Wanna test a quick Memory read/write boost, if your Board supports it?

Lupiron

Distinguished
Feb 9, 2008
1,711
0
19,780
If anyone is interested, believe it or not, this could boost your reads and writes by a small amount, couple of hundred.

If your board supports manual strap to north bridge adjustments, and you have it set at 400 Mhz because your OC is good, try simply setting it to the 333 strap and then selecting the memory speed that it was nearest to. (It wont allow for a similar selection if your memory is already high, or you are running a berserk divider, which is useless anyways.)

Technically, your MoBo May be on Auto, and using this strap, however, most of us are in sync at 400 FSB anyways, so would use the 400 Mhz strap, and so would Auto, just change it to 333, and select 801 Mhz, or its equivalent.

Make sure you Read/Write/Copy/Latency test your memory first! Then go into the bios and try it.

If its not simply using the strap to force a Trd change, then I did something else...

::Scratches head.::

And who knows what that was...

Mention if you get the speed boost...

--Lupi
 

Lupiron

Distinguished
Feb 9, 2008
1,711
0
19,780
Oh, it could cause a post failure if your Memory is running at the edge of stability!! Be warned, And test your Memory with a decent tool if you do get the speed boost. :) Wouldn't want any unstable systems out there!

--Lupi
 

Lupiron

Distinguished
Feb 9, 2008
1,711
0
19,780
Sure, most of the newer boards will have its support.

I know my Formula does, My P5K-e wifi does, the 750i, which is a 650i with new lane controller does not have it! (because it's really an old board.)

--Lupi
 

Lupiron

Distinguished
Feb 9, 2008
1,711
0
19,780
Wow, yeah, who needs a few hundred Megs read, write and copy for 30 seconds of hitting enter and the arrow key to select a new FSB strap, then saving it.

Not me! I'd rather be slower, than get free performance!

No one would ever wish to maximize their system performance!

--Lupi
 

Zorg

Splendid
May 31, 2004
6,732
0
25,790
I saw that. It appears that you can just change the tRD directly on the newer chipsets. It can be done on the P35-DQ6. I will do some benches and post them. I was only able to go from 6 to 5 clocks with 1333FSB DDR2 800 clocked to 667 (1:1) 3-3-3-10. It appears that the auto tRD in the DQ6 is pretty close to the minimum, at least with the above configuration. I'll check the higher speeds.

Like I said earlier I don't think it will give much in real world performance with the Core2 duo/quad, because they aren't affected that much by memory bandwidth/latency.

I'll bench it and give you some Sandra numbers anyway, as soon as I get some free time.
 

Zorg

Splendid
May 31, 2004
6,732
0
25,790
Here you go.

SiSoftware Sandra Benchmarks

1333/667 3-3-3-10 tRD5
Int Buff'd iSSE2 Memory Bandwidth : 6.03GB/s
Float Buff'd iSSE2 Memory Bandwidth : 6.03GB/s

1333/667 3-3-3-10 tRD(Auto)
Int Buff'd iSSE2 Memory Bandwidth : 5.98GB/s
Float Buff'd iSSE2 Memory Bandwidth : 5.98GB/s

1333/667 3-3-3-10 tRD5
Memory (Random Access) Latency : 83ns

1333/667 3-3-3-10 tRD(Auto)
Memory (Random Access) Latency : 84ns


1333/800 4-4-4-12 tRD5
Int Buff'd iSSE2 Memory Bandwidth : 6.27GB/s
Float Buff'd iSSE2 Memory Bandwidth : 6.29GB/s

1333/800 4-4-4-12 tRD(Auto)
Int Buff'd iSSE2 Memory Bandwidth : 5.90GB/s
Float Buff'd iSSE2 Memory Bandwidth : 5.91GB/s

1333/800 4-4-4-12 tRD5
Memory (Random Access) Latency : 80ns

1333/800 4-4-4-12 tRD(Auto)
Memory (Random Access) Latency : 89ns


1333/1000 5-5-5-15 tRD5
Int Buff'd iSSE2 Memory Bandwidth : 6.52GB/s
Float Buff'd iSSE2 Memory Bandwidth : 6.54GB/s

1333/1000 5-5-5-15 tRD(Auto)

Int Buff'd iSSE2 Memory Bandwidth : 6.54GB/s
Float Buff'd iSSE2 Memory Bandwidth : 6.56GB/s

1333/1000 5-5-5-15 tRD5
Memory (Random Access) Latency : 73ns

1333/1000 5-5-5-15 tRD(Auto)
Memory (Random Access) Latency : 73ns
 

Slobogob

Distinguished
Aug 10, 2006
1,431
0
19,280

This is an enthusiast forum after all...
 

Zorg

Splendid
May 31, 2004
6,732
0
25,790
As far as I can see, the differences are negligible, at 1333 FSB anyway. At 1333/1000 setting it actually lost some ground. It's possible that the GA-P35-DQ6 did a good job of setting the tRD to begin with, or the 1333 isn't high enough to realize the gains, or the SPD timings on my RAM aren't tight enough, I don't know. I'm sure you are aware that The core2 is not very sensitive to memory tweaking, so the real world differences will probably be not noticeable anyway, unless your running an Allendale. The only reason I ran the benches in the first place is because I read the same Anandtech article that lupiron posted, and of course, I was bored senseless. I lost cable due to a nasty storm last night. :cry:

Unless I got something wrong somehow, I still say it's not worth the hassle to run the calculations. Run some benches and let's see what you get. Your an enthusiast, right?