tRD values and how to set them

[newhit]

First, sorry for my lack of knowledge but,

I cannot get my head round tRD values and how you set them. I have read various articles, including the "Max xtrm" and "Rampage" ones together with the "Tango" article but I still wonder how I can physically set the actual tRD values referred to in the various articles.
Am I missing something? Maybe the fact that you must run "Memtest" to determine the performance level aka tRD?

I have not got any of the mbd's but I still need to know how the actual tRD settings are inputed.

Any help greatly appreciated

 

[nefariouscaine]

that is all motherboard & chipset dependent

not all boards will have that feature/function enabled in bios nor editable via software

with my 650i chipset that I love I have most latency settings available to me - this may or may not be your case

IIRC my tRD is called something "funky" in bios (most latencies have 1 or 2 names)

I can't say for sure with out more info &/or screen shots of your bios

 

[lopri]

Simply put, tRD value (selection) isn't open in many boards. It used to be decided by the CPU's base strap, FSB, and memory ratio and timing by the board's BIOS. It's understandable because it's a very sensitive value. Latest boards targeting enthusiasts (ASUS/DFI's offerings) finally exposed the direct manipulation of tRD value, if I'm correctly understanding.

But I don't think it matters a lot at the end of the day, because the value you can select will be limited just as it used to be before. (In other words, just because you have an option to choose tRD doesn't mean that you will be able to use that value every time you want) AT article's wise message is that raising FSB isn't the best way to achieve optimal performance, because as the FSB goes up you (or your board) will inevitably loosen the tRD value.

 

[bunnyfubbles]

well, its not that you can't change it to whatever you want, its that it won't be stable and/or it will require MCH frying voltage

 

[Sylvanas]

OP what board do you have? It depends on the board as different manufacturers call it totaly different things. If you do not have access to TrD via the BIOS, download Memset- the 'Performance level' option, to my knowledge, is TrD.

 

[nevbie]

If you have a P35-based ASUS board, my post here might be helpful: http://forums.anandtech.com/me...id=83&threadid=2160162

 

[nefariouscaine]

Originally posted by: Sylvanas
OP what board do you have? It depends on the board as different manufacturers call it totaly different things. If you do not have access to TrD via the BIOS, download Memset- the 'Performance level' option, to my knowledge, is TrD.

"Performance Level" is a board specific option in Memset

I went back into my bios and looked at my tRD again - I can't "see" it any other place but the bios and I've manually set it to 4

even though i can see the value I manually set in bios I have no other way of checking this via software *yet* that I've found

its not displayed under memset or everest but the remaining latency values seem to be there - unless it has multiple names I can recall...

I'm running at DDR2-1050 right now with 4 gigs of ballistix - settings 5-5-5-5 T2 / tRC 7 / tRD 4 / tRFC 30

all other remaining values are on Auto

I haven't played much with tRD - read that same article it looks like a few of you read too and decided to set it to 4 - it passed memtest fine with it so I left it as such. how tRD relates to performance might be related more to the specific chipset used than as a general thing.

Everyone has jumped on the tRD bandwagon since that ASUS Rampage Formula review - in theory my values should not be possible with the FSB at 375 and my 7/5 ratio for memory (according to the rules formula in that article) as 0.4286 > 1.4285 is false

 

[edrom]

Thanks for the link, nevbie to Memory and Storage : Clockspeed vs. Timings vs. Ratios.

I read the posts from you and others, and it helped me to understand part of what I'm wanting to understand.

I have some G.SKILL DDR2-1000 memory with timings of 5-5-5-15 but was planning to underclock it to DDR2-800 4-4-4-12. But according to the calculation put forth by wired247 in the thread that you listed, running it either way won't make any difference. Here is what I calculated:

DDR2-1000 operates at 500MHz thus 1/500e6 or 2.0ns per clock cycle
Timings of 5-5-5-15 would be (5+5+5+15)*2.0 = 60ns total

DDR2-800 operates at 400MHZ thus 1/400e6 or 2.5ns per clock cycle
Timings of 4-4-4-12 would be (4+4+4+12)*2.5 = 60ns total

Since I can operate my DDR2-1000 memory at either setting above then the only thing to pay attention to is how each of those settings fall into play with FSB strap setting.

Would that be correct? If so then this answers one of my questions.

I'm not trying to hijack this thread in any way, but I didn't see this thread when I posted here:
Motherboards : ASUS Rampage Formula, Q9450, strap, FSB, tRD, aaaahhhhh!

My question posted there (please read it all if you will) boils down to this:
Which of the below would be better and why?

1. Strap: 400, Divider: 1:1, Default tRD: 12, 400MHz FSB DDR-800 / 30ns
2. Strap: 200, Divider: 5:3, Default tRD: 6, 300MHz FSB DDR-1000 / 20ns

Since I can run the memory either at DDR2-1000 or DDR2-800 with identical results then I guess that the strap is the next thing to consider.

I think that related to all this is the processor speed, multiplier, and resultant system bus speed. But again I'm confused how this all fits together. 😕

I don't have to understand everything and I do plan to try different things on my own, check results with Everest, and try to decide what really works best for me. But I'd at least like to know what kinds of things I should be shooting for, that is, what kinds of things should I try that make sense.

Thanks for your help.

Edro

 

[newhit]

OK, thanx for the replies but there is still confusion.

I've got my hands on an Asus P5E for use with DDR2 and using the corresponding P5E3 article (although this uses DDR3) it seems that, from reading the paragraphs about "Ai Transaction Booster" I have to equate tRD to tCL (CAS Latency?) which should be simple enough. I will have to go with this but I suspect that I'm not seeing the whole picture.

 

[nefariouscaine]

Originally posted by: edrom
Thanks for the link, nevbie to Memory and Storage : Clockspeed vs. Timings vs. Ratios.

I read the posts from you and others, and it helped me to understand part of what I'm wanting to understand.

I have some G.SKILL DDR2-1000 memory with timings of 5-5-5-15 but was planning to underclock it to DDR2-800 4-4-4-12. But according to the calculation put forth by wired247 in the thread that you listed, running it either way won't make any difference. Here is what I calculated:

DDR2-1000 operates at 500MHz thus 1/500e6 or 2.0ns per clock cycle
Timings of 5-5-5-15 would be (5+5+5+15)*2.0 = 60ns total

DDR2-800 operates at 400MHZ thus 1/400e6 or 2.5ns per clock cycle
Timings of 4-4-4-12 would be (4+4+4+12)*2.5 = 60ns total

Since I can operate my DDR2-1000 memory at either setting above then the only thing to pay attention to is how each of those settings fall into play with FSB strap setting.

Would that be correct? If so then this answers one of my questions.

I'm not trying to hijack this thread in any way, but I didn't see this thread when I posted here:
Motherboards : ASUS Rampage Formula, Q9450, strap, FSB, tRD, aaaahhhhh!

My question posted there (please read it all if you will) boils down to this:
Which of the below would be better and why?

1. Strap: 400, Divider: 1:1, Default tRD: 12, 400MHz FSB DDR-800 / 30ns
2. Strap: 200, Divider: 5:3, Default tRD: 6, 300MHz FSB DDR-1000 / 20ns

Since I can run the memory either at DDR2-1000 or DDR2-800 with identical results then I guess that the strap is the next thing to consider.

I think that related to all this is the processor speed, multiplier, and resultant system bus speed. But again I'm confused how this all fits together. 😕

I don't have to understand everything and I do plan to try different things on my own, check results with Everest, and try to decide what really works best for me. But I'd at least like to know what kinds of things I should be shooting for, that is, what kinds of things should I try that make sense.

Thanks for your help.

Edro

in that case at the bottom of your post - the proposition of your entire system speed (ie 300mhz vs 400 mhz fsb) - regardless of what your memory is set at and even if you adjust your cpu multiplier to be running at close to the same speed your NOT going to be running at the same speed. Your memory *might* behave similarly but likely it won't.

Mainly due to a 33% difference in FSB speeds - general rule of thumb is the higher the fsb the "faster" the system. I've done my own testing on this and cranked up my fsb to 425mhz and even with my cpu clocked 175mhz less than what it is now that 50mhz fsb difference is enough for higher bandwidth thoroughput - now if you were running a different divider but still using the same 400mhz fsb the argument might be different.

Remember that your FSB effect your "entire" system clock - so the higher that is the more theoretical performance you will receive.

to kinda sum it up - latency adjustment generally isn't enough to compensate for a decreased FSB but can be for a decreased memory clock.

 

[edrom]

Thanks, nefariouscaine.

My fog is beginning to clear on this, at least enough so that I understand the kinds of things I should try when I get the new CPU and mobo hopefully later this month.

It will be fun trying different things and seeing what they yield.

Edro

 

[viqarqadir2]

On the other hand my fog is still thick and all encompassing 🙂

Let me know if you already haver figured it out and what kind of settings are you using. I think if you want to run the Q6600 in overclock, the rest of the setup just crumbles. I've seen more BSODs in the last few days than ever in my short but eventful life.

One interesting thing I noticed is that if I dont overclock the CPU and set the timings on the memory following the article. the system is ridiculously fast, Vista loads in 20 seconds and responds like the good old 98. 3dMark went crazy on the memory and showed a speed of 1.9 Ghz (!!) What is that all about though?

The overall mark with me 8800 GTX was about 11000 something. I dont trust the score though.