Memory experts, please help with timings

[tranceaddict]

So far this is what I have narrowed it down to:
http://img.photobucket.com/albums/v241/Tranceaddict69/untitled-1.jpg

- tRP is not stable at 4 even with 2.3V
- tRAS can be lowered below 10 but I read that it should be tCL+tRCD+2, is that true?
- the other sub-timing I am not too familiar with so any help would be greatly appreciated.

BTW, even though this RAM is spec. @ 2.3V, I am running this at 2.1V and it is just as stable.

[esveezy]

I'm not claiming to be an expert, but I can tell you some things that helped me.

1.) What are you using to declare it stable? I have found that P95/Orthos Large-FFT (and probably Blend as well) will find RAM instability that Memtest will not. I've also heard HCI Memtest works great.

2.) Use a program called SPDTool to read the SPD data from your modules (File -> Read, then select your module, then click the Timings Summary tab on bottom). That will give you a reference point.

3.) Don't skimp on the tRFC - that is the one that always gets me, and your Asus is going to limit you to a max of 42 (unless I'm wrong). I'd be impressed if you were stable @467 with 35 tRFC, especially with those tight main timings, unless those 1GB sticks are that much better than the 2GB ones.

good luck

[tranceaddict]

I passed Memtest86+ for 8h+ but prime failed after 4.5h. I suspect that the DRAM voltage is too low. These sticks run @ 2.3V stock and I had it down to 2.1V. I am testing it with 2.2V now and see how that goes.

[tranceaddict]

I was right, once the DRAM voltage was raised to 2.2V, it passed 10h+ of Prime testing.

[||Console||]

and just because your timmings are lower doesnt mean they are faster . 4-4-4-12 @ 478mhz is faster than 4-4-4-8 (useing everest mem read test )

[tranceaddict]

and just because your timmings are lower doesnt mean they are faster . 4-4-4-12 @ 478mhz is faster than 4-4-4-8 (useing everest mem read test )

I'm not sure what your point is, @ 478MHz my CPU would be at over 4300MHz which is not stable. I reached my max FSB/CPU speed, and now I am trying to make the tightest timings possible while still maintaining a 1:1 ratio.

[||Console||]

Im just making a point that just by lowering timing it doesnt mean it is faster
you must test it to see if the changes you are making are helping or hurting

[esveezy]

Console is right. With RAM timings, you can get slower with tighter timings (not usually, but still) and you are delaing with microscopic performance differences anyway. RAM just isn't as sexy as the heatspreaders would have you believe.

Aim for stability with reasonable timings, and spend your time tweaking other things.

[Sizmo]

Console is right. With RAM timings, you can get slower with tighter timings (not usually, but still) and you are delaing with microscopic performance differences anyway. RAM just isn't as sexy as the heatspreaders would have you believe.

Aim for stability with reasonable timings, and spend your time tweaking other things.


Yea isnt each step (number) only a few nanoseconds for ddr2?

I think it was something like: in ddr1 your first reading was 6ns (2-3-3-7 w/e) 2x6ns=12ns. with DDR2 4-4-4-12, each value of 4 is only 3ns (cut in half) ~ 4x3ns=12ns. So! going from 4 to 5 in ddr2 is only a difference of 3ns, where as back in ddr1 going from 2 to 3 was 6ns.


I think it works something like that: Dont hang me if im off a little!

[jason4207]

and just because your timmings are lower doesnt mean they are faster . 4-4-4-12 @ 478mhz is faster than 4-4-4-8 (useing everest mem read test )

Lower is not necessarily better (possibly due to recoverable errors?), but usually is. I had opposite results than you. Everest gave me better scores w/ 4-4-4-8 than it did w/ 4-4-4-12 @ 450MHz.

*Everest scores can vary run-to-run, so it's a good idea to do a few runs at a given setting to get a more accurate reading.

I can also run tRFC at 25...even 20, but 20 gives me an occasional BSOD. I've never heard of that formula for tRAS, but I always thought that twice the CAS sounds good. I just go as low as possible while using Everest as a guide, and then raise it a bit if I have issues. I was able to get 6 to run, but I'd get occasional BSOD's, so I raised it to 8.

