6400MHZ CL32 or 6800MHZ CL34 RAM in Ryzen 7000 series

The AGESA 1.0.0.7B BIOS update unlocked higher memory speeds beyond 6000MHz for Ryzen 7000 series systems. But the question remains: is it advisable to push the RAM speed to these new limits?

In this study, I will be assessing whether PC builders should lean towards 6400Mhz or 6800Mhz kits by comparing the performance on the Ryzen 7000 series 7950X3D processor. The Kit I will be using for these tests is the Kingbank SharpBlade DDR5 6400Mhz and 6800Mhz. 

I would like to thank Venus Protocol for supplying us with a grant to get things moving along and KingBank for sending us a review kit of the 6800Mhz Ram. 

1. MCLK (Memory Clock):

• Measured in MHz, it represents the actual speed at which your RAM operates.
• The MCLK half of the advertised RAM speed (6000MHz RAM will have an MCLK of 3000MHz) as DDR5 operates in dual channel.

2. FCLK (Infinity Fabric Clock):

• Measured in MHz, it acts as the communication channel between the various parts of the Ryzen 7000 CPU.
• Think of it as a highway connecting the CPU cores, the IO die (which contains the memory controller), and other internal components.
• Optimal performance is achieved when the FCLK matches the MCLK (1:1 ratio), but this isn't possible on the Ryzen 7000 series where it runs async at 2000Mhz

3. UCLK (Unified Memory Controller Clock):

• Measured in MHz, it acts as the bridge between the RAM and the rest of the CPU via the Infinity Fabric.
• You want the UCLK to be a multiple of the MCLK (e.g., 2:1, 3:1) for optimal performance.

4. Timings:

• Represent the latency between different RAM operations, measured in clock cycles.
• Lower timings indicate faster response times and better performance.

Some Theory

AMD's Ryzen CPUs are made up of core complexes called CCDs and/or CCXs and their design is commonly referred to as chipsets. The 7900(X/3D) / 7950(X/3D) have a dual-CCD design with each CCD housing 8 cores (6/6 or 8/8 cores enabled respectively). Dual CCD designs could potentially allow for greater overclocking of the FCLK.

The Default FCLK on Ryzen 7000 series CPUs is set to 2000 MHZ. For this reason, 6000MHz RAM is typically recommended for optimal performance due to being able to sync the clocks (MCLK:UCLK=1:1, FCLK=MCLK/1.5), Faster RAM could be run if you desync the MCLK & UCLK but there is a chance you could sync on 6200MHZ and 6400MHZ.

• Kingston Fury Renegade 2TB NVME
• Ryzen 9 7950X3D
• 32GB Kingbank DDR5 Sharpblade Ram
• MSI B650m Mortar (BIOS: 7D76vAB1)
• Acer Intel Arc A770 GPU
• Be Quiet! Dark Power 13 850W

Ram Configuration

I have set the UCLK to 2133MHz, as it runs stable on my Ryzen 9 7950X3D and the MCLK = UCLK and the FCLK = MCLK/1.5 which makes it ideal for 6400Mhz.

• 6400Mhz
• tCL: 32
• tRCDWR: 39
• tRCDRD: 30
• tRP: 39
• tRAS: 80
• tRC: 119
• MCLK: 3200 Mhz
• FCLK: 2133Mhz
• UCLK: 3200Mhz
• SOC Voltage 1.30V
• 6800Mhz 
• tCL: 34
• tRCDWR: 45
• tRCDRD: 45
• tRP: 45
• tRAS: 108
• tRC: 153
• MCLK: 3400 Mhz
• FCLK: 2133 Mhz
• UCLK: 1700 Mhz
• SOC Voltage 1.15V

I have set the Memory Timing Preset to tighter in the MSI Bios when running at 6800MHZ and at balanced when at 6400MHz as this is simple to do and automatically lowers the sub-timings for better latency and performance. I did not experience any stability issues with these when testing on AIDA64 and saw noticeable gains in Cache and memory Benchmark. 

When running at 6800Mhz UCLK will have to be run at half the MCLK speed. I tried a 1:1 ratio but the System will not post. This is because the UCLK tops out at around 3000Mhz. I managed to push the UCLK up to 3200Mhz but 3400Mhz just was not possible so UCLK = /2 MCLK.

I undervolted the SOC to 1.15v when I was running at 6800MHZ as the UCLK was forced into reduced speed which allowed the system to maintain stability at 1.15V. When running at 6400MHz I left it on auto which set it to 1.3V which was necessary as the UCLK was running faster.

This investigation compares the performance of 6400MHz and 6800MHz DDR5 memory on the AMD 7000 series AM5 platform. Does higher clock speeds at a non-ideal ratio translate to superior performance, or will a 6400MHz in a synchronised 1:1 ratio keep up or even outperform the async 6800MHz counterpart?

I have stuck to basic RAM overclocking for this experiment leaving potential performance on the table as getting deep into manual timing adjustments isn’t for everyone. 

Read Performance:

• Average Read Speeds:
• 6400MHz: 92888 MB/s
• 6800MHz: 95306 MB/s
• Note: The 6800MHz RAM exhibited variability, with two runs around 92000 MB/s and a third at 100000 MB/s, influencing the higher overall average.

