Intel & AMD Dual-Platform DDR5 Memory Performance Measurements - High Frequency, Low CL Value

Various memory module brands will launch various frequency (e.g. DDR5 6000~8000 MT/s) specifications or different CL value specifications under the same frequency for sale. Basically, high-end models will pursue optimization and advancement in both directions, but both high frequency and low CL value will further increase the product price because not every batch of chips can meet the requirements of high frequency or low CL value at will. Not every batch of chips can meet the requirements of high frequency or low CL value, not to mention that models that meet both high frequency and low CL value will be even more expensive. This time, we will conduct performance tests on both Intel & AMD platforms to provide you with the performance gap of the memory itself for your reference on how to choose, and the following results are only representative of the results of the tests on my platforms.

The Four Emperors appear! Test Memory and Software

This memory test "does not enable the motherboard's memory score optimization settings" or manual overclocking adjustments, so as to present the performance of directly applying the memory profile, so that there will be no motherboard factors interfering with the results, and the motherboard's Gear mode and crossover mode are kept automatically for the motherboard to adjust, and of course, will not be turned off through the E-Core or the use of the Thin Overclocking version of the system to obtain the best scores. Of course, there is no need to disable E-Core or use a thinned down overclocked system to get the best results, but rather, the test is conducted purely on a normal Win11 system with the full default processor, which is more in line with the direct usage scenario.

This article does not include games or professional rendering software because there are hundreds of games and professional software that everyone can play, so the author cannot test each and every one of them, or use one game or software to represent all of them. Therefore, it is recommended that you look at the commonly-used software to see if there are any changes in memory bandwidth performance or latency, and then refer to it with the performance of this article.

∆ This test will not turn on the memory optimization settings in the motherboard. Turning on such settings will make the score look good, but it will have the effect of overclocking of the motherboard, this figure shows the result after turning on the relevant overclocking optimization.

 

This time prepared four sets of Chi Chi G.SKILL Trident Z5 Royal Royal halberd and Trident Z5 Royal Neo Royal halberd EXPO Edition memory module for real test, detailed memory specifications are as follows, the author this time mobilized the existing or borrowed from a friend of four different specifications Royal halberd performance test, but because the production batch and the use of the particles of the lot may be different, and the performance may be different. But because the production batch and the use of particles batch may be different, and lead to the performance may be different, this test results only represent the author's hand hardware and platform, and measured results and conventional logic is very different, rational consumption.

1. Chipset Royal Halberd EXPO Edition G.SKILL Trident Z5 Neo RGB DDR5 6000 MT/s 32GB (2x 16GB) CL 26-36-36-96 1.45V: F5-6000J2636H16GX2-TR5NS(AMD EXPO Profile)
2. Chipset Royal Halberd EXPO Edition G.SKILL Trident Z5 Neo RGB DDR5 6000 MT/s 32GB (2x 16GB) CL 28-36-36-96 1.4V: F5-6000J2836G16GX2-TR5NS (AMD EXPO Profile)
3. Chipset Royal Halberd G.SKILL Trident Z5 RGB DDR5 6400 MT/s 32GB (2x 16GB) CL 32-39-39-102 1.4V: F5-6400J3239G16GX2-TR5S (Intel XMP 3.0 Profile)
4. Chipset Royal Halberd EXPO Edition G.SKILL Trident Z5 Neo RGB DDR5 8000 MT/s 48GB (2x 24GB) CL 40-48-48-128 1.4V: F5-8000J4048G24GX2-TR5NS (AMD EXPO Profile)

∆ This time, we prepared four sets of Chi Chi Royal Halberd for comparison, 6000 CL26 (2x 16GB_EXPO), 6000 CL28 (2x 16GB_EXPO), 6400 CL32 (2x 16GB_XMP), and 8000 CL40 (2x 24GB_EXPO).

 

As long as the memory module has a higher frequency or a lower CL value, the price will be higher than the lower frequency or higher CL value, while the larger the capacity, the higher the price, the capacity is based on one's own needs to decide, but the frequency and the CL value is a relatively ambiguous use of the demand for the specifications, for example, many people would say that the AMD AM5 platform is more or less the same performance as long as the frequency exceeds 6000 MT/s. I am curious if this is really the case. For example, many people would say that AMD AM5 platforms with a frequency of over 6000 MT/s would be almost the same in terms of performance. Is there any difference if the CL value is low?

For price comparison, the prices of memory modules from 1 to 4 are NTD 7290, 5490, 4899, and 8990 respectively, based on the "G.SKILL Officially Authorized Flagship Store" on Shrimp Pi at the time of the author's writing.

