Timing rules

[Velvet Wafer]

From what i was able to gather, most professional overclockers that know the secrets of ram, tend not to tell their tricks regarding that, because it seems to be one of the hardest things to get right

 

[Sasqui]

Nice thread, subb'd.

I suspect, like everything... there's a tradeoff between secondary timings and speed (and stability) + the interelationships with other timings.

If there is a general theory of relativity for DRAM timings, I'd like to see it... after reading this, I doubt it's linear like x+x+x+x, there's probably some x^(y-z) stuff going on.

 

[d1nky]

theres a couple guys on this site that seem to masters of ram, wonder if theyll chip in.


and i doubt its a secret or something to be hidden, unless its used in competition.


ram manufacturers must have a formula like sasqui said above. even a laymans guide would be good.

 

[cadaveca]

You're not going to see any amazing benefits from changing secondary or tertiary timings, even if you know exactly what you're doing.

You're looking for an easy answer where these is none. I'm just saying.

I see almost 25% more memory performance from tertiary timing adjustments at lower and at higher speeds, depending on the board used. Does it affect overall performance? marginally.


Not all boards use the exact same labels for timings, either, so that complicates things.

As to formulas for memory timings, there is not one for general use. finding out what each IC likes is easy, simply go to the IC OEM's website, download the whitepapers, and decipher the mess inside. It's not really that hard, for me.

Like Samsung DDR3 ICs...when I got them, I read the provided info, and found that there is a relation between CAS Latency, and tWCL. Nobody was talking about tWCL is any way, on any forum I could find. Nothing in extreme sites, nothing in normal sites...so I posted some info in my review on what I found during testing.


Today, you'll find countless mentions of tWCL on extreme forums, looking at efficiency tweaks or pushing a stick higher. Did I point that setting out? Maybe..but probably not. If you took the time to adjust each individual setting, you would have discovered the same just as easy.

theres a couple guys on this site that seem to masters of ram, wonder if theyll chip in.

So...yous wants dat timings guide, eh? Imma goona tell you, it doesn't exist, and that it changes with every platform, too. Intel SNB and IVB like some settings, Haswell, completely different. AMD, yet another set of settings to worry about.


I should start giving courses, with high entry fees.


Dumo is a hidden Guru on this stuff. That guy..his ram...OMG...

 

[d1nky]

So...yous wants dat timings guide, eh? Imma goona tell you, it doesn't exist, and that it changes with every platform, too. Intel SNB and IVB like some settings, Haswell, completely different. AMD, yet another set of settings to worry about

wasnt look for the answer, just direction.

seems a bit alien going in and lowering numbers thinking they have an effect when actually they could be doing nothing or worse.

i like what you said about the manufacturers schematics.

guess its in the art of tweaking and finding out for ourselves.

 

[cadaveca]

guess its in the art of tweaking and finding out for ourselves.

Yes, very much so. I could give you general guidelines, as I have for years for many members here. From general user to forum moderators, there are many people here running timings I gave them.

Like MXPhenom...his sticks wouldn't work with his board at all until I told him what timings to change...and rather than changing things to make them looser...or increasing voltage, I had him TIGHTEN secondaries. Just three settings, set tighter, and everything worked perfectly.


I don't think I'm all that bright, honestly, so if I can figure this out ,anyone can, but as you said, it does involve devoting some time and attention to it. I have literally 1000's of pages of notes from memory clocking over the past 10 years or so. Memory is what I like to play with. I got into reviewing boards, since some boards offer better options than others, and what better way to get the boards I wanted to overclock my memory with?

I've recently started to focus more on memory, and less on boards. I have about 20 Z87 boards for review, and then I might completely switch to memory only, while going to school in the fall. Memory just offers me too much fun. I'd rather have one excellent board and 5 or 6 good memory kits to play with, than all these motherboards.

 

[m1dg3t]

Dumo = RAM God!

I hate messing with RAM, frustrating as all hell to max out

 

[de.das.dude]

wow, so much math. i just try the timings for the next lowest clock. if it doesnt work, i try upping one by one

 

[LAN_deRf_HA]

I feel like there must be some catch to this tRFC thing. I dropped it from 171 to 88 and it's appears to pass memtest. Should I just take it to the floor or can it damage something?

@cadaveca
I changed my tWCL down a notch after seeing your samsung review mention that. Even though I doubt I have Samsung chips it seems whenever I look up what timings to try I end up in a thread about those chips and what works for them seems to work for my sticks.

 

[TheoneandonlyMrK]

I personally tweek for hours and generally fall upon best settings to be honest I think I do ok but on my amd rig that's taking some time as is the heat tradeoff in cpu max clock v memory , how high would you all push the voltage on 1.5v ram reasonably well cooled.

Oh and thanks I had not thought of actually looking up the ic info doh.

GDAMIT now I wanna crack my memory open.

 

[cadaveca]

@cadaveca
I changed my tWCL down a notch after seeing your samsung review mention that. Even though I doubt I have Samsung chips it seems whenever I look up what timings to try I end up in a thread about those chips and what works for them seems to work for my sticks.

Yeah, it seemed really odd to me that this crucial timing was something not looked at or at least talked about until after I posted that review. Obviously the OEMs knew, or they wouldn't have included the option as adjustable.

and yeah, the same rules, in general, apply as you scale clocks up, it's when you start to hit the limits that what you have to tweak changes, or if the IC density changes, like going from 2 GB sticks to 4 GB to 8 GB requires you adjust things as well.

