Impact of DDR5 Speed on Content Creation Performance

Image

×

Introduction

When choosing parts for a new PC, there is a massive amount of choice depending on not only the type of part, but also the brand and specific model. This is bad enough for fairly straightforward parts like the CPU and GPU, but it is far worse for system memory (RAM), which is arguably the most confusing, and often the most misunderstood, component.

What makes choosing the right RAM especially difficult is that it is extremely easy to unintentionally use RAM that is beyond what your CPU is rated for. Just like overclocking your CPU, using RAM that is outside spec can work, but in our experience, it increases the chances that your system will be unstable.

Still, this wouldn't be too difficult if there was just a single RAM frequency you should stick to based on your choice of CPU, but it is becoming more and more common that a CPU may actually support a wide range of frequencies depending on a number of factors including how many RAM slots your board has, how many RAM sticks you are using, and whether those sticks are single or dual-rank. Even worse, neither AMD nor Intel are very transparent about what the supported RAM speeds actually are!

We've been dealing with this with AMD for a while, starting with their 2nd generation Ryzen processors. And just like AMD, the spec pages for CPUs like the Intel Core i9 12900K simply give a less-than-helpful spec of "Up to DDR5 4800 MT/s". Luckily, Intel does have more information available to the public, but you have to hunt for it on page 95 of the 12th Generation Intel Core Processors Datasheet – Volume 1.

If you don't want to download that PDF, below is our conversion to what we feel is the most straight-forward interpretation of what RAM speeds the 12th Gen Intel processors support depending on the RAM configuration:

Given how new DDR5 is, it is unknown how important it is to stick exactly to these specifications. We have found that as DDR4 has matured, it has become safer and safer to go slightly out of spec when choosing RAM for AMD's Ryzen CPUs, but with how new DDR5 is, it is likely a good idea to try to adhere to these specs as closely as possible. It may be perfectly safe to use 4400MHz RAM when using two RAM sticks in a four-slot motherboard, but it is going to take time to determine the impact of going slightly beyond spec like that.

And in fact, even in the testing for this article we found several instances where being too far outside of spec resulted in the system being unstable, or even not being able to POST at all.

However, in this article, we are going to largely ignore whether or not these specifications are overly conservative or not, and focus primarily on exactly how much performance you may have to give up if your system configuration limits you to lower RAM speeds like DDR5-4000. After all, if we see minimal or no difference at lower RAM speeds, then you lose nothing by adhering to Intel's specifications, but theoretically should end up with a more stable system.

Content Creation Workstations

Puget Systems offers a range of powerful and reliable systems that are tailor-made for your unique workflow.

Configure a System!

Labs Consultation Service

Our Labs team is available to provide in-depth hardware recommendations based on your workflow.

Test Setup

Listed below are the specifications of the system we will be using for our testing:

For many of these applications, we will be using our PugetBench suite of benchmarks, while others like CineBench and V-Ray are using the official benchmarks provided by the software developers.

For the RAM itself, we are using very standard sets of RAM from Crucial. This RAM is not overclocked and doesn't even support features like XMP. Our main goal in this testing is to see how performance changes when sticking as closely to the official specifications as possible, and that includes staying true to the JEDEC RAM specifications as well. The JEDEC frequencies and timings we tested are:

• DDR5-4800 CL40
• DDR5-4400 CL36
• DDR5-4000 CL32

As for the RAM speeds we are going to test, we initially planned on going all the way from 3600MHz CL30 to 4800MHz Cl40 with both 2x32GB and 4x16GB, but we ran into a number of issues that reduced the number of configurations we could test.

First, we were unable to get our system to even post when the RAM was set to DDR5-3600 CL30. It may be that we are simply going too far outside of what the DDR5-4800 RAM we are using is intended to run at, but we have been unable to find any DDR5 RAM for sale that is actually advertised to run at anything close to 3600MHz.

Second, we had increasingly severe stability issues with the 4x16GB configuration when running beyond DDR5-4000. 4400MHz was mostly stable, with the system only crashing when running our Unreal Engine tests. 4800MHz, however, was very unstable, resulting in immediate Windows bluescreens, and numerous errors in Memtest. We tried increasing the voltage to the RAM manually, and switching to a set of similar Kingston RAM, but were unable to get the system stable enough to run our performance tests. This is likely to get better as DDR5 matures, and higher-end RAM using XMP profiles would probably be able to hit these speeds with greater stability, but with the tight supply of DDR5 RAM at the moment, we had to make do with what we have and restrict our 4x16GB testing to 4000MHz and 4400MHz.

Adobe Photoshop

Image

×

Starting off with Adobe Photoshop running our PugetBench for Photoshop benchmark, the largest performance difference we saw between RAM speeds was just 1.6%. That is arguably well within the margin of error for a real-world test like this and is a clear indicator that RAM speed doesn't matter much with DDR5 RAM when sticking to JEDEC specs.

In fact, what is interesting is that we actually saw a larger performance difference between the 2x32GB and 4x16GB configurations than we did when looking solely at how the RAM speed affected performance. Even sticking within the official specifications, the 4000MHz 4x16GB configuration was 4% faster than the 4400MHz 2x32GB configuration. Even upping the 2x32GB configuration to 4800MHz (to match what a 2-slot board fully supports), the 4000MHz 4x16GB RAM setup is still 2.4% faster.

This is an indicator that having more RAM sticks in total is more important than the speed that the sticks are running at for Photoshop, and it will be interesting to see if this continues for the other applications we are testing.