However, the DDR5 6400 MT/s 32GB (2x 16GB) CL 32 specification was only tested on the Intel platform, as the author sent the memory in for repair after testing the Z890 first (the light bar was loose), and the memory had not been sent in for repair when I switched to the X870 platform, so the AMD platform was skipped over for the 6400 MT/s test.

This test software uses AIDA64 Cache & Memory Benchmark It is mainly used to test the performance of processor caches and RAM. The read/write/copy bandwidth performance score represents the rate of transfer between CPU and memory, which means the efficiency of data throughput (the higher the score, the better), and the time delay score of accessing data indicates the responsiveness of the memory system.

as well as RAM Test Pro Memory Benchmark Designed to measure the performance of DDR5, DDR4, DDR3, and DDR2 memory, the software can test the performance of (1) sequential read, write, and copy bandwidths / (2) random read, write, and copy bandwidths / (3) read/write latency / (4) latency for random accesses to blocks of different sizes.

AIDA64 was updated in the course of the author's Z890 and X870 tests on different days, but my Z890 is finished and I'm not going to get any money for retesting it again and it's a waste of my time, so I'm not going to retest it, and also because the two platforms don't have the same version of AIDA64, based on the principle of fair comparison, this article doesn't suggest a direct side-by-side comparison between the two platforms and AIDA64 only takes a screenshot once and doesn't test many times to get the best result. AIDA64 also only carry out a direct screenshot, and did not test many times to take what the best results, for the author to run once is how much is how much, the score has a substantial change that is the software end of the problem (yes, I am talking about AIDA64).

∆ DDR5 6000 MT/s 32GB (2x 16GB) CL 26-36-36-96 1.45V for SPD Information View.

∆ DDR5 6000 MT/s 32GB (2x 16GB) CL 28-36-36-96 1.4V for SPD Information View.

∆ DDR5 6400 MT/s 32GB (2x 16GB) CL 32-39-39-102 1.4V for SPD Information View.

∆ DDR5 8000 MT/s 48GB (2x 24GB) CL 40-48-48-128 1.4V for SPD Information View.

 

Intel Core Ultra 9 285K and ASRock Z890 Taichi OCF Platform Testing

Intel Testing Platform Usage ASRock Z890 Taichi OCF motherboards with 24-core, 24-threaded Intel Core Ultra 9 285K The processor motherboard was tested, the motherboard BIOS was updated to version 3.07, the DRAM Gear Ratio setting in the motherboard is kept automatic, only the Memory Profile is turned on, the rest is left to the platform to decide automatically, and no other overclocking settings are turned on.

Testing Platform

Processor:Intel Core Ultra 9 285K QS
Cooler: Valkyrie E360 (full speed)
Water-cooled fan: LIAN LI UNI FAN P28 (full speed)
Motherboard:ASRock Z890 Taichi OCF ( BIOS version: 3.07 )
Graphics: NVIDIA GeForce RTX 4060Ti Founders Edition
Operating System: Windows 11 Professional 24H2
System Drive: Plextor PCIe Gen3 x4 M.2 2280 SSD 512GB
Game Dish:XPG GAMMIX S70 PRO PCIe Gen4 x4 M.2 SSD 4TB
Case: STREACOM BC1 Benchtable V2

 

firstly CPU-Z to view information about each memory module after opening the Profile.

∆ CPU-Z Information View_F5-6000J2636H16GX2-TR5NS.

∆ CPU-Z Information View_F5-6000J2836G16GX2-TR5NS.

∆ CPU-Z Info View_F5-6400J3239G16GX2-TR5S.

∆ CPU-Z Information View_F5-8000J4048G24GX2-TR5NS.

 

In AIDA64 tests, the Z890 platform has similar read and write performance at the same 6000 MT/s with different CL values, but the replication performance is slightly higher by about 1.5%, while the latency of the CL26 is lower than the CL28 specification by about 1.3%.

The higher frequency 6400 MT/s CL 32 has an increase in overall bandwidth performance due to the increase in frequency, but the latency is 90.2 ns. Surprisingly, the 8000 MT/s CL 40 has an increase in overall bandwidth performance, and also has the lowest latency among the four specifications.

∆ DDR5 6000 MT/s 32GB (2x 16GB) CL 26-36-36-96 1.45V_AIDA64 test results.

∆ DDR5 6000 MT/s 32GB (2x 16GB) CL 28-36-36-96 1.4V_AIDA64 test results.

∆ DDR5 6400 MT/s 32GB (2x 16GB) CL 32-39-39-102 1.4V_AIDA64 Test Results.

∆ DDR5 8000 MT/s 48GB (2x 24GB) CL 40-48-48-128 1.4V_AIDA64 test results.

 

In the RAM Test Pro Memory Benchmark, Latency and Linear Speed are similar to AIDA64, while Random Speed increases or decreases purely with the frequency specification.

