FSB & Ram timings benches @ 3.6ghz

doz · Jan 13, 2009

So people were wondering about FSB and whether it helps performance and whether timings do as well so I ran 3 tests. All this was done on the hardware listed in my SIG. Tests run were Everest memory/cache bench and 3dmark. Not alot, but its something to go off of. Is there "real world" results like someone was asking? Not sure, but im sure it will differ program to program. Do I think higher FSB = better? Im old school, YES!

CPU - 400x9 - 3.6ghz
RAM - 800mhz @ 4-4-4-12

CPU - 400x9 - 3.6ghz
RAM - 1066mhz @ 5-5-5-15

CPU - 514x7 - 3.6ghz
RAM - 1066 @ 5-5-5-15

custom90gt · Jan 17, 2009

nice, how about some super pi times.

jason4207 · Jan 17, 2009

What is your tRD (Performace Level) at each of those settings? I think you can get your latency down at the high FSB setting if you can reduce tRD some.

IMO the only way to properly compare different FSB/RAM ratios is to tweak all the timings at each setting. I still get the best bandwidth results w/ max FSB and max RAM speed w/ the tightest timings I can run at that speed.

Here's my best w/ what I have in my sig:

tRD=6

MadMan007 · Jan 17, 2009

This is pretty nice, just shows that in more real-world programs Intel has really tuned things to hide bandwidth and latency. SPi and Everest are nice to see how changes affect low-level numbers but don't necessarily correspond to real programs. It's sort of like the way that low-level hard drive measurements don't always play out with more real program tests.

What else is there besides 3DM that reasonably simulates real-worl use? I may try something like this with built-in game timedemos but I need to do more tuning first.

Priv · Jan 17, 2009

jason4207 said:
What is your tRD (Performace Level) at each of those settings? I think you can get your latency down at the high FSB setting if you can reduce tRD some.

IMO the only way to properly compare different FSB/RAM ratios is to tweak all the timings at each setting. I still get the best bandwidth results w/ max FSB and max RAM speed w/ the tightest timings I can run at that speed.

Here's my best w/ what I have in my sig:

tRD=6

What NB voltage are you using for that? That is just insane! My board wont post with PL 6 above 400 fsb, lowest I can get then is PL7.

jason4207 · Jan 17, 2009

Try the 5:6 mem ratio to get access to PL6. If you don't have D9's, though, you may not be able to get it.

I can't remember the NB voltage, but I know it was either the lowest setting or 1 above that. I was only running 2 sticks during that run. The key to what you see there is 2.3v on the RAM...that run was the only time I ever pushed them that hard. They usually sit at 2.1-2.15v...rated for 2.2v. Not all RAM responds well to voltage, but the D9's do pretty good.

MadMan007 · Jan 17, 2009

Priv said:
What NB voltage are you using for that? That is just insane! My board wont post with PL 6 above 400 fsb, lowest I can get then is PL7.

He kind of said it but not outright: he has 1GB sticks of D9GMH RAM which can overclock much better than 2GB sticks.

KTE · Jan 17, 2009

FSB, RAM and PL significantly affect Intel C2 CPU performance, yup. Many times much more significant than CPU MHz in cases. In gaming, FSB very quickly will become a bottleneck so higher FSB and lower FSB runs are never directly comparable across reviews. Thats the whole reason 400FSB CPUs formed the highest class offerings by Intel.

The problem when OC'ing RAM/FSB is, some boards are quicker at an FSB/PL than others are. Asus boards are usually left very loose on MCH timings to allow high FSBs, while many will be much tighter. Commando was a classic example of the former. They take a hit on absolute perf. though unless you tweak the MCH registers. Nv 680i, DFI P35 and Biostar P45 are usually very tight on RAM/FSB performance per MHz.

Also Asus P35 overvolted very badly on vDIMM, on at least 5-6 of their boards the trend was well known. Its the renown D9 killer. People kept blaming D9 RAM for death but MFGs warned the board was overvolting highly and the board was the culprit. That was one reason why initially some guys reported much higher RAM clocks on Asus P35 than others. Until kiwi on XS measured a high overvolt on vDIMM and others started corroborating it at +0.1-0.5v. My board was one of the samples that killed Crucial D9 due to insane overvoltage

Daily runner. Sucker was showing me 1200 4-4-4-10 at 2.28v, until Cellshock rep asked me to measure it.. 2.65v

Priv · Jan 17, 2009

I do have a set of D9GMH, they do 675 MHz @ 2.3v. Could try doing a benching run actually, see if I can hit 10k read speed, hopefully 11k

. So 5:6 ratio and PL6 above 400 fsb should be possible?

jason4207 · Jan 17, 2009

KTE said:
FSB, RAM and PL significantly affect Intel C2 CPU performance, yup. Many times much more significant than CPU MHz in cases. In gaming, FSB very quickly will become a bottleneck so higher FSB and lower FSB runs are never directly comparable across reviews. Thats the whole reason 400FSB CPUs formed the highest class offerings by Intel.

