Do memory timings really matter anyway? (Pentium 4)

[xtknight]

All tests ran with Quake 3 1024x768 High Quality

RAM: Corsair Value Select 256MBx2 Dual Kit (512MB)
Constants:
"Performance mode": Standard
Burst Length: 4
Memory clock rate: 200 MHz
Dual Channel Mode
Memory Acceleration (ASUS Hyperpath/Intel PAT): Enabled

You may think all the FPS differences are within the margin of error, so to make sure they weren't I did the following.
I ran one test 3 times and came out with the exact same score each time, to the tenth decimal (output accuracy), so essentially the EXACT same thing.

Disclaimer: I only tested this on one application, so your mileage may vary depending on which application or game you run!
Timing Format: CAS-RAS2CAS-RAS PRECHARGE-TRAS

Data

Timings = FPS
2.0-4-4-8 = 240.1
2.5-3-3-8* = 246.6
2.5-3-4-7 = 244.8
2.5-4-4-7 = 241.7
2.5-4-4-8 = 238.9
2.5-4-3-7 = 242.2
2.5-3-4-8 = 243.7
2.5-4-3-8 = 242.0

* Default SPD Value

Performance increase per -0.5 CAS : 0.50%
Performance increase per -1 RAS2CAS : 1.90%
Performance increase per -1 RAS PRECHARGE : 1.19%
Performance increase per -1 TRAS : 0.45%

Thus, we can conclude that the order of importance of the timings is as follows:
1. RAS2CAS
2. RAS PRECHARGE
3. CAS
4. TRAS

We can also conclude timings make little difference to performance, of games at least. A higher clock rate is much more valuable if all you have to do is increase one of the timings.

 

[xtknight]

Anyone have any comments?

 

[Dough1397]

lower = better overall...

 

[BlueWeasel]

Will these results generally translate into all games, or are these results game dependant?

 

[xtknight]

Originally posted by: BlueWeasel
Will these results generally translate into all games, or are these results game dependant?

Generally, this should translate into most everything, yeah. However, I'm not really sure, it's possible some of these are in margin of error but I hope not. Any CAS higher than 2.5 and my board wouldn't POST so I couldn't test it that way. Fortunately, I could go one CAS lower and see the increase in performance which was little at best.

 

[bobsmith1492]

Look at Zebo's benchmarks. They make a huge difference on the A64 platform. P-4s on the other hand - who knows??

 

[Shimmishim]

nice thread but if i were you, i'd run a bunch more benchmarks just like zebo did...

 

[Zebo]

try 2-2-2-7 1T and get back to us.

 

[xtknight]

Originally posted by: Zebo
try 2-2-2-7 1T and get back to us.

Once I get my new system in place I will be sure to. With this crappy Corsair Value RAM I can't go that low on RAM timings or my mobo won't POST. Edit: The last 1T means command rate, correct? I'm not sure I can even adjust that on the P4P800 Deluxe.

 

[xtknight]

Originally posted by: Shimmishim
nice thread but if i were you, i'd run a bunch more benchmarks just like zebo did...

Yeah I wish I would have, soon I'll update when I get my new system with the 2-2-2-5 memory. I should be able to test a lot more different configurations with it. Anyone know the best PRACTICAL benchmark (not Sandra!) for testing memory? I want to know real-world performance but still avoid the theoretical MB/sec benchmarks if possible. His Athlon 64 benchmarks report slightly larger performance deltas when using different timings. Maybe I'm doing something wrong, or maybe it's my P4.

 

[Zebo]

P4 likes bandwidth you'll find a little more than A64. I found back when I had a 2.6C anything above 260Mhz (3-4-4) owned 2-2-2 up to about 220 IIRC..