You can see the timings for 4x1GB in my sig for reference. I can go a little bit tighter on some settings w/ 2 sticks. Most notably, Performance Level = 6.

RAM tweaking can be fun, but it's not worth the time spent for most people.

[tranceaddict]

Lower is not necessarily better (possibly due to recoverable errors?), but usually is. I had opposite results than you. Everest gave me better scores w/ 4-4-4-8 than it did w/ 4-4-4-12 @ 450MHz.

*Everest scores can vary run-to-run, so it's a good idea to do a few runs at a given setting to get a more accurate reading.

I can also run tRFC at 25...even 20, but 20 gives me an occasional BSOD. I've never heard of that formula for tRAS, but I always thought that twice the CAS sounds good. I just go as low as possible while using Everest as a guide, and then raise it a bit if I have issues. I was able to get 6 to run, but I'd get occasional BSOD's, so I raised it to 8.

You can see the timings for 4x1GB in my sig for reference. I can go a little bit tighter on some settings w/ 2 sticks. Most notably, Performance Level = 6.

RAM tweaking can be fun, but it's not worth the time spent for most people.

Thanks for focusing on my actual question. As for the others, I understand RAM timings do not carry much weight, but I already maxed out my CPU/FSB/VGA so this is the only thing left to do. Anyways, I just finished stability testing at 4-4-5-8 and it's good. I just have one more variable to play with, as I maxed out all the other sub-timings, and that is the "performance level". Can you please give me a little more info on that if you can, as I am not too familiar with it. Thanks

[jason4207]

Yea isnt each step (number) only a few nanoseconds for ddr2?

I think it was something like: in ddr1 your first reading was 6ns (2-3-3-7 w/e) 2x6ns=12ns. with DDR2 4-4-4-12, each value of 4 is only 3ns (cut in half) ~ 4x3ns=12ns. So! going from 4 to 5 in ddr2 is only a difference of 3ns, where as back in ddr1 going from 2 to 3 was 6ns.


I think it works something like that: Dont hang me if im off a little!

You have to look at the speed the RAM is running. RAM running at 500MHz is switching every 2ns (1second/500,000,000cycles/second). The timings are the number of cycles the RAM has to wait after performing a specific operation before it can perform a similar operation, so a CAS of 5 means you have to wait 5 x 2ns = 10ns. A CAS of 4 would be 8ns. Only a 2ns difference, but when you consider millions or billions of tasks then the difference is a lot greater. 10ns is 25% slower than 8ns, and 8ns is 20% faster than 10ns...depends on how you look at it.

Does this mean your PC will be 20% faster...NO. That above equation assumes the worst case. The system may not have to wait on any timings if the RAM hasn't just been asked to do a similar task, and if the needed memory has already been loaded into the CPU cache then the wait is a lot less.

Getting every last ounce of performance out of your RAM isn't going to make COD4 run better, but if you're a numbers guy like me then you'll appreciate the fact that you know your RAM is performing optimally. It's mostly a placebo effect, but my PC does 'feel' faster to me!

:beer:

[jason4207]

Thanks for focusing on my actual question. As for the others, I understand RAM timings do not carry much weight, but I already maxed out my CPU/FSB/VGA so this is the only thing left to do. Anyways, I just finished stability testing at 4-4-5-8 and it's good. I just have one more variable to play with, as I maxed out all the other sub-timings, and that is the "performance level". Can you please give me a little more info on that if you can, as I am not too familiar with it. Thanks

Performance Level has a fairly large impact from what I've read. I'm no exactly sure what it actually does, but just the name, 'Performance Level', sounds like it does a lot! Lower is definitely better. In my BIOS the settings 'Static Read Control' (SRC), and 'Transaction Booster' (TB) control it. After a thread I read it became clear to me that it is pretty important...can't find it right now.

Anyway, I read that you should leave SRC disabled, and just use TB and the relax/boost options to control it. Which strap you have set, and how many sticks of RAM you have installed will determine what PL level is set w/ each TB setting. When I had 2GB the boost level set to 2, and was getting PL = 6 in memset. When I went to 4GB I had PL = 8 show up in memset. I had to increase boost level to 3 just to get a PL of 7. If I try to tighten it to PL = 6 in memset, I get a lock-up.