 

[d1nky]

can we keep this thread alive for RAM timings and alike. turn into some informational help each other out kind of thread.?! and i like posting links that i find good reads!


well i was thinking last night, and if hypothetically i manufactured some ram from scratch how would i create timings profiles for such. id guess most manufacturers have templates etc.

but after testing the ram for its maximum voltage handling/speeds etc. they get binned based on the tests.

so if your ram is X speed, from that the mb/s (transfer rate) is worked out and that sets out a formula to work the timings.

now the timings act like a set of instructions for delays and precharging cells in order for data to be transferred, based on the X speed/timings.

what ive learnt so far.... ive seen something for RAM profiling and it looks like physics... maybe DRAM physics LOL

http://www.hardwaresecrets.com/article/Understanding-RAM-Timings/26/3

http://www.hardwaresecrets.com/arti...o-Know-About-DDR-DDR2-and-DDR3-Memories/167/5

 

[LAN_deRf_HA]

I tried tRAS at half the formula recommend speed thinking it might not cause any stability issues if it's running at twice the speed instead of something irregular. I was at 9-10-10-28, tried 9-10-10-14 and it worked fine, but I can report zero performance benefit. So I'd say only bother adjusting tRAS for the sake of stability at higher overclocks.

 

[d1nky]

i havent been paying much attention lately, but if i get my ram to 2550mhz with slightly looser timings my performance is a lot better.... and benching is phenenenenenemonal, was able to hit 10k physics 3dmark firestrike, but stock speed/timings no chance.


after i finished setting a waterloop i want to learn about this and hopefully come with some sort of strategy.

i messaged gskill asking if theres things i should and shouldnt do, they replied saying its mainly in the major timings and didnt help lol

 

[Aquinus]

I tried tRAS at half the formula recommend speed thinking it might not cause any stability issues if it's running at twice the speed instead of something irregular. I was at 9-10-10-28, tried 9-10-10-14 and it worked fine, but I can report zero performance benefit. So I'd say only bother adjusting tRAS for the sake of stability at higher overclocks.

I would imagine that tRAS would improve the amount of latency to read data out of a DRAM page that the IMC isn't currently on. It's the amount of time (in clocks) to change which DRAM row (page) is being selected. This only impacts data that is accessed from that page for the first time, subsequent reads from the same page don't get impacted by tRAS.

So even though your bandwidth might be unchanged, your latency may have dropped as a result but only during random reads from different DRAM pages.

 

[LAN_deRf_HA]

I reran the tests again just for you. Multiple runs of Aida, maxxmem, and super pi all say nada. If I ran 32m all day long and averaged it out I could see something but I'd have to file that under too small to bother. Maybe it can help someone somewhere at some point but not me.

So far mhz, CL, tRCD, tRP, and tRFC are all I've found that increase scores on a large enough scale for me to confirm.

 

[d1nky]

i see the biggest improvements from upping my cpunb and with a bit more northbridge voltage.

changing speeds or timings help with the latency and slightly improve read/write/copy


does anyone know if its ok to copy the XMP profile for AMD chips? otherwise i have no settings to copy lol

 

[TheoneandonlyMrK]

i see the biggest improvements from upping my cpunb and with a bit more northbridge voltage.

changing speeds or timings help with the latency and slightly improve read/write/copy


does anyone know if its ok to copy the XMP profile for AMD chips? otherwise i have no settings to copy lol

Im using mine to get to 2145 so I doubt it can hurt plus following this thread ive messed with a fair few but I am not finding any better timings at this speed I did at 1333 , 1600 and 1800-2000 but not near my memories top frequency , mushkin got them pretty close.

Argh phones r crap.

 

[erocker]

does anyone know if its ok to copy the XMP profile for AMD chips? otherwise i have no settings to copy lol

Well... It works for some things but not for others. It's a good place to start though.

Just to add a couple cents to this thread I've found that sometimes one or two sub-timings can be the difference between a running and stable system and a system that won't boot at all.

 

[TheoneandonlyMrK]

i havent been paying much attention lately, but if i get my ram to 2550mhz with slightly looser timings my performance is a lot better.... and benching is phenenenenenemonal, was able to hit 10k physics 3dmark firestrike, but stock speed/timings no chance.


after i finished setting a waterloop i want to learn about this and hopefully come with some sort of strategy.

i messaged gskill asking if theres things i should and shouldnt do, they replied saying its mainly in the major timings and didnt help lol

What's your northbridge and ht frequency at that memory speed please and how's your memory cooled.

 

[erocker]

Leave HT frequency alone! It does nothing!!!

I found with Bulldozer and later CPU's anything over 2400mhz NB requires a lot more voltage if it's going to be stable at all.

 

[TheoneandonlyMrK]

Leave HT frequency alone! It does nothing!!!

I found with Bulldozer and later CPU's anything over 2400mhz NB requires a lot more voltage if it's going to be stable at all.

I know what your saying but due to other component speeds I can now run it higher or lower then spec but rarely on spec so I asked out of interest no need to shout
Plus as you well know pushing every clock you can as high as you can is a bit addictive but fraught with a new issue every try so high memory speeds might need lower nb speeds forsstability which is actually why I asked that and nb speed affects memory a lot.

 

[erocker]

It was just excitement!!! THIS IS SHOUTING!!!

 

[LAN_deRf_HA]

I got a consistent 0.002% performance boost by raising tREF from 8320 to 10400. Yes I had to go back and forth a lot to confirm that one. I picked that number because the highest I saw in reviews was 9360 which is a 1040 step so I just ran with that. I'm reasonably confident that's the end of the road for tweaks with my system at least.

 

[niko084]

I hope some of you have spent this much time clocking your gpu/cpu

I haven't spent a serious amount of time playing with memory timings since getting my first i7.
But now this makes me want to again, blast you benchmarkers... It's an addiction gotta push for the top!