The new G.Skill Ultra-Low Latency 2x32GB DDR5-6000CL26 in test.
- A bin that is second to none.
First of all, thanks to G.Skill for providing the memory.
You can purchase it here
https://geizhals.eu/g-skill-trident...-64gb-f5-6000j2636g32gx2-tz5nrw-a3410334.html
There are many memories on the market, but the Hynix A-Die is one of the most popular and fastest on the market. However, there are major differences in quality and that is exactly what is offered here, a quality of ICs that you will probably not find on any other bin. A higher clock rate, tighter timings and less voltage required - that's what you would expect from a good bin, and first of all, that's exactly what you get here.
Think of it this way, it's hard to get a gem of a CPU, but the chances of finding 32 ICs of this quality by chance on an A-DIE kit is really nil. I was provided with the 2x32GB G.Skill Trident Z5 Neo RGB F5-6000J2636G32GX2-T5NRW. This has RGB lighting and an Expo profile for people who don't want to do it themselves.
This is specified as DDR5 6000CL26-36-36 at 1.4V DRAM voltage, the remaining timings can be found in the SPD screenshot, making this bin the only one to offer a CL26 at 6000Mhz at only 1.4V.
Here are a few impressions.
![Advertising]()
![Electronic device Technology Label Gadget Box]()
![Label]()
![Label]()
![Advertising Label]()
![Electronic device Technology Computer hardware Gadget Electronics]()
![Label]()
![Electronic component Circuit Component Computer hardware]()
![Photograph White Text Font Screenshot]()
As can be seen from the SPD, a Richtek RTQ5132GQWF is installed as the PMIC. This allows higher VDD voltages than 1.43V. The ICs used here are SK Hynix 16 Gbit A-Die, recognizable by the part number H5CG48AGBD X018. The ICs (447V) should probably be from the 47th week of 2024, i.e. November 2024. The sticker shows 0T64, so contrary to the SPD, this is the Speed Grade 6400BN from Hynix. The kit also has a 6K5850UA0 RGB controller. This can be controlled either with the Trident Z Lighting Control software or third-party software.
The memory is equipped on both sides with a total of 16 ICs per module and dual ranked. In other words, 4 way interleaving, 4 banks are addressed, which places a greater load on the IMC and therefore does not allow extreme frequencies such as DDR5-8000+. Since AMD's 1:1 mode UCLK=MCLK is normally the fastest for gaming anyway, this is not a problem. In my tests with the limited Kingston KF580C36RLAK2-48, I had worse gaming performance at 8000-8400MT/s, especially as I had to drive over 8200 GDM ON in 2:1 mode. So I stayed with the more sensible 1:1 mode, especially as it is also much easier to keep stable.
Basically, 4 way interleaving used to be the optimum ratio between latency and transfer rate. DR was better in copy value and SR in latency, but the high bandwidth of DDR5 should have put this into perspective.
In practice, I didn't notice any real difference between 2-way and 4-way interleaving on the AM5 platform. In the mass of benchmark runs, I was slightly higher overall in the AVG of the total FPS with DR. However, we are talking about nuances here, the max and low FPS were pretty much the same. The advantage here is that you are simply better positioned with 64GB.
The Testsystem
So much for the theory, now let's get to the practice.
The maximum boot bare frequency in 2:1 mode is a remarkable 7600MT/s, which is 200MT/s more than I managed with the 2x32GB 6000C30 bin from Corsair. But what is really amazing is that it is possible for me to run 6400MT/s 1:1 in gear down disabled mode and even with Nitro 1-2-0, which I actually thought was impossible with dual ranked memory. It was even possible to boot 6600MT/s 1:1 with GDM disabled, but only GDM on with 1-2-1 is stable and I couldn't even boot that before with the other DR kit. There, a maximum of 1-3-1 with GDM ON was possible.
It is often said that the usable nitro settings are only dependent on the memory controller of the processor. This memory proves the opposite. If you are not familiar with all this, it makes the memory even faster.
![Screenshot Number]()
![Screenshot Number Font Software Computer]()
To achieve such high frequencies with DR modules, a suitable motherboard is of course required. From my own experience, I can say that the MSI Tomahawk X870 and the Gigabyte Ice B650E, for example, are also able to run stable 6600 1:1 and boot in 2:1 up to 7000+. On the Asus Strix X670E-E I was unable to achieve stable operation above 6400, whether this was purely due to the bios and can be rectified is unknown to me.
Now comes the most important part, does the memory live up to its price?
First of all, yes definitely, the quality of this bin has never been seen before. I tested once without any cooling, then with a 140 fan at 800 rpm and under H2O. Once with sub-timings (subs) and also with auto subs to rule out any errors in the timings.
For all those who want to convert the memory to water cooling, do not use the old methods! The adhesive pads are so sticky that you cannot simply remove them! Soak the tank in isopropanol for 3 hours to loosen the adhesive pads, then simply dry in the oven at 60° for 3 hours. When converting, also make sure that the higher capacitors are covered with heat conducting pads so that they cannot cause a short circuit.
Double-sided memory gets much warmer than SR modules, I would advise against operating it without any airflow beyond 1.43V and then organizing TM5 orgies. This test generates a lot of heat, more than the Karhu memory test. In my case, the dashboard of the mainboard causes a heat build-up on the 2nd module.
Each test ran for 1 hour Karhu and approx. 25min TM5 @1usmus to be able to make a rough statement. The voltages represent the minimum voltage required to pass this test! With 6400CL28 and below, it was necessary to have the VDDQ higher than the VDD, as this is simply "too low". I have also listed the temperatures of both modules, because as you can see from the values, the temperature plays a significant role.
![Technology Computer hardware Electronic component]()
For comparison
My DDR5-6000 CL30 model needed 1.55V for DDR5-6600 CL28 under H2O. In comparison, this one needs 1.42V for DDR5-6600 CL28 without any cooling, although a stick in my test system gets warmer due to the dashboard of the mainboard. The Kingston KF580C36RLAK2-48 required 1.55V for DDR5-6600 CL28-39 without cooling, which is an absolute top value for M-DIE. This makes us curious as to what the new G.Skill 2x24GB DDR5-6000 CL28 Low Latency M-DIE models would need in terms of voltage?
DDR5-7000 CL28 under H2O are already possible from 1.49V and DDR5-7200 CL28 at 1.56V. That's way beyond what I was hoping for. DDR5-6600 Cl26 @ H2O with only 1.57V, my CL30 model didn't boot even at 1.73V. Whether uncooled, with active ventilation or water-cooled, all values are more than impressive. It can be stated that these modules were able to achieve stable DDR5-6200 CL26 and DDR5-6600 CL28 even without extra cooling. If you actively cool with air, there are hardly any limits. Even DDR5-6600 CL26 is already practicable without extreme voltages. The conversion to water cooling saves another 30-40mV, which is really remarkable.
It is easy to see how much influence the temperature has on the required voltage, which should not be forgotten, especially in summer with an air-cooled graphics card. If you look around the forums, you will see that this memory could be a preview of the quality of a future generation. Of course there are still differences within a bin, but what is on offer here is like gold at the Olympics!
The performance
Now it's time for synthetic tests. I chose Geekbench3 and Maxxmem2 for this. I ran all the tests 3 times and calculated the average value. In addition, I ran these tests in diagnostic mode to minimize strong deviations due to secondary loads. Maxxmem2 shows the latency well, Geekbench3 the real memory performance. As in all tests, the 9800X3D runs with PBO +200Mhz at CO -30, i.e. with moderate voltage.
![Image]()
![Image]()
The gain is impressive, from DDR5-6000 CL40-40 without additional adjustments to DDR5-6600 CL26 with sharp subs, it is up to 30% in memory performance. This shows the real potential of manual memory overclocking.
You can clearly see how both the latency and the bandwidth improve massively. The overall memory performance receives a massive boost. Geekbench3 tests various workloads such as Image Compression, Text Compression, Cryptography, Navigation, HTML5, SQLite, PDF/Text Rendering, Compiling and also Floating Point calculations. Here is an overview of the tests included.
https://www.geekbench.com/doc/geekbench3-workloads.pdf
The timings could be optimized even further. However, a significant amount of extra work is required to ensure stability and optimum performance in gaming mode. Unfortunately, this is beyond the scope of this user review. There is definitely potential for this. As a small teaser: DDR5-6600 CL26-37, with even tighter timings.
