I can only dream of 25 per cent more performance: memory timings tried out
16.4.2020
Translation: machine translated
RAM timings determine how quickly the memory responds to requests and executes tasks. With a little fine-tuning, you can tease more performance out of your modules. How much depends on the individual case.
"For the average user or gamer, it is perfectly sufficient to set the correct XMP in the bios." That was the conclusion of my explanatory article on memory timings. User Annihilator disagreed with me:
you should rather ignore such articles that show zero basic knowledge... and yes, I have meticulously tightened timings and depending on the application and/or game there are even almost 25% extra to 3800 with XMP timings.
25 per cent? I want to try that too. The tool that should help me achieve tighter timings is the "DRAM Calculator for Ryzen 1.7.0". I don't want to change the megatransfers per second (MT/s) and the Infinity Fabric clock speed (FCLK) (for now). I just want to see what tighter timings bring me in contrast to the JEDEC settings and the XMP (Extreme Memory Profile).
Explanation: JEDEC is the standardisation association that defines timings for RAM. If you do not change the RAM settings in the BIOS, your system will most likely boot with these settings. There is also XMP (Extreme Memory Profiles) - also called DOCP (Direct Over Clock Profile) or EOCP (Extended Over Clock Profiles) by some motherboard manufacturers. This is information on memory timings and megatransfers with higher performance than that of JEDEC. This information comes directly from the RAM manufacturers.
I am carrying out the test on my PC, in which the following components are installed:
I am aware that I would probably achieve tighter timings if I only used two RAM bars instead of four. This is because my motherboard uses Daisy Chain and not T-Topology. Daisy chain motherboards are more suitable for two RAM bars instead of full configuration. It is better to use a T-Topology board for a full configuration. However, as I need more RAM for certain programmes and am unfortunately not sitting on a goldmine, I currently rely on four bars.
DRAM Calculator for Ryzen
I'm not a professional overclocker and therefore don't have an overview of all the memory timings. Many people are probably like me. Designations such as tCL, tRCDWR, tRCDRW not only sound cryptic, their explanations are also difficult to understand.
This is where the DRAM Calculator comes in. It automatically calculates the appropriate timings based on your RAM. You can choose between the three profiles Safe, Fast and Extreme. For the test, I will try the Safe and Fast profiles. According to the programme, the Extreme profile requires additional RAM cooling.
The tool also gives specific recommendations on the appropriate voltage values for the memory (DRAM Voltage) and the SoC (SoC Voltage) as well as the required resistors. Let's see if it works.
I'm following these instructions from Youtuber JayP Tek.
Test methodology
As it always depends on the use case whether the timings actually help you, I run four benchmarks: the AIDA Cache and Memory Benchmark, Cinebench R20 (Multi Core), the Riverside Benchmark from "Wolfenstein: Youngblood" and the slightly more CPU-intensive Gathering Storm graphics benchmark from "Civilization VI". Both games run in 1440p. In "Wolfenstein: Youngblood" I use the "My Life!" presets. DLSS and RTX are activated with the "Quality" preset. In "Civilization VI" I set all graphics settings to maximum.
I run all benchmarks three times to eliminate errors and calculate an average value rounded to whole numbers. As the RAM with the JEDEC standard only runs at 2400 MT/s, I also set it manually to 3200 MT/s so that the JEDEC timings at least have a fair basis. My RAM is one with 3200 MT/s. This makes a total of five different settings: JEDEC, JEDEC with 3200 MT/s, XMP, DRAM Calculator settings Safe and DRAM Calculator settings Fast.
On JEDEC standard, the RAM runs in addition to 2400 MT/s with the following timings: 17-17-17-39. With the JEDEC settings and 3200 MT/s with 22-22-22-52. The DRAM Calculator calculates the following settings for the Safe profile:
For the Fast profile, the timings are as follows:
The results
Without further ado, here are the results of the benchmarks:
As you can see, the results of the DRAM-Calculator defaults are almost missing. Unfortunately, I didn't even make it through the POST (Power On Self Test) process with these defaults. Maybe it's the daisy-chain topology, maybe it's the RAM or maybe it's the mainboard. I have no idea why it didn't work. I tried different settings for a few hours, but nothing worked. Do you have any idea what the problem could be? Write it in the comments column.
What did it achieve apart from wasting time with the Fast profile? The differences between the safe settings of the DRAM-Calculator and XMP are minimal. There is almost no difference in the games. Even the JEDEC standard with 3200 MT/s can keep up. This may be due to the fact that the graphics card handles the majority of the calculations at 1440p. At 1080p, the CPU and therefore the RAM is used more.
For fun, I run the benchmark in "Civilisation VI" in 1080p with Safe profile, XMP and JEDEC at 3200 MT/s. I achieve 100 FPS with XMP, 99 FPS with JEDEC standard 3200 MT/s and 101 with the Safe profile. The performance increase is one per cent compared to XMP. I'm still a long way from User Annihilator's 25 per cent
In Cinebench, the difference is not even one per cent. The differences are also very small in the AIDA64 memory benchmark. The differences are only more noticeable when it comes to copying and latency.
What about stability? The safe settings work, but not always. Occasionally the PC stalled during POSTing and switched back to the JEDEC standard. Maybe it's because I'm using four bars and not just two.
I'll repeat the game with two bars instead of four.
Even two doesn't help
With the Fast profile I have the same problem with two instead of four RAM bars. I can't get through the POST process. Nevertheless, the computer boots perfectly with the Safe settings. I run all the benchmarks again with the JEDEC standard at 3200 MT/s, XMP and the safe settings. I leave the JEDEC standard without any changes. The first test showed that it is much slower.
The differences in the benchmarks are also negligible here. What is evident, however: The latency is lower, but the read, write and copy speeds are slower with two instead of four bars through the tape.
I'm sticking with XMP - for now
In my case, the performance gain from tighter timings with the DRAM Calculator is devastatingly small. Regardless of whether with two or four RAM bars. I can only dream of 25 per cent, as user Annihilator writes. I can't achieve any more with my configuration. That's why I'm sticking with XMP and my four bars for the time being. The test showed me that the system doesn't run any faster with two bars, but it is more stable. As soon as I can afford it, I'll try it with two 16 GB bars - and correspondingly more overclocking-friendly RAM.
Why didn't I manage more? I can see two reasons for this: My RAM is aimed more at beginners. It is therefore to be expected that not much can be done. On the other hand, my mainboard could be the reason. It's not designed for a lot of gaming either. Where do you think the reason is that I didn't achieve more? Was it my inability? Write it in the comments column.
Kevin Hofer
Senior Editor
kevin.hofer@digitecgalaxus.chFrom big data to big brother, Cyborgs to Sci-Fi. All aspects of technology and society fascinate me.
