X38 DQ6 - High DDR2 OC WILL ROB Bandwidth @ 1200Mhz+

OCChronic

Member
May 7, 2008
83
0
0
Can anyone elaborate on what is happening or at least post similiar experiences with the same motherboard,

GA-X38-DQ6(DDR2 1066 version)

At 1200 Mhz and up, the X38 MCH loosens the DDR2 strap considerably. With all other settings being identical, when bumping up from 1197 to 1200 Mhz, I lose 1065MB/s(18.4%) memory bandwidth(6210 MB/s vs 7350 MB/s). Also, memory latency is increased by 10ns(68-78)

My theory is that the strap is going into DDR3 mode even on boards with DDR2 support, once you hit the 1200Mhz DDR mark and beyond.
 

Sylvanas

Diamond Member
Jan 20, 2004
3,752
0
0
You are probably right in your assumption and unfortunately there is nothing really you can do unless Giga release a BIOS that allows TrD adjustment or if Performance level is adjustable through Memset. Either that or just stick to 1197mhz :p.
 

VirtualLarry

No Lifer
Aug 25, 2001
56,571
10,207
126
You mean to tell me that you're running at 600Mhz FSB on your CPU? Which CPU might that be??? You do know that running memory faster than 1:1 provides no real benefit, right?
 

Sylvanas

Diamond Member
Jan 20, 2004
3,752
0
0
Originally posted by: VirtualLarry
You mean to tell me that you're running at 600Mhz FSB on your CPU? Which CPU might that be??? You do know that running memory faster than 1:1 provides no real benefit, right?

You do not have to run 1:1 in order to get 1200mhz, thats what divers are for. There are benefits but you wouldn't really notice them in the real world, high frequency DDR2 is only really useful for benching SuperPI etc- and even then, you can get similar results with DDR2 800 Cas 3 1T.
 

OCChronic

Member
May 7, 2008
83
0
0
My e4500 has A FSB wall of 359 MHZ and cannot get DDR2 800 @ 1:1 ratio so it isn't in fact pointless to run the 3.33C divider. That's the best possible performance setting for my CPU/CHIPSET/MEMORY combination. Also, my Corsair DDR2 1066 Dominators won't run with a 1T command rate or CAS3 at DDR2 800 on INTEL chipsets. That is precisely why different dividers and straps are available to accomodate many different hardware confogurations. It's more confusing and has more variables than the AMD systems out there with an on-die memory controller.

 

cactusdog

Member
Apr 28, 2008
52
0
0
This happens on any board. As the frequency goes up the timings needs to be looser (usually). Its not just the DQ6. If you look at timings on standard memory sticks it like 800Mhz 4-4-4-12, 1066 Mhz 5-5-5-15 1333Mhz 7-7-7-21 1800Mhz 9-9-9-25 etc.

Having said that the DQ6 seems to be unable to hold good latencies at high frequency compared to the P5E. You can get tight timings on the DQ6 but the latency seems to go to hell. That was my big problem with the DQ6 and the reason i changed to the P5E.

My memory performance with the P5E is so much better than the DQ6 i had before and thats using the exact same ram.



 

OCChronic

Member
May 7, 2008
83
0
0
Originally posted by: cactusdog
This happens on any board. As the frequency goes up the timings needs to be looser (usually). Its not just the DQ6. If you look at timings on standard memory sticks it like 800Mhz 4-4-4-12, 1066 Mhz 5-5-5-15 1333Mhz 7-7-7-21 1800Mhz 9-9-9-25 etc.

Having said that the DQ6 seems to be unable to hold good latencies at high frequency compared to the P5E. You can get tight timings on the DQ6 but the latency seems to go to hell. That was my big problem with the DQ6 and the reason i changed to the P5E.

My memory performance with the P5E is so much better than the DQ6 i had before and thats using the exact same ram.

It wasn't the meory timings that robbed me of performance at 1200Mhz and above. It was the NorthBridge strap loosening up quite considerably. It's as if the X38 chipset went into a DDR3 timing mode.

From 1197Mhz -> 1200Mhz @ the same exact mem timings(5,5,5,18,2T/perf. enhance:TURBO, etc.), I lost 1000MB buffered mem bandwidth. I was wondering if anyone else could duplicate this phenomenon.

Can incresing the MCH/FSB voltage(s) bring this back into line or is the chipset hardwired to react this way without any way to remedy it?

 

OCChronic

Member
May 7, 2008
83
0
0
*** UPDATE ***

I've changed from an E4500 @ 3150Mhz to an E8400 @ 3880MHz. The X38-DQ6 has so far not dissapointed me. My Sisoftware synthetic memory benches are now in the general neighborhood of 9500 MB/s read/write bandwidth, 61.6 ns latency and rock stable.

Current settings:

+.25v on the FSB voltage
102Mhz PCI-E BUS
INTEL CORE 2 DUO E8400 @
1.3825v(act. 1.360v)/3880MHz/485FSB/
2 x 1 GB Corsair Dominator DDR2 1067/C5 @
2.25v/1164Mhz/CL5,5,5,18,2T, TRD7('turbo')
3 x 500GB RAID '0' array
8800GTX 640MB/112SH overclocked @
600Mhz-GPU/1350Mhz-SHADERS/1800Mhz-DDR3

Everst 4.5 Ultimate Benchmarks:

Memory Read: 10099 MB/s
Memory Write: 10283 MB/s
Memory Copy: 10196 MB/s
memory Latency: 50.7 ns



So far, everything is rock-solid stable and I've given it the Prime95 12 hr. blended test, on both cores, while doing my ebay thang on the net and watching movies.