Timings vs. MHz

[18 is # 1]

Since the question/misperception of timings vs. speed (MHz) is frequently asked/misstated, I wanted to start a thread to give a simple example of why speed is most important. I will keep my comments directed to those of us who run Intel processors and chip sets with DDR2 since that can keep the generalizations from being too generalized.
Theoretically the same FSB will work with RAM to create the same performance such as this:
475 FSB 4-4-4-12 1:1 ratio 950 MHz tRD 12
http://i383.photobucket.com/albums/oo280/rkduncan/4-4-4-122x2.jpg

And this 475 FSB 5-5-5-15 5:6 ratio 1140 MHz tRD 12
http://i383.photobucket.com/albums/oo280/rkduncan/5-5-52x2PL12.jpg
And as you can see the differences are not great.


BUT-
The looser timings allow is the tRD or Performance Level (PL) to be lowered, allowing significant gains in bandwidth: 475 FSB 5-5-5-15 5:6 ratio 1140 MHz tRD 8
http://i383.photobucket.com/albums/oo280/rkduncan/5-5-52x2.jpg
Read, Copy, and Latency all improve and SuperPi gains .08 seconds which is almost a half cycle (counting to 1M).
My old Abit Pro mobo changed the tRD automatically and didn't provide an option for manipulation.
I haven't found many who don't gain from increasing their RAM speed. "Real World" results depend upon your applications, but empirical results are almost uniformly positive.

More on tRD here: http://www.ocztechnologyforum.com/forum/showthread.php?t=36361

[mastrdrver]

Note: I've read the articles multiple times over on OCZ several times but there are a few things I'm running into that I havn't seen anyone comment on. Maybe I'm just missing it and not catching it when reading them(?).

I've been wondering if there is a relationship between memory timings, fsb strap, and tRD?

I've run into problems where if I run the memory at CL6 with certain fsb over 400 and memory other than 1:1, I get freezes and bsods. I do it to try and keep the memory loose and keep from causing problems, but keeping it loose seems to cause the problem. I can play around with the tRD, but it doesn't help. Also, if I switch the memory back to CL5, it seems the problems go away. Increasing\decreasing vDimm doesn't help.

I know I don't have the greatest memory, CL5 800mhz XMS2, but as long as I keep them at CL5, up to 920 now, I don't run into the problems. It only seems to show up if I loosen the timings to CL6 or 7. I've noticed also that having the tRFC too loose can cause problems too as backing it down will fix it.

Example: 333 strap 1:1, 12-14 tRD, CL6, and memory over 410(can't remember exact) causes problems. Switched to 400 strap 1:1 helps, but still there. Change to CL5 and problems disappear.

[Shiggity]

Nice post :thup:

I always thought tighter timings / latency improved the 'snapiness' of things and more Mhz gave you more bandwidth throughput / IO.

Since almost nothing in the desktop environment really requires a huge amount of memory bandwidth, I always figured it was better to have tighter timings.

Memory guru's, am I mistaken in those assessments?

[Neuromancer]

Looks right to me shiggity.

timings vs speed, is a platform specific as well as use specific.

MOST applications see more benefit from tighter timings than higher bandwidth running a everest memory benchmark is only useful in determingin which has better bandwidth. (the latency of each setup is justa little number tucked on the side as if it is unimportant.)

Intel rigs definitly prefer memory bandwidth. They make up for a higher latency by getting lots and lots of memory throughput. AMD systems like tighter timings.

General desktop usage benefits (noticeably) from tighter timings and it is why I am back on AMD after a brief hiatus playing with the Core2 series. Snappy desktop and application loading, like I have not seen since my first 939 setup (winnie 3200 at 2700 ish and 250MHz 2-2-2-5 Geil ram, my everest latency test was 35ns. Was a very "snappy" system.

Thought it was the switch to DDR2 that caused the loss of quickness. Even my 3Cas 667 Team Group ram did not feel quick on my Q6600/Q9300 or numerous E series chips. However, moved back to AMD with phen and phen two and 4-4-4-10-12 800mhz at 1T.. even stock CPU I feel the speed again. Benchmarks may favor intel, but the nose knows.

[MongGrel]

Actually I agree, I've personally started setting things up more and more on my personal preferences on how things work a bit lately for my system.

Benchmarks are nice in general but just using things yourself and tweaking here and there works also :)

:beer:

[freeagent]

if the ram is up to it, i prefer mhz over tightness :)

actually, i prefer fsb, and mhz on the stix :)

[18 is # 1]

I always figured it was better to have tighter timings.
Memory guru's, am I mistaken in those assessments?

With your set up, MHz should increase your performance over timings. Measure and see. Everest, SuperPi, and Sandra are benchmarks to compare your settings. There is not much you can do to get empirical data from everyday applications.

[slap]

text

I've been wondering if there is a relationship between memory timings, fsb strap, and tRD?

I've run into problems where if I run the memory at CL6 with certain fsb over 400 and memory other than 1:1, I get freezes and bsods. I do it to try and keep the memory loose and keep from causing problems, but keeping it loose seems to cause the problem. I can play around with the tRD, but it doesn't help. Also, if I switch the memory back to CL5, it seems the problems go away. Increasing\decreasing vDimm doesn't help.




A good read at anand http://www.anandtech.com/mb/showdoc.aspx?i=3208&p=9 about tRD, nice article.