The problem when OC'ing RAM/FSB is, some boards are quicker at an FSB/PL than others are. Asus boards are usually left very loose on MCH timings to allow high FSBs, while many will be much tighter. Commando was a classic example of the former. They take a hit on absolute perf. though unless you tweak the MCH registers. Nv 680i, DFI P35 and Biostar P45 are usually very tight on RAM/FSB performance per MHz.

Also Asus P35 overvolted very badly on vDIMM, on at least 5-6 of their boards the trend was well known. Its the renown D9 killer. People kept blaming D9 RAM for death but MFGs warned the board was overvolting highly and the board was the culprit. That was one reason why initially some guys reported much higher RAM clocks on Asus P35 than others. Until kiwi on XS measured a high overvolt on vDIMM and others started corroborating it at +0.1-0.5v. My board was one of the samples that killed Crucial D9 due to insane overvoltage

Daily runner. Sucker was showing me 1200 4-4-4-10 at 2.28v, until Cellshock rep asked me to measure it.. 2.65v

Well, I decided to try and measure my vdimm. I can't find it, though. I found this vdimm point on the P5K-Deluxe...I would assume the Premium would be the same, but that point is showing 1.94v for me w/ a BIOS setting of 2.15v.

Here is the thread that shows the vdimm measure point I tried:

http://www.xtremesystems.org/forums/showthread.php?p=2389473

I also tried some other MOSFETS in that area, but nothing close to 2.15v.

In that same thread they said the vdimm measurement they did was spot on w/ the BIOS setting on the Deluxe. I've been running these Ballistix for over 1-1/2 yrs at 2.1-2.25v BIOS on a daily basis. If this board was horribly overvolting the RAM I'd assume they'd be dead by now. This board is an RMA, and the first board was the same except that I couldn't get my WD HDDs to run in RAID on the 1st board.

Priv said:
I do have a set of D9GMH, they do 675 MHz @ 2.3v. Could try doing a benching run actually, see if I can hit 10k read speed, hopefully 11k . So 5:6 ratio and PL6 above 400 fsb should be possible?

It was easy as pie for me. :beer:

Priv · Jan 18, 2009

I had 10.8k read speed as my highest result and I learned that my fsb wall of my Q6600 is around 520. So I could run 515x7 if I want

. I lost my proof though ffs, the overclock caused a corruption in my raid controller and my raid 0 setup went to smithereens. So I had to reinstall windows and lost ALL my data that I did not back up...

doz · Jan 18, 2009

Well, I changed the PL from 13 to 10 (wow, it was high!). I havent ever tweaked the ram before (tried basics but failed). I havent changed any other settings as of yet. What a difference that made in the memory department! Here is a before and after.

jason4207 · Jan 18, 2009

Priv said:
I had 10.8k read speed as my highest result and I learned that my fsb wall of my Q6600 is around 520. So I could run 515x7 if I want . I lost my proof though ffs, the overclock caused a corruption in my raid controller and my raid 0 setup went to smithereens. So I had to reinstall windows and lost ALL my data that I did not back up...

I learned my lesson about fiddling w/ RAM timings on RAID0 a couple years back. Sorry to hear you just learned a lesson yourself.

Until recently I used a spare HDD to play on and disconnected the RAID0 while I was testing. Now I just use a single Velociraptor, and back up images to my file server. 5-10 min restore and I'm back in business.

jason4207 · Jan 18, 2009

doz said:
Well, I changed the PL from 13 to 10 (wow, it was high!). I havent ever tweaked the ram before (tried basics but failed). I havent changed any other settings as of yet. What a difference that made in the memory department! Here is a before and after.

Heck yeah! Even the L2 cache saw a nice improvement.

doz · Jan 18, 2009

Cannot get to tRD of 9 though

Running my ram currently at 2.2v and my NB is at 1.28. Maybe ill give my NB a bump?

Also lowered my tRFC to 50 (instead of 54), and im running 5-5-4-15 now (tried 5-4-4-15 and 5-5-4-12, neither would go).

Tried tRD 9 again, and couldnt get it. Running in memset:

5-5-5-4-15, 5-45-10-11-11-3-7-5-5-4171-2t

FSB & Ram timings benches @ 3.6ghz

doz

Member

custom90gt

Member

jason4207

Senior Member

MadMan007

Magical Leopluridon Senior

Priv

Member

jason4207

Senior Member

MadMan007

Magical Leopluridon Senior

KTE

Member

Priv

Member

jason4207

Senior Member

Priv

Member

doz

Member

jason4207

Senior Member

jason4207

Senior Member

doz

Member

Similar threads