looser timings, better performance?

I was running 3dmark05 at 2-2-2-5 timings stable, then after looking around on different forums I though I might try 2-2-2-6 and I got a higher score. Obviously not by much (like 20-30 points). I'm currently running the RAM at 450mhz (HTT @ 225) @ 2.96v. My 3dmark05 score is around 7030. Any thoughts on this? both timings were prime stable.
I'm still a noob at this so any feedback would be good.

DuckOfDeath said:

I was running 3dmark05 at 2-2-2-5 timings stable, then after looking around on different forums I though I might try 2-2-2-6 and I got a higher score. Obviously not by much (like 20-30 points). I'm currently running the RAM at 450mhz (HTT @ 225) @ 2.96v. My 3dmark05 score is around 7030. Any thoughts on this? both timings were prime stable.
I'm still a noob at this so any feedback would be good.

Click to expand...

Lowering tras does not give better performance, its similar to tref, most settings will work and a few give slightly better stability/bandwidth. Tweeking and testing is th ekey. The change may have affected bandwidth slightly or the change may be within margin or error. All settings are not as simple as lower = better.

Lowering tRAS actually can improve performance, but there's a threshold. If it's too low (5 at 2-2-2 is too low) then the memory will try to perform a new read before the last read is finished. As a result, you end up with read errors and the reads need to be performed again. That's what causes the performance hit in 3DMark - your RAM isn't entirely stable, and the system is having to work harder to correct errors. If you go up to 2-2-2-7, you'll probably see worse scores again because the RAM is waiting longer than it needs to.

Thats right. I don't understand why people insist on running 2-2-2-5 instead of 2-2-2-6.
Maybe it is because they buy into lower must be better, since the marketing surrounding the product they are sold is rated at 2-2-2-5.

Any tRAS setting lower tRCD + CAS + 2 cycles will allow the memory controller to close the page “in your face!” over and again and that will cause a performance hit because of a truncated transfer that needs to be repeated. Along with those hassles comes the self-explanatory risk for data corruption. That one is not a real problem as long as the system is kept running but in case it is shut down and the memory content is written back to the hard disk drive, the consequences can be catastrophic. For the drive, that is.

The above taken from Mushkin's offical website!

In theroy tRCD + CAS + 2 cycles, if you were to run 2-3-2-6, then the best setting would actually be 2-3-2-7

Lol. My ballistix is default rated at 2-2-2-8? Your opinions on that?

azu said:

Lol. My ballistix is default rated at 2-2-2-8? Your opinions on that?

Click to expand...

Easy run 2-2-2-6 if you can, and I see you are in your sig.

I've done some pretty extensive testing (see guide in my sig) and as long as you don't set it too short (as microfire points out) then it doesn't make much difference.

Several years ago when we were using the XP and Abit's NF7 a tRAS of 11 was the rage. Turned out if you did enough tests (repeat the test) and did statistics on the results, any "improvement" was less than the variation you would get from test to test.

I've not tested again in the last year but would be surprised if things have changed. As always, the bottom line is to test your equipment to see how your own hardware works.

I see you have a DFI nF4 motherboard.

Play with your tRAS a bit, run a standardized bench across a few settings to find your hotspot - I think you will find that 3, 6, or 7/8 is ideal.

These DFI boards have some weird voodoo going on with some of the memory timings; values are available that shouldn't exist, and some of them behave in a peculiar manner .

I run a tRAS of 3, works for me .


EDIT: tRAS not tREF, my bad!

felinusz said:

I see you have a DFI nF4 motherboard.

Play with your tRef a bit, run a standardized bench across a few settings to find your hotspot - I think you will find that 3, 6, or 7/8 is ideal.

These DFI boards have some weird voodoo going on with some of the memory timings; values are available that shouldn't exist, and some of them behave in a peculiar manner .

I run a tRef of 3, works for me .

Click to expand...

tREF? It usually has possible settings in the thousands. You must have one of the other settings in mind.

Since there seem to be several million timing choices available these days it isn't hard to get them mixed up --- or do I have them mixed up?

Whoops!~

I meant to say tRAS, not tREF! tRAS values like 00, 01, 02, do something "special" with the DFI boards - Oskar Wu apparantly coded these values in, as "tweaked" tRAS values, much like CAS 1.5, which is a "tweaked" CAS 2 timing. TCCD is said to love a tRAS of 00, for example.

Since there seem to be several million timing choices available these days it isn't hard to get them mixed up --- or do I have them mixed up?

Click to expand...

I **know** that I'm mixed up.

tras of 3?? wow, that's a new one. Maybe I'll give that a shot. Would car chase and lobby in 3dmark2k1 be good tests?

The add up your timings and that should be your best tras setting theory is just not true. There is no formula for timings. As Eldonko pointed out, you need to tst it to see what's best. Shortcuts generally lead to bad results.