In order to test the performance of the heatsinks, the RAM kit is also forced to sweat in Testmem5 with profile “Extreme1@Anta777” today. For this purpose, external type K temperature sensors are used together with an Elmorlabs KTH for temperature measurement. One of the sensors is placed in the center of the outer (warmer) module at the height of the ICs, between the board and the heatspreader. With another one the ambient temperature is measured, the values are logged 2 times per second each and so the delta is formed. This test was of course performed before the teardown, so the results really describe the modules in their original state. In addition, the heat test was performed on the Asus Maximus Z690 Apex, as this board has the smallest spacing between DIMM slots and thus poses the biggest challenge to the heatsinks.
In addition to the ADATA XPG LANCER kit, I also included the Corsair Dominator Platinum RGB kit from the previous review in the diagram for comparison. Of course, these two kits are only comparable to a limited extent due to their different XMP voltages and memory ICs used, but this way we at least have a clue. With a 120 mm fan on the modules (active cooling), the XPG modules level off at about 13 °C above the ambient temperature. Without cooling, the temperature is just under 30 °C, but even that still leaves a lot of room for overclocking. While the Hynix M-Die ICs don’t have internal temperature sensors either, as we saw in Igor’s thermal imaging tests, a 15-20 °C markup on the SPD Hub Temp, which our external measurements are very close to, is relatively reliable for determining IC temperatures.
Overclocking – DDR5-6800 CL30 please?
Since DDR5 is really still in its infancy, the compatibility between motherboard, BIOS version, CPU and RAM kit is currently still quite reminiscent of playing the lottery, with the chances of winning decreasing the higher the selected frequency. Accordingly, it is of course very helpful if you can try out several of these combinations until you find the best possible one, as in today’s test. In my case, this consists of the MSI MEG Z690 Unify-X with BIOS A12U4 and my “SP93” i9-12900K CPU. With this, the LANCER modules can be stably overclocked to DDR5-6800 at 1.45 V, which is just above the JEDEC voltage limit and relatively humane. Based on this, there is once a setting with Command Rate 2T and tCL 30, and on the other with Command Rate 1T and tCL32, whereby the other timings differ only slightly.
The latter with Command Rate 1T produces much more waste heat than the 2T counterpart despite identical voltages and was therefore equipped with active cooling, i.e. a 120 mm fan above the DIMMs, for stabilization. While the 2T configuration also passively stays below 65 °C SPD hub temp according to HWInfo, the 1T config would march tightly towards 80 °C. With a fan, the temperature remains just under 40 °C. Here you can once again see that today’s standard passive RAM coolers often reach their limits relatively quickly if left on their own. For enthusiasts and overclockers, active RAM cooling via air or even custom waterblock is therefore often the means of choice, which is why I have also included this overclocking config in today’s tests, marked with a UTF-8 snowflake ❄. The larger DIMM slot spacing of the Z690 Unify-X motherboard naturally also plays a non-negligible role here.
All in all, I am very satisfied with the overclocks. With a better CPU or future BIOS optimizations, it will probably be possible to break through the clock barrier of 7000 Mbps stably. The biggest part of these high clock rates and performance results is of course due to the Hynix M-Die memory ICs, which are the clear winners in the first generation of DDR5, at least in terms of clock potential and efficiency. It is all the more unfortunate that ADATA also sells modules with Samsung B-Die memory modules under the same SKU number, which makes it very difficult for end customers to see what they are buying before they have it shipped.
Other RAM manufacturers do the same and not only since DDR5, but it is still a point of criticism in terms of consumer friendliness that should be mentioned. If you really want to be sure that you get Hynix ICs, you currently have to buy a RAM kit with an XMP spec of more than 6000 Mbps, which is often also much more expensive.
- 1 - Unboxing and design
- 2 - Dimensions and SPD information
- 3 - Teardown and PCB analysis
- 4 - Test systems and methodology
- 5 - Heatsink peformance and overclocking potential
- 6 - Synthetics – AIDA64, Geekbench 3, SPI32M, LinpackXtreme
- 7 - Gaming – SoTR, CS: GO, Cyberpunk 2077
- 8 - RGB lighting and conclusion
42 Antworten
Kommentar
Lade neue Kommentare
Urgestein
1
Mitglied
Urgestein
Urgestein
Veteran
Veteran
Urgestein
1
Urgestein
Urgestein
Urgestein
Mitglied
Mitglied
Mitglied
Mitglied
Mitglied
Mitglied
Mitglied
Alle Kommentare lesen unter igor´sLAB Community →