Clock rates and ranks in comparison
Let’s start with Full HD or 1080p: Here Cyberpunk wants one thing above all else, bandwidth! All five dual-rank configurations can be found at the top, with the 10900K ahead of the 5950X. As mentioned earlier, on the 5950X in dual-rank we were only able to use Geardown Mode, which should give it a latency advantage over 2T. Still, the 10900K is consistently faster, with DDR4-4400 marginally ahead of DDR4-3800. With a run-to-run variance of our benchmark of about 2%, however, one can speak of a tie here.
Surprisingly, the fastest single-rank config is DDR4-4400 with the 5950X and asynchronous ratio of Infinity Fabric and RAM, despite the large latency disadvantage that comes with it. Cyberpunk 2077 in 1080p really lives on bandwidth! This is followed by the rest of the configurations with the 10900K and then the rest from AMD. In addition to DDR4-4400, we also tested DDR4-3800 in 1:2 mode for verification. So the latency disadvantage compared to the synchronous DDR4-3800 does make itself noticeable in 1080p in the form of about 6% in AvgFPS and about 9% in 1%-Low FPS, resulting in last place.
The measured difference between the fastest and slowest config with the Ryzen 5950X in average FPS are a whopping 15.9 FPS, and 13.1 FPS when excluding asynchronous configs. That equates to 17.7% and about 14.6% difference respectively with identical CPU and GPU just by changing the RAM configuration. On the Intel system, the largest measured gap is 13.2 FPS, which is 13.8%. So RAM configuration makes a noticeable difference at 1080p Ultra Raytracing and comparable settings.
When it comes to frame time variances in 1080p, the rankings are a bit more mixed. Higher bandwidths and dual-rank configs tend to be found higher up, lower bandwidths and single-rank configs lower down. We cannot detect other patterns also because of run to run variances. The only constants: The Intel system with DDR4-4400 is again the leader and the AMD system with asynchronous DDR4-3800 is the taillight.
At 1440p we are increasingly GPU limited, but can still see differences greater than the 2% run to run variance. Intel clearly tops this chart, with all dual-rank ahead of single-rank configs. In the AMD results, DDR4-4400 with asynchronous IF comes out on top again, followed closely by the dual-rank configs. The last five places are occupied by the remaining AMD single-rank configurations.
In terms of frame time variances, only the two Intel configs with DDR4-4400 can really stand out. The rest follows with similarly constant frame times. Only the 5950X with single-rank DDR4-3200, which is also theoretically the slowest config, is at a clear disadvantage in the frame time variances and comes in last place.
In UHD, we are now almost completely bottlenecked by the GPU. Only the three dual-rank configs of the 10900K can measurably and repeatably stand out at the top. Behind that, the results largely blur in run to run tolerance. Only the two dual-rank configurations and the asynchronous DDR4-4400 configuration, both with the 5950X, stand out positively in the 1% low FPS.
In the frame time variances, the dual-rank configs are now at the top together with the asynchronous DDR4-4400 config. Doesn’t that sound familiar? Yes it does, because that’s exactly how the Average FPS ranking looked in Full HD. So while due to high bandwidth we benefit from more Average FPS in 1080p, we see an advantage in frame time variances in 4K. The single-ranked configs, whether with Intel or AMD line up behind it.
Conclusion and recommendations
Of course, to gamble in and around Night City, you’ll need a big CPU and an even bigger GPU. Especially at higher resolutions and graphics settings, the GPU bottleneck becomes noticeable, slowing down the entire render pipeline no matter how much data the CPU can deliver. However, as our test today has shown, you still should not disregard the working memory. 22.8% is the difference in 1080p Average FPS between the fastest and slowest configuration tested, which is practically the same as the difference between an RTX 2080 Ti and RTX 3090.
Now you still have to keep in mind that DDR4-3200 CL16 is definitely not slow RAM either, and a large part of Cyberpunk gamers on PC are probably running at even lower XMP settings or even JEDEC clock rates. At these lower clock rates even greater differences in 1080p Average FPS and 2160p Frametime variances can then be expected.
In today’s test, we’ve left latency out of the equation as much as possible with the goal of being able to compare the effects of RAM bandwidth due to clock speed and number of ranks in isolation. Now, due to the platform, we were partially limited in the working settings, to the detriment of the perfect comparability of the different configurations, but we can already state without a doubt that more RAM bandwidth has a positive effect on the gaming experience in Cyberpunk 2077 in any case.
And since the latency between the tested DDR4-4400 and DDR4-3600 remains effectively the same, the load on the RAM also remains the same. This means, for example, that you can achieve a considerable performance boost with a DDR4-3200 CL14 XMP kit just by increasing the memory clock and timings in the same ratio. And you probably won’t even have to adjust the RAM voltage, since the latency effectively stays the same.
Even though we “only” used three different resolutions with the “Raytracing Ultra” preset for our test, we can already draw conclusions about the expected performance with other settings. The tendencies are clear: The more RAM bandwidth, the more FPS you get at low resolutions and settings, and the more constant the frame times are at higher resolutions and more demanding graphics settings. The logical next step would be to compare different latencies at constant bandwidth, from 1T and 2T command rates and re-test upcoming patches of the game. But if you have any other suggestions, ideas or improvement requests, feel free to let us know in the forums!
Kommentieren