RAM OC/Timing Order & Why?

[Daharen]

So conventional wisdom across the guides I've been reading seems to be to 'work your way down' when trying to maximise your timings. That is do your primary timings first, then secondary, and tertiary last. Quite a few articles, blogs, posts, and threads do discuss the fact that primary timings sometimes can actually be tightened further after you do your secondary and tertiary timings... But then, if you do that, you also need to work your way back down all your secondary and tertiary timings to ensure their as tight as they can be, as that change can sometimes also provide more room for further tightening from those timings...

Now... This 'implies' to me that it should be possible to do timings in reverse order, that is tertiary first, then secondary, and primary last... I don't know if there is a benefit to this or not, I haven't seen anyone who says they did it, I haven't seen any explicit reason not to do it, and I haven't seen any good explanation as to why to do it in the order convention suggests we ought to. Since I'm trying to get the best latency I can at a reasonable frequency out of my RAM, I'm already certain I'll be making at least two passes over all the timings, but I am also wondering if the reason it's done the way people modify it currently is purely because of convention, or if there's some good mechanical or causal reason that going from primary to tertiary timings leads to tighter results?

[Jurrunio]

5 minutes ago, Daharen said:

This 'implies' to me that it should be possible to do timings in reverse order, that is tertiary first, then secondary, and primary last

possible but not the good way. Primaries have bigger performance impact than secondaries, then tertiaries. 

[Daharen]

Also, while I'm on this topic, my tRAS is currently 38, and my tRP is 18... I keep hearing for stability purposes, that tRC shouldn't drop below tRAS + tRP... However, I've done 24 hours of memtest86 and 12 hours of Prime95 with zero errors with my tRC at 41, which is obviously less than those combined... So what gives with this calculation? Is there something I'm missing here?

[Daharen]

2 minutes ago, Jurrunio said:

possible but not the good way. Primaries have bigger performance impact than secondaries, then tertiaries. 

Will tighter tertiaries and secondaries not be possible because of loose primaries though? The way I look at it, you're right, the primaries are definitely more performance impacting, but the principle only holds true if the 'stable bounds' of tertiary and secondary timings are determined by primary timings both down AND up... That is, looser primary timings prevents tighter secondary and tertiary timings... If they don't, and if tight secondary and tertiary timings make primary timings able to become 'tighter' then you will get more out of your primary timings AFTER tuning your tertiary and secondary timings...

In other words, which direction do the dependencies flow? 

[Bananasplit_00]

Ever worn a slalom boot? if you have you might understand that its possible to max out how tight everything is at once but its a lot fucking harder than doing them gradually.

 

Also training is a thing, there are almost always some timings set to auto that the motherboard/imc need to figure out on its own and going gradually helps shove it in the right direction afaik

[Daharen]

2 minutes ago, Bananasplit_00 said:

Ever worn a slalom boot? if you have you might understand that its possible to max out how tight everything is at once but its a lot fucking harder than doing them gradually.

 

Also training is a thing, there are almost always some timings set to auto that the motherboard/imc need to figure out on its own and going gradually helps shove it in the right direction afaik

I mean that does make sense, assuming that tertiary and secondary timings are bound on both sides (From getting larger or smaller beyond a certain range) based on the primary timings... However, if the training is only trying to find the lowest stable configuration in a quick and dirty fashion, and the primary timing stability actually relies on how low those secondary and tertiary timings are, then it makes a difference still... IE, is the training more to compensate for the differences in hardware interaction and within model silicon quality (IE not variable once all hardware and silicon is determined), or is it also equally bios dependent?

[Jurrunio]

8 minutes ago, Daharen said:

Will tighter tertiaries and secondaries not be possible because of loose primaries though?

I dont think so, but why would you use loose primaries?

 

12 minutes ago, Daharen said:

That is, looser primary timings prevents tighter secondary and tertiary timings... If they don't, and if tight secondary and tertiary timings make primary timings able to become 'tighter' then you will get more out of your primary timings AFTER tuning your tertiary and secondary timings...

haven't tried that myself, but logically that's not what you should do. You wont loosen secondaries and tertiaries all the way to the floor just to get the best primaries either right?

 

Think of it as relationship between core frequency and uncore frequency. 5GHz core 4.7GHz uncore is definitely better than 4.7GHz core 5GHz uncore (if you can actually do that), but 5.2GHz core 4GHz uncore will be slower than both. You keep a balance by not delibrately making things worse than auto

 

Oh and among secondaries, make sure tREFI as high as possible. Higher tREFI means less time refreshing, which is a period of time that it can't do work.

[Daharen]

11 minutes ago, Jurrunio said:

I dont think so, but why would you use loose primaries?

 

haven't tried that myself, but logically that's not what you should do. You wont loosen secondaries and tertiaries all the way to the floor just to get the best primaries either right?

 

Think of it as relationship between core frequency and uncore frequency. 5GHz core 4.7GHz uncore is definitely better than 4.7GHz core 5GHz uncore (if you can actually do that), but 5.2GHz core 4GHz uncore will be slower than both. You keep a balance by not delibrately making things worse than auto

 

Oh and among secondaries, make sure tREFI as high as possible. Higher tREFI means less time refreshing, which is a period of time that it can't do work.

That makes sense, obviously to some level the primaries do always need to come first, as I could obviously tank my frequency, and skyrocket my primaries, and then do the tertiaries and secondaries extremely stable at ridiculously low timings, but doing so is meaningless relative to the values I selected...

I suppose the best method is still top to bottom and then repeat to get it tighter. 

[Benjeh]

You need to set your desired speed first as the motherboard will auto train all you other timings loose to start, sadly sometime you will need to manually set up your timings when aiming for the max speed you wish to hit but this becomes a trade between latency and speed.

 

once speed is set you will need to juggle the primary timings together, you will get a feel of what works and what doesn't based on experimentation, you will now experience failed posts or long posts or failures to get into windows, which will require you play with the 3 ram voltages to find what you're lacking. secondary timings are next but unlike previous ram it doesn't massively effect DDR4 i'm told at least, and to be honest in my experience it hasn't effected me too bad.

 

terciary timings do have a decent impact however and should't be ignored but you will need to get a good feel for these also, there are some equations to help you find a decent setting but they're in guides.

Uncore will help your memory read and writes and copy etc but by how much i'm not sure, you should never sacrifice raw core speed for uncore.

 

Quote

Oh and among secondaries, make sure tREFI as high as possible. Higher tREFI means less time refreshing, which is a period of time that it can't do work.

 You need to be careful with this point, it can create a great deal of issues if the motherboard can't support the timing set, while I agree in theory, its worth keeping in mind that memory should be stress tested as much as possible in order to ensure the tREFI window isn't staying open too long to cause issues.