![Text Font Number Screenshot]()
However, due to the 3D cache of the 9800X3D, this will not be realized in gaming mode.
In order to demonstrate the influence of the RAM, I used in-game benchmarks for the most reproducible results possible. I used Shadow of the Tomb Raider, Far Cry Primal and Final Fantasy XIV. These were also run 3 times and the results were averaged to reduce fluctuations caused by accidental secondary loads. All tests were performed in 1280x720p resolution with low settings to avoid limitation by the graphics card and to illustrate the pure CPU performance increase.
As with all tests, the 9800X3D again runs with PBO +200Mhz at CO -30, i.e. with a moderate voltage.
![Technology]()
![Screenshot]()
![Font Screenshot]()
Significant gains are also visible here, but on a smaller scale. The minimum rather than the average FPS increases. Older titles, such as Far Cry Primal, which benefit significantly from the processor's large X3D cache, barely scale with the RAM. You may be slightly faster, but this would not usually be noticeable, at least with an X3D processor. If a CPU or CCD without a 3D cache comes into play, things can look different again. Then there should be a much higher increase.
My conclusion
The kit currently costs around 390 euros. That's not a small amount of money. But this bin is clearly aimed at enthusiasts who want the best of the best. In DDR4 times, the 32GB DDR4-4000 CL14 top models even cost around 500 euros. Therefore, I think the price is definitely justified for such a quality, which you may never find on another bin, especially with dual ranked memory.
As the platform will continue to be supplied with processors by AMD until at least 2027 and the memory can easily cope with DDR5-7200+, you are also well equipped for future generations. Currently, 32GB of RAM is practically mandatory, but with 64GB you are on the safe side in the long term.
The memory is also the first choice for anyone who wants to run high-performance settings without extra cooling.
I am more than satisfied with the kit. Whether you need it or not, the inner Monk in me says "It's Done". The search is over.
P.S. To check the results and for anyone who would like to be inspired by my settings, I have attached screenshots of all the tests.
Download
- A bin that is second to none.
First of all, thanks to G.Skill for providing the memory.
You can purchase it here
https://geizhals.eu/g-skill-trident...-64gb-f5-6000j2636g32gx2-tz5nrw-a3410334.html
There are many memories on the market, but the Hynix A-Die is one of the most popular and fastest on the market. However, there are major differences in quality and that is exactly what is offered here, a quality of ICs that you will probably not find on any other bin. A higher clock rate, tighter timings and less voltage required - that's what you would expect from a good bin, and first of all, that's exactly what you get here.
Think of it this way, it's hard to get a gem of a CPU, but the chances of finding 32 ICs of this quality by chance on an A-DIE kit is really nil. I was provided with the 2x32GB G.Skill Trident Z5 Neo RGB F5-6000J2636G32GX2-T5NRW. This has RGB lighting and an Expo profile for people who don't want to do it themselves.
This is specified as DDR5 6000CL26-36-36 at 1.4V DRAM voltage, the remaining timings can be found in the SPD screenshot, making this bin the only one to offer a CL26 at 6000Mhz at only 1.4V.
Here are a few impressions.
As can be seen from the SPD, a Richtek RTQ5132GQWF is installed as the PMIC. This allows higher VDD voltages than 1.43V. The ICs used here are SK Hynix 16 Gbit A-Die, recognizable by the part number H5CG48AGBD X018. The ICs (447V) should probably be from the 47th week of 2024, i.e. November 2024. The sticker shows 0T64, so contrary to the SPD, this is the Speed Grade 6400BN from Hynix. The kit also has a 6K5850UA0 RGB controller. This can be controlled either with the Trident Z Lighting Control software or third-party software.
The memory is equipped on both sides with a total of 16 ICs per module and dual ranked. In other words, 4 way interleaving, 4 banks are addressed, which places a greater load on the IMC and therefore does not allow extreme frequencies such as DDR5-8000+. Since AMD's 1:1 mode UCLK=MCLK is normally the fastest for gaming anyway, this is not a problem. In my tests with the limited Kingston KF580C36RLAK2-48, I had worse gaming performance at 8000-8400MT/s, especially as I had to drive over 8200 GDM ON in 2:1 mode. So I stayed with the more sensible 1:1 mode, especially as it is also much easier to keep stable.