Write Performance:

• Average Write Speeds:
• 6400MHz: 95115 MB/s
• 6800MHz: 97822 MB/s

Copy Performance:

• Average Copy Speeds:
• 6400MHz: 82131 MB/s
• 6800MHz: 81905 MB/s
• Note: Results across three runs were consistent, showing marginal speed differences.

Latency Performance:

• Observation: Clear disparity in latency attributed to the UCLK running at half the MCLK speed at 6800MHz.
• Average Latency:
• 6400MHz: 61.03
• 6800MHz: 69.06
• Explanation: The additional latency at 6800MHz is due to the UCLK processing memory readings being out of sync with the faster MCLK, causing a latency penalty.

These results indicate significant variations in read and write speeds between the RAM speeds, while copy speeds show negligible differences. The noticeable increase in latency at 6800MHz is explained by the UCLK's synchronization disparity with the faster MCLK, resulting in the observed penalty.

This has changed a lot since its first appearance. Instead of running one render it now runs for 10 minutes minimum and renders as much as it can and upon completion, it resets calculating the average score in real time as it progresses.

• Scores
• 6800MHz: 1914 points
• 6400MHz: 2037 points

There's a 123-point disparity favouring the 6400MHz RAM, contrary to expectations of the faster RAM yielding a higher score.

This was an extended 8GB Benchmark that I ran 5 - 8 times and got the average.

• 6800MHz Scores
• Average: 727.67 GFlops
• Maximum: 730.37 GFlops
• 6400MHz Scores
• Average: 723.31 GFlops
• Maximum: 737.63 GFlops

Linpack Xtreme x64 indicates relatively similar performance with marginal differences favouring either speed in specific metrics.

The Gaming Benchmark has a focus on my configuration and I will be running these are the Lowest preset and the Highest preset at the native resolution. The point of this experiment is to see if I will receive any performance differences between 6400Mhz and 6800Mhz RAM. 

One big bottleneck is my Intel Arc A770 Graphics Card but this offers reasonable performance for the money and I don’t think many of will be rocking at RTX 4090 RX7800XTX.

RAM prices have been getting cheaper and there isn’t much difference between 6400 or 6800MHz kits but there is a huge price difference for GPU performance.

A game I’ve previously completed, it looks and runs well. The graphical difference between Low and Ultra High is noticeable, Low kinda makes the textures look like Zelda so a good, playable balance at 60FPS would be ideal for me.

Ultra High Preset:

• Average FPS: 61 (6800MHz) / 61 (6400MHz)
• Max FPS: 194 (6800MHz) / 204 (6400MHz)
• Min FPS: 37 (6800MHz) / 36 (6400MHz)

Low Preset:

• Average FPS: 92 (6800MHz) / 90 (6400MHz)
• Max FPS: 114 (6800MHz) / 119 (6400MHz)
• Min FPS: 62 (6800MHz) / 60 (6400MHz)

Very similar performance between the two RAM kits across both Ultra High and Low presets with minimal discernible difference in FPS.

This game looks amazing and it is one I’ve started but never finished, I want to finish this. It runs well on the Intel Arc A770

Highest Preset:

• Average FPS: 84 (6800MHz & 6400MHz)
• Max FPS: 132 (6800MHz) / 120 (6400MHz)
• Min FPS: 66 (6800MHz & 6400MHz)

Lowest Preset:

• Average FPS: 151 (6800MHz & 6400MHz)
• Max FPS: 235 (6800MHz) / 234 (6400MHz)
• Min FPS: 128 (6800MHz) / 121 (6400MHz)

Both kits showcase almost identical performance, with a slight improvement in minimum FPS on the 6800MHz kit. The average FPS remains consistent.

This is the Newest Game on the list and especially the most demanding game. It is tough to run at 60FPS+ on arc A770 even at the lowest preset without resorting to resolution scaling. I am not sure why that is, I don’t think it's the best-looking game here but maybe a lot is happening in the Avatar world.

Ultra Preset:

• Average FPS: 29 (6800MHz & 6400MHz)
• Max FPS: 66 (6800MHz & 6400MHz)
• Min FPS: 66 (6800MHz & 6400MHz)

Low Preset:

• Average FPS: 45 (6800MHz & 6400MHz)
• Max FPS: 33 (6800MHz & 6400MHz)
• Min FPS: 36 (6800MHz) / 35 (6400MHz)

No performance difference between RAM kits; the game's performance is low and it is primarily being constrained by the GPU, leading to similar results.

This is an intense Benchmark with the Enhanced Edition of the gaming coming with RayTracing support. There is a noticeable stutter during the benchmark and the GPU’s coilwine would sutter with it which was fascinating to see. This might be a GPU-related issue, the Intel Arc A770 isn’t the most well-optimised GPU but it may even be the Ryzen 9 CPU. I’ve seen suggestions on the forums on how to address this but for this review, I decided to leave everything as is.

Extreme Preset:

• Average FPS: 42.41 (6800MHz) / 41.55 (6400MHz)
• Max FPS: 105 (6800MHz) / 103.53 (6400MHz)
• Min FPS: 18.16 (6800MHz) / 19.76 (6400MHz)

High Preset:

• Average FPS: 77.54 (6800MHz) / 75.94 (6400MHz)
• Max FPS: 131.93 (6800MHz) / 119.59 (6400MHz)
• Min FPS: 22.73 (6800MHz) / 16.21 (6400MHz)

The faster 6800MHz RAM exhibits slightly better performance, especially noticeable in the High Preset with higher average, maximum and minimum FPS. The stuttering persists in both configurations.

On Ultra this game looks out of this world and it runs pretty well on the ARC A770. With Ray Tracing on it looks unbelievably good but for this review, I have kept it off as it defaults to off in the ultra preset. 

Ultra Preset:

• Average FPS: 91 (6800MHz) / 93 (6400MHz)
• Max FPS: 115 (6800MHz) / 116 (6400MHz)
• Min FPS: 64 (6800MHz) / 62 (6400MHz)

Low Preset:

• Average FPS: 102 (6800MHz & 6400MHz)
• Max FPS: 130 (6800MHz) / 131 (6400MHz)
• Min FPS: 66 (6800MHz) / 69 (6400MHz)

Both RAM speeds show very similar results, with slight differences in FPS with the 6400MHz taking the edge.

This is quite a long benchmark at around 3 minutes and once started you cannot cancel out. RayTraced Vehicle Shadows and VRS were turned off with Dynamic Resolution Scaling turned on. The rest was at the default Ultra Low and Very High presets.

Ultra High:

• Average FPS: 56.3 (6800MHz) / 56(6400MHz)
• Max FPS: 66 (6800MHz) / 65.1 (6400MHz)
• Min FPS: 50 (6800MHz) / 42.1 (6400MHz)

Low Preset:

• Average FPS: 207.8 (6800MHz) / 208.7 (6400MHz)
• Max FPS: 230.3 (6800MHz) / 230 (6400MHz)
• Min FPS: 191.3 (6800MHz) / 184.7 (6400MHz)

Marginal differences were observed in minimum FPS for the 6800MHz kit, while the average FPS remains consistent between both RAM speeds. The Ultra High preset shows a slightly higher minimum FPS with 6800MHz RAM.

Overall Analysis:

The results here can be improved by tightening the timings but this is a long, and usually trial-and-error process. For the system builders that want minimum memory configuration the read/write and copy speeds are very similar to each other and yes you can run faster memory than 6000MHz on AM5.

Running the EXPO profiles I can see that the latency is high and it's higher on the 6800Mhz kit. This is attributed to me running the UCLK, MCLK and FCLK async, although it works it isn’t typically recommended because of this latency penalty.

Being able to run the UCLK at /2 MCLK with the 6800MHz kits allowed me to under-voltage the SOC on the CPU to 1.15v. This could lower your power consumption and having a lower SOC v could increase the lifespan of the CPU. 1.15v SOC was unstable at 6400MHz (3200MHZ UCLK) but fully stable when running at 7800MHz (1700MHz) UCLK.

Ryzen is said to prefer lower latency, especially in gaming, but higher RAM latency on the X3D CPUs is reportedly less of a bottleneck and my results demonstrate this. This could explain why the performance difference with the higher latency 6800MHz kit appears marginal across most games and presets.

The GPU seems to be the primary bottleneck in achieving higher frame rates, indicating that the RAM speed does not significantly impact performance in these configurations.

With the price of raming coming down and there being little cost difference between 6000,6400 and even 6800MHz kits deciding what to do for can be tough.

After my research and analysis, the 6400MHz CL32 kit emerges as a particularly promising choice. Its potential to harmonize with a 1:1 MCLK:UCLK ratio, accompanied by an overclocked FCLK of 2133MHz, presents an enticing equilibrium between speed and latency. Notably, the Ryzen 9 and Ryzen 9 X3D models stand a higher chance of attaining this optimal configuration, enabling FCLK at 2133MHz and UCLK at 3200MHz. In cases where 6400MHz doesn't operate at a 1:1 ratio, configuring it at 6000MHz should pose minimal challenges, especially given the SK Hynix A die chips utilized in these KingBank RAM Sticks.

For those inclined toward a conservative approach, the AMD-recommended 6000MHz remains a robust and safe option.

My ongoing efforts will concentrate on refining RAM timings to achieve the lowest attainable latency, with forthcoming results to be shared in a subsequent article, enhancing our understanding of optimal RAM utilization.

Deliberating on the prospects of 8000MHz RAM unveils intriguing theoretical advantages, notably in synchronizing UCLK with the default FCLK of the Ryzen 7000 series CPUs. However, achieving stable operation at 8000MHz hinges significantly on the interplay between the CPU, motherboard, and RAM modules. This presents a formidable challenge, and any progress made in this pursuit will be promptly shared and updated.

Create your eSports profile on our website and stay tuned for exciting virtual tournaments. Incredible prize pools await the champions. Don't miss out by signing up today!

On launch the game warns me that the GPU driver is outdated however I am using the Latest Driver from Intel so I simply ignored this message. I experienced no issues with the game during benchmarking.