I picked this Unreal Engine 4 based game because it (not only) contains a very consistent and long internal benchmark scene that is reproducible almost to the decimal point and I know that there will definitely be no memory or bandwidth limitations out of order. Three TGP classes for the GeForce RTX 3090 are used this time. For the 420 watts I use a MSI RTX 3090 SUPRIM, the 350 watts I solve with the original RTX 3090 Founders Edition (FE) and the 300 watts also with the FE including a slightly limited power target (but without changing the boost clock manually).
The 300 watts then corresponds to the power consumption of the Quadro RTX A6000 and we can easily observe how both cards act with almost identical power consumption. The two RTX 3090 cards with 350 and 420 watts beat the RTX A6000 reproducibly in this run, but the 300 watt variant then has to admit defeat in the “1:1” comparison. We can see that the differences are very small, which is why I put so much emphasis on reproducible benchmark runs, where even the leaves keep falling in the same pattern and there is no open world.
The FPS curves over the (long) runtime coincide very nicely, but also show the (marginal) differences in the cards’ behavior.
The percentiles now show (better than the bars) that only the 420 watt variant of the GeForce and the Quadro RTX A6000 have roughly evenly sloping curves, but the FE buckles slightly in both watt classes from 350 watts downwards.
The frame time, on the other hand, reflects the results from the FPS bar chart quite well. But even here you can see where the disadvantage of the GeForce RTX 3090 lies at only 300 watts.
In terms of variances, the Quadro RTX A6000 is now just behind the MSI RTX 3090 SUPRIM, but well ahead of the two variants with the GeForce RTX 3090 FE. This finding is also based on the elaborate calculation and shows in clear bar form what can be subjectively perceived well even from the corner of the eye. Thus, the variances are once again more important than the pure frame time progressions.
Now we come to the power consumption, where all cards were loaded with almost 100%. You can see that the two 300 watt cards really almost make a precision landing and that the two more drinkable cards also hit the limit stored in the BIOS pretty accurately. So, from that point of view, there were no surprises this time, as there were in WWZ.
You can also see this in the CPU’s power consumption, which differs in frames of less than 2 watts. Here, the Quadro RTX A6000 is slightly behind for the time being, but still within the possible error tolerances.
Finally, let’s compare efficiency. The winner, once again and not entirely unexpectedly, is the Quadro RTX A6000, which is the amped-up version with the statutory abstinence.
Here again the gallery for all detail lovers with the frame times in relation to the power consumption
32 Antworten
Kommentar
Lade neue Kommentare
Veteran
Urgestein
Urgestein
Veteran
1
Veteran
Mitglied
Veteran
Urgestein
Urgestein
Urgestein
Mitglied
Urgestein
Veteran
Veteran
Mitglied
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →