keyword: Zen 3

We’ve been pounding on Cyberpunk 2077 benchmarks with our GPU testing and CPU thread fixes. But what about memory? Does your PC’s memory configuration matter? Of course, the answer is yes, but we wanted to see just how much it matters.

We’ve run a handful of benchmarks on an Intel Core i9-10900K system, not because it’s necessarily the fastest CPU for playing Cyberpunk 2077, but because Intel’s Z490 platform tends to have better compatibility with memory kits. The results here should apply to X570 Zen 3 and Zen 2 CPUs and other platforms as well, but we haven’t taken the time to confirm that yet. Consider these preliminary findings as yet another data point if you’re trying to build the ultimate Cyberpunk 2077 PC.  

Our test PC, with full specs listed on the right, is basically maxed out in nearly every area. The Core i9-10900K is Intel’s fastest current CPU (for gaming purposes at least), we’re using the RTX 3090 Founders Edition to ensure the GPU minimizes bottlenecks as much as possible, and we’ve got a high-end motherboard, liquid cooling, storage, and power keeping everything running.

Then we’ve got the three memory kits that we’ve used for testing. At the top of the ladder sits the 4x8GB Patriot Viper DDR4-4000 CL19 kit. We’ve run this in both 4x8GB and 2x8GB. Next, we have a Corsair 2x16GB kit rated for DDR4-3600 CL16. Finally, we have an old 2x4GB kit of DDR4-2666 CL16 memory from Corsair at the bottom end of the spectrum. It really is an old kit, as it was once used in an X99 Haswell-E build. This is something of the worst-case scenario for a modern platform.

We don’t have exactly matching memory kits of different capacities, but to test how much memory capacity matters compared to speed and timings, we used the 2x8GB and 2x16GB kits with the same detected timings and speed of the 2x4GB. The motherboard didn’t clear RAM timings when we swapped kits, so this is about as close as we can come to ‘equivalent’ speeds, with the only variable being capacity. However, the 4GB and 8GB sticks are single bank, while the 16GB sticks are dual bank, so there’s a slight advantage for the latter.

We’re using the same benchmarking methodology as in our other Cyberpunk 2077 performance articles. We walk around Night City just outside of V’s apartment, following the same route each time, and log frametimes using OCAT. We test at 1080p, 1440p, and 4K using the medium, ultra, and RT ultra presets. Each test is repeated twice, plus an extra run at 1080p that gets discarded, and we reboot after switching the graphics preset. Here are the results, broken down into medium, ultra, and RT ultra testing.

Starting with 1080p medium, we see some of the largest differences between the memory kits. That makes sense as this is the setting most likely to hit CPU bottlenecks, which go hand in hand with memory bottlenecks. The fastest configuration is the 4x8GB DDR4-4000 CL19 kit, followed closely by the 2x16GB DDR4-3600 kit. 32GB of high speed memory ended up improving performance by up to 20 percent compared to the slowest kits, though capacity is also a factor. The 2x16GB DDR4-2666 kit is about 12 percent slower than the same kit running at DDR4-3600 speeds.

1440p doesn’t change things much, but 4K medium pushes the bottleneck over to the GPU enough that there’s only about a 3 percent spread from the fastest to the slowest kit. That’s almost margin of error for these tests, since Cyberpunk 2077 is an open world that doesn’t guarantee the same NPC behavior between runs.

Bumping up to the ultra preset, 1080p remains similar to what we saw above, but now 1440p and 4K run into the GPU bottleneck and end up with only a 2 percent spread from fastest to slowest RAM. So, if you don’t have 32GB, you’re not necessarily giving up a ton of performance, especially since your GPU will be a major factor. We’re using an RTX 3090, but a more sensible choice like the RTX 3060 Ti would push the bottleneck over to the GPU much sooner (i.e., at 1440p medium or 1080p ultra, we’d likely get a very limited spread in performance).

Given what we’ve seen so far, you might think enabling ray tracing and DLSS would end up with everything tied, regardless of resolution. That doesn’t happen at all at 1080p using the RT Ultra preset, and in fact, the spread is slightly larger this time (26 percent from the fastest to slowest GPU). Capacity clearly starts to matter more with ray tracing, and the 32GB DDR4-4000 kit is about 9 percent faster than the same kit with only 16GB (two DIMMs).

1440p RT Ultra does narrow the spread to 8 percent, and 4K RT Ultra only shows a 4 percent difference between the fastest and slowest memory kits. Still, 2x4GB is clearly scraping the bottom of the barrel, and given memory prices, most people with a rig capable of handling Cyberpunk 2077 should be using at least 16GB, if not 32GB.

Want to see more memory testing? Check out our Best DDR4 Memory guide. And if you want to see more Cyberpunk 2077 memory testing, like on the AM4 platform, let us know in the comments.