With still 2.97% performance loss, the disadvantage compared to PCIe 3.0 @8 is still visible with this fast card, even after the CPU limit is removed. This is measurable but not quite as dramatic compared to the run with @16. The difference in both PCIe 4.0 measurements with the Radeon RX 5700XT, however, is again significantly below one percent, which one has to impose as a measurement tolerance, at 0.51%. Since the RX 5700 XT is inherently much slower, a difference of meanwhile 7 FPS to the Quadro RTX 6000 @8 documents that the bandwidth advantage is not everything when the graphics card gets more to do and therefore the large card does not have to suffer from the CPU limitation. Fewer drawcalls.
You can see the difference quite well again on the FPS curves:
Now let’s take a look at the percentiles, because the 99th percentile is the most important one. Percentile isn’t everything. What you see is the now break-in of both cards, no matter what connection, when it is over 99. …goes out. Whereby the Radeon is once again somewhat more affected.
At Frame Time, the fully connected Quadro RTX 6000 wins for the first time over the halved version. The RX 5700XT performs approximately the same with both connections. Also here I have the whole progressions later as single graphics for each map.
The evaluated variances can then again illustrate this a little more precisely, because the Quadros are again the measure of all things. The two measurements on the PCIe 4.0 are virtually identical.
Individual graphics for each run as picture gallery
Kommentieren