6000 CL30 EXPO 1:1 5H16M DR, 2000 IF – Fully loaded
The DDR5-6000 CL30 configuration is now available in a double version, i.e. with 4 RAM modules instead of the usual 2. Yes, with X670 and Zen 4, 64 GB of DDR5 RAM can actually be run stably at 6000 Mbps. Due to the dual-channel architecture, this configuration then effectively represents dual-channel and is marked accordingly in the diagrams. Thus, we can immediately see whether dual-rank brings a reasonable performance increase compared to single-rank.
Besides the G.Skill 6400 CL32 kit, a Teamgroup 6000 CL40 kit is used for this. Both kits use the same Hynix M-Die memory chips and can be combined without any problems despite their different PCB design.
6000 CL30 EXPO 1:1 5H16M SR, 2167 IF – Maximized Infinity Fabric
This is the single-rank config with 6000 Mbps and EXPO timings for the last time, but this time with the full Infinity Fabric clock at 2167 MHz. So we can see one more time how much performance the faster connection between IO die and chiplets really brings, and furthermore if this behavior differs between JEDEC and EXPO RAM settings.
6400 CL32 * 1:1 5H16M SR, 2133 IF – Fully optimized single-rank Hynix M-Die
Now we finally get to the manually optimized configurations, whereby the maximum achieved clock rate of 6400 Mbps is unfortunately a bit lower than I would have liked. Here, either the early AGESA code still has a bit of room for improvement or my mainboard has a hard limit here. However, since this limit is also noticeable with only a single RAM module in slot B2, I tend towards the former. Reports from testers with motherboards from other manufacturers also confirmed a hard limit of currently 6400 – 6600 Mbps to me, at least at normal temperatures.
Assuming DDR5-6400, all timings were then tightened and, mind you, still in 1:1 mode with unchanged voltages. Unfortunately, there still seems to be a bug in the Ryzen Master software, so that the timings are no longer displayed correctly here. Accordingly, you’ll have to make do with screenshots from the BIOS from here on.
However, the Infinity Fabric now seems to be more heavily loaded due to the higher data volume, which means that one step less and thus only 2133 MHz can be operated stably. Even with higher voltages at SOC, VDDG etc., 2167 MHz can no longer be stabilized on the FCLK with such fast RAM. Sometimes the system still boots, but always ends up in the Bluescreen of Death a short time later.
6400 CL32 * 1:1 5H16M DR, 2133 IF – Fully optimized dual-rank Hynix M-Die
The whole thing is correspondingly analog once again in full configuration with 4x 16 GB, whereby the tertiary timings ending on Dd (different DIMM) become relevant here accordingly and can no longer be minimized to 1. In addition, the double number of ranks requires a bit more SOC voltage, here 1.4 V, although this is still in the safe range according to several sources.
6400 CL36 * 1:1 5S16B SR, 2167 IF – clock lid as salvation for Samsung’s DDR5?
In our DDR5 tests with kits using Samsung’s 16 Gbit B-Die memory ICs, their relatively low clock potential was always a major drawback. But how do the ICs behave when the clock is specified at 6400 Mbps and the timings are ultimately the defining characteristic? Especially the tighter tRCD, tRP and tRRD timings should benefit the Samsung ICs here.
Furthermore, 2167 MHz can also be stabilized on the Infinity Fabric with slightly higher voltages – less is sometimes more. In addition to 1.4 V SOC, 1.15 V VDD Misc, 1.25 V VDDG CCD, 1.25 V VDDG IOD and 1.15 V VDDP voltage are used here.
By the way, I also tested Hynix 16 Gbit A-Die, but it behaves the same way as Hynix 16 Gbit M-Die and cannot be clocked higher than 6400 Mbps. Furthermore, Micron 16 Gbit RevA with Zen 4 behaves the same way as with Intel Alder Lake and stops working with DDR5-5800 and higher. Accordingly, there are no configurations and results for the time being, until this may change in future AGESA updates.
- 1 - What's new with Ryzen 7000?
- 2 - Test Setup and Software Tools
- 3 - BIOS Settings (1/2) – DDR5
- 4 - BIOS Settings (2/2) – CPU-OC and other
- 5 - Tested Configurations (1/2)
- 6 - Tested Configurations (2/2)
- 7 - Synthetics (1/2) – LinpackXtreme, AIDA64, Geekbench 3
- 8 - Synthetics (2/2) – SuperPi 32M, PyPrime 2.0 2B, Timespy CPU
- 9 - Gaming QHD, FHD – ACC, CSGO, SoTR
- 10 - Summary and Recommendations
221 Antworten
Kommentar
Lade neue Kommentare
Veteran
Veteran
Mitglied
Urgestein
1
Urgestein
Veteran
Urgestein
Urgestein
Veteran
Urgestein
Veteran
Urgestein
Urgestein
Urgestein
Neuling
Mitglied
Mitglied
Alle Kommentare lesen unter igor´sLAB Community →