Overview of average FPS and percentiles
Let’s start by leaving the percentiles out of the equation and compare the average FPS within the two cards. The first finding is that the average barely changes for the AMD cards, no matter how much memory was available for the game. This is very different from what happens with NVIDIA. In percentage terms, however, you can see that the P1 low drops much more on AMD below the 10GB mark in relation to the average FPS. But if the average looks so good at AMD, we could already conclude that there must have been longer jerks at AMD for the P1 Low to turn out worse.
If we now look at the FPS progression over the 163 seconds, we can see from the curves of both cards where the cuts are and where reloading was also necessary. While NVIDIA only experienced real drops in the 10 GB variant (i.e. the equality between available memory and memory requirements), AMD shows the same drops from 12 GB downwards. Always at the same point (from zoom into the long shot, complete cut). With NVIDIA, we also see larger differences on average, but fewer real drops.
Therefore, we now look at the distribution of the percentiles. At 99% (i.e. the P1 Low), both cards show the biggest drop. However, we can also see that the drop of the last 0.1% is lower with AMD in relation to the slowest 1% than with NVIDIA. We’ll take a closer look at that in the variances. So please keep this in mind!
Frame times and variances
The frame times on the AMD card seem more balanced, although if you look closely, you can see red stripes on the far right of the first four bars (as 12 GB memory expansion and less), which indicate real drops. You don’t see this effect with NVIDIA. And yet, the frame times don’t really help us in this bar consideration. They are shown on the next page as nice single graphics with curve progression.
We have to look at the variances. We see that AMD shows extreme variances from 12 GB downwards, which visually express themselves in occasional stutters or real hangs. With NVIDIA, this is only so extreme at 10 GB, even though it occasionally starts to twitch slightly at 12 GB. But you will hardly notice that. However, we also see that below 10 GB, i.e. the area of the memory shortage, the gradient “smoothes out” again, even if this advantage then has to be bought visually. A short counter-test with the Radeon Pro W7900 did not show any difference either. NVIDIA’s “smart” memory management from the advertisement thus works quite well, even if there are pixel grits in the textures and reductions in the content from time to time.
On the next page, we will now compare the separate FPS trajectories of each memory upgrade. This took some effort, but especially from the memory shortage area, we will then see the differences best!
113 Antworten
Kommentar
Lade neue Kommentare
Veteran
Urgestein
1
Veteran
Veteran
1
1
Veteran
Neuling
Urgestein
Mitglied
Veteran
1
Veteran
Neuling
Urgestein
1
Veteran
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →