In this article, we’ll investigate the impact of RAM speed on the Intel Core i9-11900K and i9-10850K in gaming. It’s well known that Ryzen Zen 2 and Zen 3 CPUs perform optimally with higher speed RAM, but we’ve often heard it said that ‘RAM speed doesn’t matter as much with Intel’. We wanted to test that.
With the 11th Gen Rocket Lake CPUs Intel introduced the notion of ‘Gear 1 and Gear 2’ settings for the memory controller. This is analogous to Ryzens’s uClk setting, in which it’s important to match the memory controller clock and infinity fabric speeds to RAM speed for optimal performance. Now, Intel has also handed control of this behaviour to the end-user – or the motherboard manufacturers – so you can adjust this behaviour in 11th generation K series CPUs.
We’ll start by looking at RAM speeds and controlling Cas Latency, then move on to looking at Gear 1 vs Gear 2 and what that means for performance.
The Test Setup:
Whilst testing speeds we used the ASUS ROG MAXIMUS XIII Hero Z590 Motherboard with the recent 0605 BIOS which included Rocket Lake-S specific microcode and improved memory stability and flexibility. We kept CL timings fixed at 16-16-16-32 except for the higher clock speeds where loser timings were required for stability. We used our 16Gb Samsung B-die RAM kit, capable of 4400MHz speeds but changed RAM settings to the primary timings and speeds shown.
i9-11900K & i9-10850K RAM Performance Analysis
Shadow of the Tomb Raider: i9-10850K
Looking first at Shadow of the Tomb Raider on the Intel Core i9-10850K you can see how marked the trend is.
This is the benchmark reported CPU performance, independent of GPU. Moving from 2400MHz CL16 RAM to 3600MHz with all other settings controlled yields a 30 FPS performance increase. Further increasing speed to 4000MHz and reducing latency, but by a smaller amount owing to the lift in CL timings, gives another 4% or 8 FPS average. At 4400MHz CL19 we see our highest performance at 195 FPS average, which’s 43FPS or 28% faster than the base settings. To be clear, 2400MHz or even worse 2133MHz is what will happen by default if you fail to set XMP or the XMP profile on your RAM kit isn’t stable and it resets itself. It can cost you significant performance.
Here’s Rainbow 6 Sieges benchmark, a fast-paced shooter where every FPS matters. Again, this benchmark is consistent and RAM speed-sensitive.
Here we get a slightly different trend, with the best performance at 3600MHZ CL16, and then performance softens at 4000MHz CL17 and 4400MHz CL19. Again, here our overall latency and the looser timings required to keep RAM stable at those very high speeds becomes detrimental to performance.
Shadow of the Tomb Raider: i9-11900K
Moving on to Intel’s latest CPU, the i9-11900K and using the same benchmarks, we can see the same trend with a couple of caveats. Note these tests were performed in Gear 2 mode because Gear 1 was unstable at the higher memory speeds. That’s also why 4400MHz is missing from these results, it wasn’t possible to stabilise it in the gaming benchmarks at that speed.
Here again, we see the serious boost in performance moving from 2400MHz up to 3000MHz, then to 3600MHz adds a total of 55FPS average to the CPU performance, with corresponding hikes in minimum and maximum performance metrics. There’s a 41% boost to performance moving from 2400MHz to 3600MHz CL16. At 4000Mhz Cl17 performance tails off marginally – again we’re approaching the limits of the memory controller and slacker secondary and tertiary timing dent performance overall.
Confirming this behaviour in Rainbow 6 Siege, working through the same settings yields the following results:
Again we see peak performance centred around 3600MHz CL16, and a softening towards 4000MHz but were not giving much away between 3000Mhz and 3600Mhz with reasonable timings, 10 FPS or so.
And finally, just to confirm this trend and to show a little data for 4400MHz RAM on the 11900K, we were able to get the Time Spy CPU benchmark to complete at 4400MHz CL19, building the following set of results:
This is perhaps less informative aside from the fact of having that highest speed represented as it’s just a score, but it is indicative of aspects of the CPU performance and we know from another testing that this benchmark does scale well with RAM latency. It looks like we’re really at the bitter end of the memory controllers performance at these higher speeds. It would take a significant investment of time in refining RAM timings manually to, first of all, stabilise the system, and further improve performance. Nevertheless, 4400MHz CL19 does yield the highest score here but it’s not stable enough to run any games.
Gear 1 vs Gear 2
Moving on then to the question of Gear 1 Vs Gear 2 on capable Rocket Lake 11th Generation CPUs, this is a simple toggle in BIOS that runs the memory clock at either half memory speed in Gear 2, or at the same speed in gear 1. Gear 1, therefore, reduces latency, as there’s no missed cycle in communication. The penalty for this is the memory controller becomes less flexible in settings, so we’ve got limited data. Like Ryzens uCLk and FCLk settings, it also appears to be ‘unhappy’ at higher frequencies, and it wasn’t possible to stabilise behaviour at 4000MHz RAM speeds/2000MHz memory controller speeds in the time available. We satisfied ourselves with conducting A-B testing to see what kind of performance difference we could see at settings we could apply consistently across the tests.
Here we can see small but significant performance benefits from running at Gear 1 and 3600Mhz, with a 10FPS uplift in average frame rates on Rainbow 6 Siege, and we verified this uplift at 2400MHz base settings as well although of course, overall performance is much lower due to the overall speed impact discovered above.
Finally, for a real-world demonstration of this benefit, Flight Sim 2020 is a highly demanding simulator that is frequently CPU bound. Running our standardised benchmark flight over Manhattan yields the following results:
Here, we can see that moving to Gear 1 and matching memory and memory controller speeds yields a small but valuable lift across the full suite of metrics and translate directly into improved performance.
Verdict
This set of results demonstrates how important overall memory latency, as indicated by speed and timings, is for Intel CPUs. Clearly, it can have a significant performance impact in any situation where you’re wholly or partially CPU limited.
It highlights the importance of ensuring that your Memory is correctly set up. If you didn’t enable the XMP profile on your RAM during set up, or it’s been reset in a BIOS update or by clearing your CMOS, you could be hurting performance significantly. It takes a moment to check this using CPU-Z or Hwinfo64 and doesn’t necessitate entering BIOS.
For both Comet Lake and Rocket Lake K series CPUs, there appears to be a sweet spot in the region of 3600MHz CL16, and that’s a good starting point for refining system performance. To clarify the settings we used here were ‘quick and dirty’ because of time constraints, with the motherboard left to decide most secondary and all tertiary settings itself. Time spent tightening those timings could yield further performance improvements, but it’s a time-consuming process and tends to offer diminishing returns, and you have to check and validate results not only for performance but for stability as well.
For most users with K series CPUs, it seems that much like Ryzen you’re best off buying a kit of 3600MHZ CL16 RAM or something with a similar total latency, and if you do that and enable XMP you’ll have gained the bulk of the performance benefit for virtually zero effort. If you spend as much time in BIOS as your operating system, then buying even faster RAM and using the headroom to lower latency as much as possible can yield significant rewards on these CPUs.