∆ DDR5 6000 MT/s 32GB (2x 16GB) CL 26-36-36-96 1.45V_Test Results.

∆ DDR5 6000 MT/s 32GB (2x 16GB) CL 28-36-36-96 1.4V_Test_Results.

∆ DDR5 6400 MT/s 32GB (2x 16GB) CL 32-39-39-102 1.4V_Test_Results.

∆ DDR5 8000 MT/s 48GB (2x 24GB) CL 40-48-48-128 1.4V_Test Results.

 

AMD Ryzen 9 9950X3D and GIGABYTE X870 AORUS TACHYON ICE Platform Testing

The AMD AM5 platform is using GIGABYTE X870 AORUS TACHYON ICE motherboards with 16-core, 32-threaded AMD Ryzen 9 9950X3D processor and update the motherboard BIOS to F4b version for testing, the UCLK DIV1 MODE setting in the motherboard is kept automatic, only the memory profile is turned on and the rest is left to the platform to decide automatically, no other overclocking settings are turned on additionally.

Testing Platform

Processor:AMD Ryzen 9 9950X3D
Radiator:LIAN LI GA II LITE 360 RGB (full speed)
Motherboard:GIGABYTE X870 AORUS TACHYON ICE (BIOS version: F4b)
Graphics: NVIDIA GeForce RTX 4060 Ti Founders Edition 8GB
Operating System: Windows 11 Professional 24H2
System Dish:Kingston FURY Renegade PCIe 4.0 NVMe M.2 SSDs
Power supply:MONTECH TITAN PLA 1000W
Case: STREACOM BC1 Benchtable V2

 

firstly CPU-Z To view the information of each memory module after opening the Profile, the F5-8000J4048G24GX2-TR5NS will run with CL 40-48-48-127 on GIGABYTE platform because the upper limit of tRAS in GIGABYTE platform's BIOS is 127, which is the same even when applying the memory profile directly.

∆ CPU-Z Information View_F5-6000J2636H16GX2-TR5NS.

∆ CPU-Z Information View_F5-6000J2836G16GX2-TR5NS.

∆ CPU-Z Information View_F5-8000J4048G24GX2-TR5NS.

 

On the AMD AM5 platform, the CL26's read/write bandwidth is a bit better than the CL28's, while the copy performance is a bit better for the CL28, and the latency results are about the same. However, the performance difference between these two specifications in AIDA64 is actually not that bad, and the latency results are not any better because of the CL value.

There is an argument that the AM5 platform does not have much performance improvement after exceeding 6000 MT/s. In terms of write performance, which has seen the greatest improvement, the 8000 CL40 has a 15.4% performance improvement over the 6000 CL28, and the bandwidth improvement is indeed not as much as compared to the Z890 platform, but there is still a performance improvement, only that the latency score has also increased to 84.9 ns relatively. The latency score has also increased to 84.9 ns.

∆ DDR5 6000 MT/s 32GB (2x 16GB) CL 26-36-36-96 1.45V_AIDA64 test results.

∆ DDR5 6000 MT/s 32GB (2x 16GB) CL 28-36-36-96 1.4V_AIDA64 test results.

∆ DDR5 8000 MT/s 48GB (2x 24GB) CL 40-48-48-128 1.4V_AIDA64 test results.

 

In the RAM Test Pro Memory Benchmark, the F5-8000J4048G24GX2-TR5NS, which is also a DDR5 8000 MT/s, performs better in all bandwidth-related scores, but the latency scores are higher than the 6000 MT/s specification.

∆ DDR5 6000 MT/s 32GB (2x 16GB) CL 26-36-36-96 1.45V_Test Results.

∆ DDR5 6000 MT/s 32GB (2x 16GB) CL 28-36-36-96 1.4V_Test_Results.

∆ DDR5 8000 MT/s 48GB (2x 24GB) CL 40-48-48-128 1.4V_Test Results.

 

Conclusion

The main purpose of this article is to find out whether each specification really has lower latency due to lower CL values in the common memory AIDA64 Benchmark software, and whether there is really no performance gain for AMD platforms using 8000 MT/s memory.

In the 6000 MT/s CL28 and CL26 tests, the CL26 does not have a lower Latency in the two Benchmarks due to the lower CAS Latency(CL) specification, while the 8000 MT/s does have a higher latency on the X870 platform due to the increased frequency, while the 8000 CL40 on the Z890 platform has a better latency performance than the other specs. The 8000 MT/s does increase the bandwidth performance due to the increase in frequency, which is indeed higher on the X870 platform, while on the Z890 platform the latency of the 8000 CL40 is a bit better than the other specifications, and the above results are only representative of the test results on my own platform.