Basically, 4 way interleaving used to be the optimum ratio between latency and transfer rate. DR was better in copy value and SR in latency, but the high bandwidth of DDR5 should have put this into perspective.
In practice, I didn't notice any real difference between 2-way and 4-way interleaving on the AM5 platform. In the mass of benchmark runs, I was slightly higher overall in the AVG of the total FPS with DR. However, we are talking about nuances here, the max and low FPS were pretty much the same. The advantage here is that you are simply better positioned with 64GB.
The Testsystem
| Processor | 9800X3D (PBO +200, CO-30) |
| Mainboard | MSI MEG X870E Godlike |
| Graphics card | GeForce RTX 4090FE |
| RAM | F5-6000J2636G32GX2-T5NRW |
| Cooling | 2xLaing DDC 3.25, 2x Mora 420,CPU TechN AM5, GPU Alphacool Aurora, Ram EK Monarch+X4 Ram Block |
| Operating system | Win11 with all updates but core isolation is off, just as I use it in everyday life |
So much for the theory, now let's get to the practice.
The maximum boot bare frequency in 2:1 mode is a remarkable 7600MT/s, which is 200MT/s more than I managed with the 2x32GB 6000C30 bin from Corsair. But what is really amazing is that it is possible for me to run 6400MT/s 1:1 in gear down disabled mode and even with Nitro 1-2-0, which I actually thought was impossible with dual ranked memory. It was even possible to boot 6600MT/s 1:1 with GDM disabled, but only GDM on with 1-2-1 is stable and I couldn't even boot that before with the other DR kit. There, a maximum of 1-3-1 with GDM ON was possible.
It is often said that the usable nitro settings are only dependent on the memory controller of the processor. This memory proves the opposite. If you are not familiar with all this, it makes the memory even faster.
To achieve such high frequencies with DR modules, a suitable motherboard is of course required. From my own experience, I can say that the MSI Tomahawk X870 and the Gigabyte Ice B650E, for example, are also able to run stable 6600 1:1 and boot in 2:1 up to 7000+. On the Asus Strix X670E-E I was unable to achieve stable operation above 6400, whether this was purely due to the bios and can be rectified is unknown to me.
Now comes the most important part, does the memory live up to its price?
First of all, yes definitely, the quality of this bin has never been seen before. I tested once without any cooling, then with a 140 fan at 800 rpm and under H2O. Once with sub-timings (subs) and also with auto subs to rule out any errors in the timings.
For all those who want to convert the memory to water cooling, do not use the old methods! The adhesive pads are so sticky that you cannot simply remove them! Soak the tank in isopropanol for 3 hours to loosen the adhesive pads, then simply dry in the oven at 60° for 3 hours. When converting, also make sure that the higher capacitors are covered with heat conducting pads so that they cannot cause a short circuit.
Double-sided memory gets much warmer than SR modules, I would advise against operating it without any airflow beyond 1.43V and then organizing TM5 orgies. This test generates a lot of heat, more than the Karhu memory test. In my case, the dashboard of the mainboard causes a heat build-up on the 2nd module.
Each test ran for 1 hour Karhu and approx. 25min TM5 @1usmus to be able to make a rough statement. The voltages represent the minimum voltage required to pass this test! With 6400CL28 and below, it was necessary to have the VDDQ higher than the VDD, as this is simply "too low". I have also listed the temperatures of both modules, because as you can see from the values, the temperature plays a significant role.
For comparison
My DDR5-6000 CL30 model needed 1.55V for DDR5-6600 CL28 under H2O. In comparison, this one needs 1.42V for DDR5-6600 CL28 without any cooling, although a stick in my test system gets warmer due to the dashboard of the mainboard. The Kingston KF580C36RLAK2-48 required 1.55V for DDR5-6600 CL28-39 without cooling, which is an absolute top value for M-DIE. This makes us curious as to what the new G.Skill 2x24GB DDR5-6000 CL28 Low Latency M-DIE models would need in terms of voltage?
DDR5-7000 CL28 under H2O are already possible from 1.49V and DDR5-7200 CL28 at 1.56V. That's way beyond what I was hoping for. DDR5-6600 Cl26 @ H2O with only 1.57V, my CL30 model didn't boot even at 1.73V. Whether uncooled, with active ventilation or water-cooled, all values are more than impressive. It can be stated that these modules were able to achieve stable DDR5-6200 CL26 and DDR5-6600 CL28 even without extra cooling. If you actively cool with air, there are hardly any limits. Even DDR5-6600 CL26 is already practicable without extreme voltages. The conversion to water cooling saves another 30-40mV, which is really remarkable.
It is easy to see how much influence the temperature has on the required voltage, which should not be forgotten, especially in summer with an air-cooled graphics card. If you look around the forums, you will see that this memory could be a preview of the quality of a future generation. Of course there are still differences within a bin, but what is on offer here is like gold at the Olympics!
The performance
Now it's time for synthetic tests. I chose Geekbench3 and Maxxmem2 for this. I ran all the tests 3 times and calculated the average value. In addition, I ran these tests in diagnostic mode to minimize strong deviations due to secondary loads. Maxxmem2 shows the latency well, Geekbench3 the real memory performance. As in all tests, the 9800X3D runs with PBO +200Mhz at CO -30, i.e. with moderate voltage.
The gain is impressive, from DDR5-6000 CL40-40 without additional adjustments to DDR5-6600 CL26 with sharp subs, it is up to 30% in memory performance. This shows the real potential of manual memory overclocking.
You can clearly see how both the latency and the bandwidth improve massively. The overall memory performance receives a massive boost. Geekbench3 tests various workloads such as Image Compression, Text Compression, Cryptography, Navigation, HTML5, SQLite, PDF/Text Rendering, Compiling and also Floating Point calculations. Here is an overview of the tests included.
https://www.geekbench.com/doc/geekbench3-workloads.pdf
The timings could be optimized even further. However, a significant amount of extra work is required to ensure stability and optimum performance in gaming mode. Unfortunately, this is beyond the scope of this user review. There is definitely potential for this. As a small teaser: DDR5-6600 CL26-37, with even tighter timings.
However, due to the 3D cache of the 9800X3D, this will not be realized in gaming mode.
In order to demonstrate the influence of the RAM, I used in-game benchmarks for the most reproducible results possible. I used Shadow of the Tomb Raider, Far Cry Primal and Final Fantasy XIV. These were also run 3 times and the results were averaged to reduce fluctuations caused by accidental secondary loads. All tests were performed in 1280x720p resolution with low settings to avoid limitation by the graphics card and to illustrate the pure CPU performance increase.
As with all tests, the 9800X3D again runs with PBO +200Mhz at CO -30, i.e. with a moderate voltage.
Significant gains are also visible here, but on a smaller scale. The minimum rather than the average FPS increases. Older titles, such as Far Cry Primal, which benefit significantly from the processor's large X3D cache, barely scale with the RAM. You may be slightly faster, but this would not usually be noticeable, at least with an X3D processor. If a CPU or CCD without a 3D cache comes into play, things can look different again. Then there should be a much higher increase.
My conclusion
The kit currently costs around 390 euros. That's not a small amount of money. But this bin is clearly aimed at enthusiasts who want the best of the best. In DDR4 times, the 32GB DDR4-4000 CL14 top models even cost around 500 euros. Therefore, I think the price is definitely justified for such a quality, which you may never find on another bin, especially with dual ranked memory.
As the platform will continue to be supplied with processors by AMD until at least 2027 and the memory can easily cope with DDR5-7200+, you are also well equipped for future generations. Currently, 32GB of RAM is practically mandatory, but with 64GB you are on the safe side in the long term.
The memory is also the first choice for anyone who wants to run high-performance settings without extra cooling.
I am more than satisfied with the kit. Whether you need it or not, the inner Monk in me says "It's Done". The search is over.
P.S. To check the results and for anyone who would like to be inspired by my settings, I have attached screenshots of all the tests.
Download