Adobe After Effects

Image

×

Next up is Adobe After Effects, running our PugetBench for After Effects Benchmark. Here, the RAM speed makes a bigger difference than it did in Photoshop, but it is still fairly minimal.

When going from 4000MHz to 4400MHz with both sets of RAM, we only saw about a 1% increase in performance, and when going from 4000MHz to 4800MHz with the 2x32GB configuration, we saw a 3.5% increase in overall performance. However, this still isn't much and is again well within the margin of error for this kind of test.

Unlike Photoshop, we didn't see a significant increase in performance when using four sticks of RAM. Four 16GB sticks at 4000MHz were about 2% faster than two 32GB sticks at 4400MHz, but compared to two sticks at 4800MHz, the difference was less than 1%.

Adobe Premiere Pro

Image

×

Adobe Premiere Pro rounds out our Adobe-specific testing, and here we are running our PugetBench for Premiere Pro benchmark. Overall, the results are very similar to the Photoshop results. Between 4000MHz and 4800MHz, the largest difference we saw was just 2%, which is small enough to make all the results effectively the same.

Also similar to Photoshop, we again saw a larger performance difference when simply using more sticks than we did with the different RAM speeds. Here, 4x16GB at 4000MHz was about 2.5% faster than 2x32GB at either 4400MHz or 4800MHz. Once again, however, this is well within the margin of error, so we wouldn't recommend reading too much into a single result like this. It is an indicator of a slight trend, but it is so minimal that it likely isn't going to be a significant factor.

DaVinci Resolve Studio

Image

×

With our PugetBench for DaVinci Resolve benchmark running DaVinci Resolve Studio, we see more of the same with DDR5-4800Mhz resulting in at most just a 2% improvement in performance over DDR5-4000MHz.

We also once again saw a larger benefit from having more total RAM sticks than the speed of the actual sticks. This time, 4000MHz 4x16GB was 2% faster than 4800MHz 2x32GB, and almost 3% faster than 4400MHz 2x32GB.

Unreal Engine

Previous Next

System Image

×

Previous Next

The results for Unreal Engine change quite a bit depending on the specific workflow, so here we decided to split it up into three charts. Something to note is that, unfortunately, the system was too unstable with 4 sticks of RAM at 4400MHz, so we were only able to test four sticks at DDR5-4000MHz.

Compiling shaders is a task that every user will experience to one degree or another in Unreal, and here, RAM speed does have an impact on performance. Going from 4000MHz to 4400MHz resulted in a 5% performance improvement, while 4800MHz was 9.4% faster than 4000MHz. This makes sense as this task is basically compiling thousands of mini-programs, so there is a large amount of data being transferred into and out of RAM. Conversely, baking lighting (chart 2) sees no difference in RAM speed – although it does see a slight bump with 4 sticks instead of 2.

Lastly, RAM speed had virtually no impact on the average FPS across the four scenes we tested. The only time there was a difference in FPS was in a scene that was CPU limited, which saw a 4% increase in FPS, going from 450 FPS to 470, which is not exactly a difference anyone would notice.

CineBench R23

Previous Next

System Image

×

Previous Next

CineBench R23 has two modes to examine CPU rendering performance within Cinema4D: single-core and multi-core. Unlike our previous tests, the difference between all the configurations we tested was less than 1% in all situations. Really not much to say here – RAM speed seems to make no difference in CineBench.

V-Ray CPU Rendering

Image

×

Rounding out our testing is the V-Ray 5 benchmark. This benchmark has a variety of modes that tests CPU and GPU rendering performance, as well as a combination of the two. Out of all the modes, the straight CPU mode showed the biggest difference in performance, but even that was less than 1% across each of the RAM speeds we tested.

Interestingly, unlike CineBench, we did see a very small increase in performance with the 4x16GB configurations compared to the 2x32GB configuration. The difference was only 2% at most, but this is once again more than what we saw when looking at different RAM speeds with the same number of sticks.

Does DDR5 RAM Speed Matter for Content Creation?

If you average the results from all of our tests, we found that the difference between DDR5-4000MHz CL32 and DDR5-4800 CL40 was just 2.0%, which is small enough that it isn't worth worrying about in most cases. There was only a single instance (compiling shaders in Unreal Engine) where the speed of the RAM made a significant difference, with 4400MHz and 4800MHz being 5% and 9% faster than 4000MHz respectively.

Except for the one Unreal Engine test, the results from our tests were well within the margin of error for the kinds of real-world benchmarks we ran. This means that if you are sticking to the JEDEC RAM specifications for frequency and timings, for content creation, you are very unlikely to see any significant performance difference between the supported DDR5 RAM speeds.

Of course, it is very possible – and even likely – that using even higher speed RAM like 5200MHz or 5600MHz with tighter timings could give a small performance boost. But, the goal of this testing was to determine if the various RAM speeds supported by the Intel 12th Gen CPUs would have any impact on performance in these applications and workflows. And for that specific question, we can very conclusively say that outside a few very specific circumstances, RAM speed doesn't matter.

In our opinion, this solidifies our current recommendation of following the official Intel specifications for RAM speed. Especially since we had significant stability issues when trying to run four sticks at 4800MHz, we highly recommend using 4000MHz when using four sticks of DDR5 memory with the Intel 12th Gen processors, or 4400MHz CL36 / 4800Mhz CL40 when using two sticks.

Custom Workstations

Puget Systems offers a range of powerful and reliable systems that are tailor-made for your unique workflow.

Configure a System!

Labs Consultation Service

Our Labs team is available to provide in-depth hardware recommendations based on your workflow.