Care should be taken when loading XMP profiles on both platforms. Because on the Intel platform, most motherboards with Rocket Lake-S CPU only use the multipliers based on the 133 MHz reference clock and together with the Gear 2 you just can’t get to the DDR4-4400 in the XMP profile. Asus on the Maximus XIII Apex tries to counter this by raising the BCLK to 103.2 MHz, which may be well-intentioned, but it also indirectly overclocks the CPU and makes it unstable.
No, Asus, this isn’t just “slightly adjusted” and overclocks the CPU from 5.1 to 5.26 GHz in boost. If you really want to achieve DDR4-4400 on Rocket Lake, it is much more practical to use the 100 MHz reference clock and add a bit more System Agent voltage to it to make the memory controller run stable. So I was able to get the XMP profile running stably with 1.4 V SA and 1.45 V IO Mem on Rocket Lake with relative ease.
And also the AMD platform compatibility is not optimal. The 1:2 mode is known to be particularly suboptimal, and the odd CAS latency is doubly unpleasant, as it rounds up from 19 to 20 with the native geardown mode. The result is a bit more access latency and less performance, but at least the XMP profile is stable here at Team Red without any further intervention.
The unfavorable XMP behavior on both platforms is not really the fault of Crucial or Intel or AMD, but rather an unfavorable chain of circumstances. But the consequence is that buyers should already have some idea about RAM tuning and OC to avoid unpleasant interactions, especially with Intel Rocket Lake.
Overclocking Intel and AMD
But if we now invest some time in manual tuning, the kit starts to show its true performance potential. Due to the RevB ICs, the modules can be clocked similarly well as Hynix DJR-based ones, but with somewhat tighter timings and thus lower latencies. And even 1.6 V voltage is tolerated by the DIMMs without any problems, whereby the heatsinks easily dissipate the additional heat.
Finally, the maximum stable overclock on the Intel Rocket Lake platform with the Asus Maximus XIII Apex is a whopping DDR4-5066 18-23-23-42, and that’s with 32 GB total capacity mind you. As often happens, silicon lottery in the quality of the memory controller of the CPU and compatibility with the motherboard BIOS also play a major role. 1.475 V System Agent and 1.55 V VCCIO Memory or VCCIO 2 I have to feed the CPU here for complete stability, which is definitely at the upper end of what is reasonable on a daily basis.
And as most of you might have noticed, on the Maximus XIII Apex with BIOS 1007 I can’t manually set tREFI, tRFC, tWTR_L and _S or tWRRD_sg and _dg without destabilizing the OC in every case. On boards of other manufacturers the problem cannot be reproduced so far. So the cause seems to be not the RAM but the board or BIOS. Nevertheless, all tests are carried out on the Asus motherboard due to better comparability, just with a corresponding performance disadvantage, which should be kept in mind. The rest of the timings were all manually optimized, see screenshot above.
Even on the AMD platform with our B550 Aorus Pro V2, the RAM kit can be overclocked without any problems, even if only to DDR4-4800 due to the 4-slot topology of the board. However, the timings are also significantly lower with 20-21-21-41 and also the voltage only requires 1.5 V on the RAM, 1.15 V for SOC, 1.0 and 1.1 V for VDDG CCD and VDDG IOD respectively.
However, the probably more interesting variant for AMD users might be the 1:1 configuration, with DDR4-3800 limited by the Infinity Fabric of our 5950X test CPU. Timings 16-17-17-34 can be set here, with identical voltages as the previous config, now these are really values you would almost only expect from Samsung’s B-Die.
By the way, there are no problems with tRFC or tWTR timings on the AMD platform and on this board – tREFI is fixed anyway – and the compatibility with the CPU seems a bit more reliable here. So here all timings could really be optimized to the limit of the modules.
Intel’s Rocket Lake is, as we all know, a half-step up from Alder Lake and as the first CPU with a transmission or gearing in IMC, new territory for Team Blue. But even in today’s test, we saw that the platform still has some issues in the Gear2, with motherboard manufacturers seemingly being left alone to patch the microcode. Accordingly, the behavior between boards or in some cases even between reboots is different, as we had seen in the last test with the DJR-based kit from G.Skill.
Here’s hoping that Intel can fix the various inconsistencies a la Early Access for the upcoming Alder Lake platform and it won’t need day-one microcode patches again. Because it seems that DDR4 will remain relevant for the time being and high-end kits, like the one from today’s test, have to be able to be operated reliably by the CPU. AMD does a much better job here at times.
- 1 - Unboxing and first Impressions
- 2 - Dimensions and RGB
- 3 - SPD and Heatsink Performance
- 4 - Teardown and PCB Comparison (vs. RevE)
- 5 - Test Systems and Methods
- 6 - XMP Behavior and Overclocking
- 7 - Synthetic Benchmarks – AIDA64 und Geekbench 3
- 8 - Gaming – Cyberpunk 2077 in UHD, QHD, FHD
- 9 - Summary and Coinclusion
14 Antworten
Kommentar
Lade neue Kommentare
Moderator
Urgestein
Mitglied
Urgestein
Urgestein
Urgestein
Urgestein
Urgestein
Urgestein
Veteran
Urgestein
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →