Question High CAS Latency with slow DDR ?

[jaydub868]

I'm putting together an arrow lake rig on a Z890 Asrock motherboard (Pro RS). The QVL memory list for that motherboard shows I can use G.Skill 8400 CAS 40 2x24GB memory sticks. My question is, if I subsequently decided to run the memory at DDR 5600 (for stability), would CAS at that speed likewise go down in lockstop, or would I be stuck with the crappy combination of slow DDR and high latency of 40? On 5600 sticks, you can get CAS down to 30-32.

 

[Lutfij]

Why do you need to look at DDR5-8400MHz rams? You can easily pick out DDR5-6000MHz rams that's known to be the sweet spot for any DDR5 platform(both Intel and AMD). In fact you can pick out any DDR5-6000MHz or slightly higher frequency ram kit with tight latencies(that has Intel's X.M.P advertised on it) to work with your platform.

 

[jaydub868]

Why do you need to look at DDR5-8400MHz rams? You can easily pick out DDR5-6000MHz rams that's known to be the sweet spot for any DDR5 platform(both Intel and AMD). In fact you can pick out any DDR5-6000MHz or slightly higher frequency ram kit with tight latencies(that has Intel's X.M.P advertised on it to work with your platform.

Yes, I could do that. But I'd like to get some kind of answer to the question posed.

 

[CountMike]

Yes, I could do that. But I'd like to get some kind of answer to the question posed.

Cas and other latencies are higher on faster RAM so all modules ca be synchronized properly. With lover frequency come lower latency, so yeah you can lower Cas and others appropriately. As that fast RAM most probably doesn't have XMP as low as 5600MHz, you can just set everything manually according to JEDEC standards.
PS. Cas Latency is not only one and not even most important one of all latencies and other settings. Lowering just Cas would do little to no good for RAM performance and even less for overall system performance, all other timings and settings will have to be adjusted accordingly.

View: https://www.youtube.com/watch?v=H-Ofj5nIYfY

 

[CompuTronix]

In answer your question, G. Skill 48GB DDR5 8400 CL40 1.4 volt kits have a True Latency of 9.524 nS (nanoseconds).

T / Ghz x CL = TL (True Latency) ... where Time divided by Single Data Rate (SDR) Frequency in Ghz times CL (Column Latency) = TL (True Latency)

... or ...

1 (Second) / 4.2Ghz (Single Data Rate) x 40 (Column Latency) = 9.524 nanoseconds (True Latency)

In the big scheme of DDR5 memory, while not exactly sloppy, 9.524 nS is nothing to write home about. Despite advancements in increasing memory band width (Frequency) over the past few decades, a True Latency of 8 nS has remained pretty much a hard wall for SDR, DDR, DDR2, DDR3, DDR4 and especially for DDR5. Here's some common values:

DDR 400 @ CL2 = 10 nS
DDR2 800 @ CL4 = 10 nS
DDR3 1600 @ CL8 = 10 nS
DDR4 3200 @ CL 16 = 10 nS
DDR5 6400 @ CL 32 = 10 nS

In addition to memory performance benchmarks, and for quick comparisons, 7.5 to 8 nS is absolutely top tier, while 8 to 8.5 nS is fast, 8.5 to 9 nS is good, 9 to 10 nS is fair, and 11 to 12 nS is loose and sloppy. Anything greater than 12 nS you shouldn't be purchasing, unless it's for a low budget build.

As CountMike has pointed out, larger capacity DDR5 kits inherently have looser timings, . For example, the afore mentioned 48GB DDR5 8400 kit is 40-52-52-134. which suggests that DDR5 memory technology has yet to reach full maturity.

Regardless, with a Z890 motherboard, you can manually downclock the memory to 6000, as Lutfij advised, then adjust the Primary Timing to 28-40-40-122, which will give you 9.333 nS. This is slightly faster in terms of True Latency than stock settings at 9.524 nS. Moreover, at this lower Frequency, you should also be able to reduce DRAM voltage from 1.4 to about 1.38 or perhaps even as low as 1.36 to 1.35 volts.

You could also try 6000 @ 27-39-39-121 which would give you 9.0 nS (5.82% faster than stock), but it might require the voltage to remain at 1.4. Of course, running several passes of various memory tests will be necessary to validate the stability of the most ideal settings with the lowest possible True Latency.

As long as it's comparatively cost effective, I have used this same approach to install memory that's faster than the CPU's IMC is capable of running, downclock enough for IMC compatibility and stability, then tighten the timings for a better True Latency result at a lower voltage, while running a few degrees cooler.

Additionally, this approach will allow you to use the same memory kit on a future CPU upgrade with a faster IMC.

CT