Duckydude

For quite I while now I have been debating whether to run my memory at tight timings or at high speeds. I decided to do some benchmarks myself to find out the answer. I ran four different benchmarks: SiSoft Sandra Memory bandwidth test, Everest Memory and Cache test, Superpi 1M Calculation and 3DMark06. I kept everything constant except the memory speed and timings for each test. I decided to run my graphics card at stock settings (700/2100) for the 3Dmark test to avoid potential hassles because sometimes the settings won't stick, etc. For each test that I ran, I set my memory to several different settings:

Super Tight Timings I: Speed: 667Mhz Timings: 3-3-3-3 Command Rate: 1T

Super Tight Timings II: Speed: 800Mhz Timings: 3-3-3-3 Command Rate: 1T

Super Tight Timings III: Speed: 833Mhz Timings: 3-3-3-3 Command Rate: 2T

Tight Timings I: Speed: 1000Mhz Timings: 4-4-4-4 Command Rate: 2T

Tight Timings II: Speed: 1066Mhz Timings: 4-4-4-4 Command Rate: 2T

Stock Settings: Speed: 1066Mhz Timings: 5-5-5-15 Command Rate: 2T

Moderate Timings: Speed: 1112Mhz Timings: 5-4-4-4 Command Rate: 2T

I decided to post my results for the stock settings appart from the others.

Alright let's get started on the tests:

Sisoft Sandra:


Everest Memory Test Benchmark Table:


Everest Memory Test:


Everest Memory Latency Test:


SuperPi 1M Calculation:


All benchmark tests ran at stock configuration:


I must note that these 3DMark06 tests were perfomed with a different processor overclock than
the rest of the tests. These test are just trying to show that memory speeds or timings do not have any effect in gaming performance.

This test was done with the following settings:
Speed: 756Mhz Timings: 3-3-3-3 Command Rate: 1T
3DMark06 (Default Settings and Video Card Was Left On Stock Settings):


This test was done with the following settings:
Speed: 907Mhz Timings: 4-4-4-8 Command Rate: 1T
3DMark06 (Default Settings and Video Card Was Left On Stock Settings):


This test was done with the following settings:
Speed: 1210Mhz Timings: 5-5-5-8 Command Rate: 2T
3DMark06 (Default Settings and Video Card Was Left On Stock Settings):


As you can see the higher speed has a greater effect (with intel systems) when testing with memory benchmarking software. As you can see from the 3DMark06 scores. memory speed and timings doesn't really affect the performance in games at all (Most of you probably already know this). Although it is interesting that the 3DMark06 score is a little higher with my RAM at 1T, 907Mhz, 4-4-4-8 timings; I thought it was pretty remarkable to get my RAM to even boot at 1T, so I think this is the reason behind that score. The Latency scores were all over the place, so I didn't really factor them in to my conclusion. Although there are some slight variations in the scores, the main consensus taken from these benchmarks is that higher speed wins over tighter timings (With Intel Systems At Least). Unless you can run your RAM at some crazy tight timings over a slightly higher speed, I would overclock your RAM at the higher speed. Thanks for listening folks and I hope that this was an interesting read.

__________________
Crucial Ballistix Club

My First Guide: The Perfectionist's Guide To Ripping Audio CD's

Check This Out (Intel Users Only):
The Definitive Answer To: RAM Overclocking: Higher Speed vs. Tighter Timings

rcdriver0604

Do a superpi run for me

__________________
Check out my photos http://flickr.com/photos/photosbyjacob/
What I need- Raised $100
1. ATH-A700's
2. Xtrememusic

Duckydude

Good idea.

__________________
Crucial Ballistix Club

My First Guide: The Perfectionist's Guide To Ripping Audio CD's

Check This Out (Intel Users Only):
The Definitive Answer To: RAM Overclocking: Higher Speed vs. Tighter Timings

Duckydude

SuperPi tests now added.

__________________
Crucial Ballistix Club

My First Guide: The Perfectionist's Guide To Ripping Audio CD's

Check This Out (Intel Users Only):
The Definitive Answer To: RAM Overclocking: Higher Speed vs. Tighter Timings

Cyant

Your 907Mhz 1T test is slower than 1200 because your FSB is not 1:1. You get more benefit with a FSB:Memory ratio of 1:1 as outlined here (a test I did a while ago at 425Mhz mem 4-4-4-16 1T with matching 425 Mhz FSB):

xenophobe

You didn't test at stock settings?

That throws objectivity out the window. Why? You have no baseline for comparison.

Because, you would see that the difference from stock timings to super tight timings to be a huge gap, while increasing speed and loosening up those timings only provides a tiny bit extra than just tightening the timings.

The difference in SuperPi from super tight timings to overclocked only netted you a 0.054 decrease in your time to completion, which is hardly noticable.

What's the difference in SuperPi from stock speed and timing to super tight timings? I bet that difference will be far greater.

DesertRat

quite possibly true. But we know that tighter timings or faster clocks will be better than stock and we can see that faster clock is better than tighter timings. SO we can conclude that faster clocks are better. However! You are completely right that you should have a baseline for comparison. If this we're to be a official review It'd need it

__________________

xenophobe

Well an extra 400MHz+ to the memory (over a 50% overclock mind you) only decreased SuperPi by 0.054 seconds. That increase is relatively meaningless.

How much did SuperPi decrease from stock timing/speed to super tight timings? Until that is shown, the 0.054 decrease has no real significance.

Duckydude

The only problem is that I can't run my memory at stock settings with this overclock and the only step in between is at 1167Mhz, which I can't get to even boot at cas 4, so it would sort of be pointless. Maybe I'll have to do it with my cpu at 3.0Ghz or something. What would you recommend the different steps in memory should be? And the 1:1 ratio thing is totally bogus by the way, sorry.

__________________
Crucial Ballistix Club

My First Guide: The Perfectionist's Guide To Ripping Audio CD's

Check This Out (Intel Users Only):
The Definitive Answer To: RAM Overclocking: Higher Speed vs. Tighter Timings

Cyant

Did you look at the screenshot by the way?