Overclock.net - An Overclocking Community - View Single Post - ROG Crosshair VI overclocking thread
View Single Post
post #21451 of (permalink) Old 06-29-2017, 07:06 AM
New to Overclock.net
kaseki's Avatar
Join Date: Jan 2017
Posts: 721
Rep: 75 (Unique: 37)
Originally Posted by ItsMB View Post

Originally Posted by kaseki View Post

I think this message can prove helpful to new overclockers here. However, I need to point out a couple of related things. When The Stilt referred to the real-time-termination loadings as {disabled, disabled, RZQ/5} I don't think he meant that the first two parameters were zero ohms. More likely 'disabled' sets them at infinity ohms.

To high a terminating resistance does not increase circuit "stress." It may degrade signal to reflected signal ratio, just as too low a terminating resistance may also degrade signal to reflected signal ratio. This is also true of procODT.

What specifically did you find in the B-die specs about RTT values? I looked and failed to find anything, perhaps from data saturation ennui.


This is a message in general, xd. I know that Proct is a termination resistance an does not have any effect on the current, but im showing my results llooking at voltages and temperatures cause i dont know the values that and specific procdt value can influde after a train, i know is just for noise etc.. like on any comunication bus, industrial, or any ethernet etc... More or less that i try to say is that i dont have enough information about all this stuff.

The Rtt values maybe are that infinite i dont know what he mean, maybe a disabled is infinite. I mean that 48 is more close to 0 than 240, its only a conclusion that can be wrong, xD

The B-die document about im talking is the Spec samsung support file on samsung site about our chips, but they only talk about speed to 2666 max, so 3200 is not specified in any caculation. I might be provided by Gskill to amd, and from amd to asus i guess, this is why im talking about the responsability of the vendors of suplly the proper specs. Excuse my confusing words im not a perfect english speaker, and imagine the expressions, xDDDD

I found that I had another different Samsung document than I referred to earlier. An extract of the state table for RTT is below. One can see that with The Stilt's recommendation {nom disabled, wr disabled, park RZQ/5} that RZQ/5 is always used. RZQ/5 is 48 ohms per this document. This value might be expected to reasonably match likely trace impedances, although what the relevant C6H trace impedances are is unknown to me.

Conversely, more or less, the Ramad/Jackalito/HarrySun values of {nom RZQ/3, wr RZQ/3, park RZQ/1} when applied to the state table yield a termination resistance that fluctuates. I'm not sure what advantage that would have.

I'm not certain what your recommendation of {auto, auto, auto} actually does, so I'll just add that to my test matrix. Edit: Discovering an earlier message I missed, I think this is The Stilt's {disabled, disabled, 48 ohm} condition.

For reference, I am presently running RZQ/2 on 1403 for all three parameters from a month-ago recommendation by BoMbY. He didn't answer my question about his rationale for these values. My present thinking is that these impedances may be too high for good matching.

And then there is The Stilt's spasm version of 1403 still to test. Sigh.

from DDR4 SDRAM Specification, Device Operation & Timing Diagram, 2014, page 178

The circuit diagram (simpified, I'm sure) extracted below suggests why there may be a change in DRAM power usage. The lower the terminating resistor, the more current flows to the signal lead and thence to eventually to ground. There are a lot of terminating resistors.

Rig running Ryzen [email protected] GHz, [email protected] MT/s, Linux Mint 18.1 on Asus C6H
kaseki is offline