Given the large number of different RAM configurations we're operating with, we've chosen three games for our testing: Ashes of the Singularity Escalation, Far Cry 5 and Crysis 3. This should give us a good range of results that should be broadly applicable to AAA games made in the last few years, with the expectation that more CPU-dependent games and configurations will see a greater advantage from RAM frequency increases.
Please keep in mind that, while the games and their settings are the same as our processor tests, our results here aren't intended to be comparable to our CPU results. That's because we've locked our Ryzen 3700X to an all-core turbo of 4.2GHz to minimise run-to-run variance; we'd expect the processor to boost intelligently past this mark and achieve better frame-rates under its default configuration.
The results here are shown with our familiar in-house performance analysis tools. On a full-fat desktop or laptop PC (as opposed to a smartphone or laptop), you'll see a YouTube video for each game. Click the play button, and you'll see the results from each run play out in real time in the frame-rate and frame-time graphs below the video. We've selected four runs to play by default, but you can tick or untick the boxes on the right side of the screen to choose the results that you're most interested in.
Below the real-time metrics, you'll see a bar chart of the overall frame-rate. Here, you can hover over the results to switch between the averages and the (frequently more illuminating) best or worst one per cent scores. You can also click on the bar chart to switch between the recorded frame-rates and the relative differences from the result you're hovering over and the others we recorded. If you're on mobile, you'll just see the bar chart, which will be more static, so it's worth revisiting the page on desktop if you want to go deep into the results!
Real time strategy title Ashes of the Singularity Escalation is the first pick, thanks to its integrated CPU test. This benchmark was programmed to perform all of the processing of the real game, so it's a truly representative measure of in-game performance - with the only wrinkle being units can't die, so unit counts remain extremely high throughout the test, a kind of nightmare scenario for any RTS player.
Let's look at the 4000MHz results first, as they're quite interesting. The slowest result of the three is unsurprisingly the RAM at its XMP latency of CL19. This results in a baseline speed of 46fps, just a hair faster than the 3600MHz results with tighter CL16 timings. Swapping to tighter timings at 4000MHz boosts the average frame-rate to 48fps, a three per cent improvement. Our final trick is ramping up the fabric clock from 1800MHz to 1900MHz, which nets us a further two per cent improvement and a final fps average of 49fps. So if you use this particular kit at XMP, you're losing out on a solid five per cent of additional performance.
Looking at the full range of results, you can see that performance uplifts sharply from our starting point of 2133MHz. We get eight per cent faster frame-rates at 2400MHz, a further six per cent from 2400MHz to 2800MHz and a further two per cent at 3200MHz. From here, the improvements continue at roughly the same rate, with an extra two or three per cent of performance from each 400MHz jump. All told, from 2133MHz C16 to 4000MHz C16, we get a 22 per cent improvement in performance - not an insignificant improvement! So if you have RAM of any kind, do be sure to check that it's running at its rated XMP speed!
Finally, let's see how the infinity fabric overclocks worked. The 3800MHz memory with 1900MHz fabric clock should be a sweet spot, and indeed we see the joint-highest frame-rates here: 48.3fps, effectively tied with the 4000MHz RAM with the same 1900MHz fabric clock at 48.5fps. So we can see that the penalty for an uncoupled fabric clock is measurable, but a boost of even 100MHz provides enough extra performance to make it worthwhile.
Perhaps unsurprisingly then, our 2400MHz results with overclocked infinity fabric are very impressive. This configuration saw a four per cent jump in performance, nearly as much as we saw from increasing RAM frequency by a solid 400MHz. So if you have relatively slow RAM, then it's well worth testing an infinity fabric overclock to see if it improves frame-rates in the games that you play most frequently.
Ashes of the Singularity: CPU Test
Next up, we have Far Cry 5. This game is the most recent release in our small suite and one that relies heavily on single-threaded performance for its world processing, making it a natural candidate for memory testing.
Starting again with the 4000MHz results, we see a massive six per cent advantage from tightening our timings from CL19 to CL16 - not a bad way to start! Overclocking the infinity fabric from 1800MHz to 1900MHz results in a further one per cent leap in performance, making it only just worthwhile. Still, our optimised configuration is still seven per cent faster than the RAM at its stock speeds, moving the average frame-rate from 122fps to 130fps.
We saw a 22 per cent improvement from 2133MHz to 4000MHz at CL16 in Ashes of the Singularity, and we accomplish nearly as much in Far Cry 5 with a 20 per cent improvement from the lowest to the highest frequencies. Again, the biggest jumps come at the bottom of the stack, with around a six per cent improvement for each of the first three comparisons - 2133MHz vs 2400MHz, 2400MHz vs 2800MHz and 2800MHz vs 3200MHz. Interestingly, the game doesn't get any faster at 3800MHz or 4000MHz compared to 3600MHz - in fact, it gets a bit slower. It's likely this is down to the uncoupled fabric clock, which we can redress this by overclocking the infinity fabric to 1900MHz. That makes our second 3800MHz result slightly faster than our 3600MHz one, but with no 2000MHz fabric clock possible we're not able to achieve a similar average frame-rate at 4000MHz.
Finally, the 2400MHz memory with overclocked infinity fabric again shows a worthwhile improvement of around three per cent, compared to around five per cent for jumping from 2400MHz to 2800MHz without an infinity fabric overclock.
Far Cry 5: Ultra, TAA
Finally we have Crysis 3. This game is something of a bona fide classic, having been released in 2013, but it remains a Digital Foundry mainstay thanks to its challenging jungle scenes (oh, and memes about Alex's Crysis shrine, obviously).
Crysis is the most consistent when it comes to its frame-rates, showing that frame-rates do slowly improve as frequencies ramp up but the spread from top to bottom is just eight per cent - nearly a third of the increase we saw in our other games. Interestingly, the 2400MHz result with overclocked infinity fabric is actually faster than 2800MHz with the default infinity fabric setting, making this overclock much more significant than usual. We see around a one per cent improvement from our 3200MHz and 3800MHz infinity fabric overclocks, while the 4000MHz infinity fabric overclock doesn't result in a higher frame-rate - perhaps the latency penalty and the frequency advantage are essentially cancelling each other out.
Crysis 3: Very High, SMAA T2X
OK, with that our gaming testing is complete - at least for the moment. Let's move onto the conclusions, where we'll lay out our RAM buying advice for Ryzen systems.
Testing 4000MHz RAM: Are higher frequencies worth it?