Recommendations and conclusion
What we could just read in the data, I would now like to summarize again in a few concise recommendations for the RAM OC with Rocket Lake, whereby with OC of course also XMP is meant.
Ranks: More ranks are always better, as long as you don’t have to sacrifice speed for them, keyword silicon lottery at the IMC. And even then, the advantage of multiple ranks can outweigh clock advantages of up to 1000 MHz. But that’s actually cold coffee and was already the case for previous CPU generations, as we could already show in the past: Performance increase for gamers: Memory with single-rank, dual-rank and quad-rank explained and compared | Practice with benchmarks
Gearing: If you don’t change the clock speed, but just load a XMP profile, you should choose the lower working gear. But if you dare to set the clock rate yourself, a compromise and maybe even an underclocking can make sense, e.g. if you run a DDR4-3800 XMP kit, for which you would need Gear 2, one step slower with DDR4-3733 with Gear 1 instead. You gain some latency and depending on the game or application also up to about 5% of performance.
Reference Clock: If you want to achieve a certain clock speed, you often have no choice with the Reference Clock, because only very few clock speeds can actually be achieved with 133 MHz as well as with 100 MHz. But if you have the choice or if you set the RAM clock yourself, the Reference Clock 133 MHz is recommended. This is because you can get by with a lower QLCK ratio at the same clock rate and thus reduce the load on the IMC.
SA (System Agent) Voltage: While previous generations often had the motto “A lot helps a lot!”, the new Rocket Lake IMC requires a few hundred mV less System Agent voltage for the same RAM clock. Thus, the ideal range in my tests is 1.25 – 1.3 V, even for RAM clock rates beyond 5 GHz. ASRock solves this well in the BIOS of the Z590 PG Velocita with a maximum Auto SA voltage of 1.28 V. If it’s a bit more with other board manufacturers and the system does not post, even a manual lowering of the SA voltage into the range just mentioned is recommended.
VCCIO Memory Voltage: This voltage is new to Rocket Lake and has been separated from the SA voltage of previous generations. You can’t affect POST and memory training behavior with this, but boosting the voltage can help stabilize a RAM OC for stress testing and daily use. Currently, however, it is not yet possible to read out this voltage with software tools like HWinfo.
VCCIO Voltage: This voltage had no influence on RAM overclocking in my tests. In previous generations, this voltage was used to stabilize higher clock speeds on the cache or ring of the CPU, but since we left this part of the CPU at the standard 4 GHz clock in today’s test, it’s unnecessary. Regardless, on my particular CPU, the cache clock is limited by the core voltage first anyway.
As I mentioned between the lines, the mainstream DDR4-4000 tested today is nowhere near the limit of RAM clock for Rocket Lake. On the contrary, Intel has probably made even bigger steps forward with the memory controller than with the CPU cores. And so with the right RAM kit and a 2-DIMM motherboard even clock rates beyond DDR4-5000 can be operated without any problems.
With the ADATA XPG D50 Spectrix Xtreme 4800 CL19 kit, which we recently reviewed, even DDR4-5600 with timings 22-30-30-50 can be operated, mind you without hacks in Windows and with a relatively moderate 1.65 V RAM voltage. If you consider that the Z590 platform and its BIOS versions are still in their infancy, you can look forward to the future with excitement.
As an enthusiast, I was all the more pleased to hear that ASRock will once again be reaching for the absolute spearhead in RAM-OC with the latest Intel platform with the announced Z590 OC Formula. Fittingly, various RAM manufacturers have also already announced further single and dual-rank kits with Hynix DJR ICs and XMP clock rates of over DDR5-5000. DDR5 clock rates and timings with DDR4 already today, Rocket Lake makes it possible!
The motherboard was provided to us by ASRock without requirements nor obligation to publish. Moreover, no influence on the tests and their results was exerted.
65 Antworten
Kommentar
Lade neue Kommentare
Moderator
Mitglied
Moderator
Veteran
Mitglied
Mitglied
Veteran
Mitglied
Mitglied
Veteran
Mitglied
Veteran
Neuling
Urgestein
Mitglied
Mitglied
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →