demanding for testing. The TimeSpy benchmark is well suited for this, but also any game that brings the ARC close to the load limit like RDR 2 or A Plague Tale.
Undervolting can sometimes achieve remarkable results. I was able to run the standard clock stably in all applications with 940 mv and partly save significantly on power consumption. Here are some impressions, note that the given values always represent the ASIC, not the power consumption of the entire GPU. At 1000 rpm, the GPU is then also really inaudible at moderate temperatures.
77 Antworten
Kommentar
Lade neue Kommentare
Veteran
Urgestein
Mitglied
Urgestein
Urgestein
Urgestein
Veteran
Urgestein
Urgestein
Urgestein
Urgestein
Urgestein
Mitglied
Veteran
Urgestein
Urgestein
Urgestein
Mitglied
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →