Intels 12th generation thrives on faster RAM – but how fast should you go for best value?
Intel’s 12th Generation CPUs have impressed across the board with their performance. To get the most out of them, it makes sense to pay close attention to your choice of RAM. In this article, we’ll explain how RAM specification can help improve CPU performance, show you our testing which demonstrates the effect of different RAM speeds, and then make our recommendations as to the RAM that will help you get the most out of an Alder lake 12th Generation Intel CPU.
DDR4 Vs DDR5
One of the features of the 12th generation CPUs is their support for DDR5 RAM. This is dependent on the motherboard you choose, with your motherboard choice tying you to either DDR5 or DDR4 RAM.
The Case for DDR5 isn’t made out in this generation: DDR5 ram boasts impressive headline frequencies, but the timings show that it’s not a clear cut picture. This is borne out in testing, with DDR5 outperforming DDR4 in certain high memory bandwidth-intensive tasks like file compression and video rendering. DDR4, particularly well-specified faster DDR4, holds its own, particularly in gaming tests. Of course, there’s also the fact that DDR5 is three to four times as expensive as even high-quality DDR4 kits. We’re safe to say that for the 12th generation, you should opt for a DDR4 kit unless you have excellent reasons to need DDR5, and very deep pockets as well.
Once you’ve decided DDR4 is right for your Alder Lake Build, what then? Which is the best DDR4 RAM to choose from with this platform? We conducted a number of tests to find out the best RAM for the Intel Core i5-12400, i5-12600K, and Core i7-12700K for 2022 builds.
Testing Set up:
All of our testing was conducted on the MSI Tomahawk Z690 DDR4 motherboard (B09GLD72QH). We used an Intel i7-12700K (B09FXNVDBJ) at stock/default settings. This was paired with an EVGA RTX 3080 XC3 Ultra. Cooling was provided by an Arctic Liquid Freezer II 240 AIO (B07WSDLRVP). The power supply is a Fractal Design Ion+ 860W platinum PSU.
RAM used was:
- Crucial Ballistix 3200Mhx CL16 2x8GB, (XMP and JDEC) (B083TRRT16)
- G.Skill Ripjaws V 3600MHz CL16 2x8GB (XMP and JDEC) (B07X8DVDZZ)
- Patriot Viper Steel 4400Mhz Cl9 2x8GB (XMP and Manual OC) (B07KXLFDL6)
Firstly, let’s define a few terms so that we can interpret our results.
MHz speed is often the headline specification of RAM, but it’s somewhat misleading in that on its own it’s not particularly helpful in determining RAMs potential for performance. It’s the frequency the RAM runs at, but more correctly should be ‘MT/s’ or Million Transfers per Second, as RAM is Double Data Rate – it is accessible twice on every clock cycle. Therefore DDR4 3200Mhz RAM actually runs at 1600MHz, which is the speed you will see in various RAM diagnostic utilities. Faster RAM has the potential to perform better, as the rate of access increases. This primarily increases Bandwidth, and to a degree also lowers latency since the memory controller is waiting less time for an available access window to transact with the RAM. However, faster isn’t always better. Very high-speed RAM (greater than 4000MHz) is difficult for memory controllers to run, and may not be stable particularly in larger capacities. If you’re looking for 32GB or 64GB of RAM, you may find you need to reduce speeds for stability.
CL timings or CAS latency are the numbers you see quoted in listings for RAM: 16-16-16-32 or similar. These lists the primary timings and they are the number of clock cycles it takes to perform certain important functions. Lower numbers are therefore better. By multiplying clock speed with CL, you gain an indication of the ‘first word’ latency of the RAM in nanoseconds. Since it’s measured in time, again lower is better.
Gear 1 vs Gear 2. This is Intel’s naming for the ratio of your memory controller speed to the RAM speed. Ideally, for the lowest latency, it should be 1:1 and this is known as ‘Gear 1’. If your RAM frequency exceeds the stable speed of the CPUs memory controller, then ‘Gear 2’ may be employed, halving the memory controller speed in relation to the RAM. This induces a latency penalty. For Alder Lake, the memory controller appears to be happy in Gear 1 up to around 4000MHz, but individual CPUs may vary.
1. Synthetic Benchmarks
The CPU test in Time Spy is highly Multithreaded, using our 12700K to the full. As it turns out it’s also sensitive to memory speed. You can see here how detrimental to performance running 2400MHz CL16 RAM is – the score is 1,500 points down on the next set of results. This is what happens if you fail to set XMP, your RAM runs at these basic ‘JDEC’ defined speeds and that can really hurt performance. Both 3200Mhz and 3600Mhz CL16 RAM kits are commonly available and currently cheap. You can see they perform identically in this test returning 16,000 points each.
Looking at the ‘high performance’ RAM kit, this is a 4400 MHz RAM Kit, using Samsung B-Die chips, that we tweak to further improve performance.
Firstly, at XMP settings this kit runs at 4400Mhz CL19, but in gear 2. This, combined with the looser timings, offsets the higher frequency operation. Tweaking further brings us to a 4400Mhz CL17 settings and our third-highest result. Finally, Tightening timings to 4000Mhz Cl15-16-16-36 brings us to our best score of 17026, but it’s by a marginal 100 points. All three configurations perform near identically.
Rendering is another task that can be memory intensive, but is it dependent on memory speeds? We ran a couple of tests to check.
Cinebench R20 uses a tile-based renderer to complete a scene as fast as the CPU allows.
Ordered by result, we can see that there is essentially no trend here. 60 points are well within the margin of error for Cinebench R20 and we can see that the ‘worst’ RAM specification, 2400Mhz JDEC settings are butting up against one of the best: Manually tuned RAM at 4000MHz Cl17. 2600Mhz CL16 ties 3600Mhz Cl16 as well. This test doesn’t demonstrate any coherent scaling with RAM Speed. The lesson here is that many tasks simply aren’t dependent on RAM speed, and also that you shouldn’t use Cinebench to look for performance improvements if you’re tweaking RAM!
Cinebench R23 shows similarly unhelpful results. Again, this is all within the margin of error for this test and there’s no clear trend aligning with RAM performance.
And finally, another rendering test, this time Blender, a popular 3D Creation platform.
We rendered out two scenes on the i7-12700K, with the following results:
We’ve put a bit of a spoiler here in the form of some DDR5 testing as well. You can see that the results in all three instances, 2400Mhz, 400MHz optimised, and even DDR5 at 6000MHZ is the same: This is another task that does not scale with RAM speed to any appreciable degree.
2. Gaming Benchmarks
So, some synthetic tests show memory scaling, and some don’t. But what about gaming?
Well, games as a rule respond very well to reduced RAM latency on previous intel platforms. Getting data to the CPU with less delay allows it to process the game world faster, returning higher frame rates.
We can demonstrate this using a number of tests.
1. Shadow of the Tomb Raider
Firstly, Shadow of the tomb raider makes things easy for us by showing the effects of RAM speed on the game engine thread itself, in the in-game benchmark.
Using the min, average and maximum framerates, we can see a clear trend in performance as we move up to faster RAM specifications. Whilst JDEC specification 2400MHz RAM languishes behind at 202FPS average, the 3200MHz Ram improves to 227FPS, the 3600 MHz kit to 239FPS, the 4400 MHz kit to 244FPS, and the manually tweaked 4000 MHz kit run to 257 FPS on average. Note that minimum and Maximum frame rates increase accordingly.
Now, this is all well and good but is it representative of anything in the real world? Well, the overall FPS results of this benchmark confirm the Ram speed scaling and demonstrate the extent to which it can help:
Here, we can see that the 4400MHz Kit actually returns the highest results, with the 4000MHz RAM kits actually trailing slightly behind. The XMP 3600Mhz kit is also close to the peak result. You can again see the clear detriment of slow RAM, with 2400Mhz a clear 10% slower than any of the better options.
As for the reversal of fortunes for the 4000MHz manually tweaked ram? – Well, XMP has settings for a number of timings that we did not have time to optimise for this testing. The Render thread often acts as a limiting factor and depends more on Ram Bandwidth than latency. Overall, you can see the broad impact of tweaking RAM, but also the subtle nature of changes in timings and settings that can add noise to the results.
Nonetheless, this test demonstrates how faster RAM assists CPU performance and allows games to run with less restriction from CPU limitations, resulting in higher frame rates.
2. Rainbow 6 Siege
Again in Rainbow 6 Siege, we can see a broad trend favouring faster RAM, with 2400MHz the slowest at 500FPS, the 4000MHz and 4400MHz options posting similar results from 530-540 FPS average, and the 3600Mhz CL16 RAM kit posting the best results at 564 FPS. Again, XMP likely takes optimisations to tRFC and other secondary and tertiary settings that deliver more consistent all-round performance here. Nevertheless, such is the performance of the i7-12700K we’re really splitting hairs looking at RAM speed impacts on this benchmark.
3. Microsoft Flight Simulator 2020
Flight Sim is a Game, or simulator if you want to be serious about it, that places unique importance on the CPU performance. It’s often the limiting factor in frame rates in this game, and lifting CPU capability directly translates to more frames on screen in a lot of circumstances.
In this test, we fly a Daher at a low altitude across Manhattan and log frame rates for 3 minutes.
Here again, we see the detrimental effect using slow ram, or not setting XMP on your RAM, has on performance. 2400MHz languishes at 98 FPS. Faster RAM scales near linearly across the speeds on test, with 3600Mhz bringing a decent lift to 107 FPS, and 4400MHz and faster continuing to increase performance. Remember, with gear 2 and loose timings this 4400MHz kit isn’t the fastest RAM on test here. The effect is subtle, but the trend is clear: Faster RAM helps this title, which is unsurprising given the CPU heavy nature of performance limits in Flight Sim 2020.
4. Forza Horizon 5
This modern AAA open-world racer has a detailed inbuilt benchmark that lets us peek under the hood and see how performance is impacted.
Firstly, we can see that the sim engine itself benefits hugely from improving RAM speeds:
This is pretty unequivocal. Going from 2400MHz JDEC specification to 3600Mhz Cl16 nets you a huge 70FPS average gain. Pushing to optimised 4000MHz RAM at CL16 makes that a 110FPS margin, over a third faster again.
Once the game has processed the world, it prepares the render to pass to the GPU, this is the CPU render stat also shown in the results:
Here we see a less dramatic but still significant performance trend: The rendering process gains around 20 FPS or 10% going from the slowest RAM to moderately well optimised 3600Mhz Cl16 RAM.
And finally, do these under-the-hood numbers make any real-world difference? Well, they can do yes, as shown by our results:
Optimising RAM sees an easy gain of around 15FPS in the overall benchmark score.
Conclusions:
Throughout these tests, we’ve seen an overall trend of faster ram assisting performance, both in some synthetic tests and also in games. We hope you’ll excuse the somewhat ‘messy’ nature of the data. It is what it is, this is real data, and the nuances of RAM speed and timings and how they interact with different software performance is a complex web. There is no one best set of RAM for all circumstances, but you can certainly pick RAM that will give you better results more of the time. As our results consistently show, you can also pick bad RAM or fail to set it up correctly, and this can bring tangible harm to your system’s performance.
We’d also like to address the issue over the gaming benchmarks. We’ve purposefully run these tests at 1080p, High settings, to bring you data that is at least somewhat relevant. But we are still using an RTX 3080 at 1080p. It’s quite correct to say that at 1440p and up the limiting factor becomes the GPU making these results less relevant. However: What is important is that RAM speed sets a baseline for performance in many games. You buy yourself more headroom by choosing better RAM. And that’s important in demanding CPU bound situations or as a system ages. A system with bad RAM may well show performance problems where one with well-chosen RAM won’t, and that could mean the system as a whole survives another generation of graphics cards upgrades.
Alder lake CPUs are exceptionally high performance, and it would be a real shame to limit their potential with poorly chosen RAM.
Recommendations
Overall, on the basis of cost, availability and ease of set-up, we still recommend 3600Mhz CL16 RAM for Alder lake CPUs. This RAM brings you the bulk of performance improvements with a minimum of money and time invested. There are Crucial Ballistix and G.Skill Ripjaws V kits available in 2x8GB Format at under $100. They’ve got decent timings, and represent great bang for buck.
If you need 32 GB, around $150 to $180 gets you G.Skill or those Ballistix kits in 2x16Gb Format.
There are also currently Crucial Ballistix Max 4000Mhz Cl18 Kits available at around $90 for 16Gb. They also represent a nice option, as they run in Gear 1 mode and give roughly the same total latency as a 3600Mhz Cl16 kit. Grabbing those and tweaking timings downwards could yield some pretty impressive results, but they’ll work just fine at XMP.