E6750 & Gigabyte P35C-DS3R Overclocking Thread

[mrfatboy]

ok, new results from the new driver with my current sig settings.

3DMark Score 12304 3DMarks

SM 2.0 Score 5522 Marks

SM 3.0 Score 5315 Marks

CPU Score 3238 Marks

Not bad. 200 extra marks just on a driver change.

I beat Gaucho's score but my vid card is way more OC'd than his. That XFX 8800gts that he has must be faster than the standard Nvidia I guess. I would like to hear somebody's opinion on this.


EDIT
I just ran 3dmark 06 with (707/1093) and got these scores. sort of bizarre.

3DMark Score 12308 3DMarks

SM 2.0 Score 5521 Marks

SM 3.0 Score 5318 Marks

CPU Score 3241 Marks

 

[Conjugal Visit]

Originally posted by: mrfatboy
@Gaucho01
How did you get a 3dmark06 score higher than me? All my componets are clocked higher than yours. Were they clocked higher when you did the benchmark? What were the video care & ram settings then? Maybe it's time for me to move to the f5 bios If you needed to, I would not hesitate to put your vcore & (g)mch up 1 notch each to get 450 or higher.

@mrfatboy - not questioning you, just continuing to learn here...do you really believe that a bios change to F5 is going to impact your O/C, either positive or negative? I've swapped out all the Bios revisions from F2 to F5, and I don't see any difference at all in stability, etc. Perhaps this has more to do with using just a simple O/C scheme of 1:1 ratio, adjusting the DD2 up to 440 (or 450) and "watching" this take the FSB with it - 3.52 and 3.6. And I'll say it again - the F2 bios WILL allow you to fully implement EIST and C1E, unless F2 is fooling CPU-Z. Meaning, F2 (regardless of what the Gigabyte CPU compatibility charts say) recognizes my E6750's cpuid at post, in bios, and in CPU-Z and CrystalCPUID, AND allows C1E and EIST to simultaneously reduce the multiplier to x6 AND undervolt the CPU to 0.963 at idle. If I move to any of the higher bios revisions, multiplier reduction to x6 works, but the voltage does not drop from Overclocked values. I may be wrong, but that seems to defeat the purpose of Speedstep? Anyway, I haven't found anything compelling in the higher bios revisions - I see three changes, basically. One, there is an expansion of options under "System Memory Multiplier" that includes stepping info, but I'm 1:1 (2:2) and I'm pretty sure this doesn't apply to me, and Two, there are some changes in the SMART Fan section (and they don't make a difference to me) and Three, DRAM voltage on the PC Health page (something I already know to be 2.14v when +0.2v is applied to stock voltages (something you've explored extensively.) Personally, I await a bios revision that restores C1E/EIST's ability to undervolt the CPU at idle. More recent revisions don't give me that, and as stated, these recent revisions don't give me anything else either, so why switch up just because it's newer? The keyboard compatibility thing? You been having problems with your keyboard? Me neither.

@Gaucho - my direct experience and reading seems to indicate that pushing the E6750 much past a 3.6 overclock gets pretty tough without a lot of voltage being pushed to the CPU, which generates a LOT of heat. I haven't found a review/comparison that makes a case for going higher anyway - 2-4% gains in apps...3-4 fps in games that are CPU-bound? Not worth to me, factoring in heat, stress, noise, etc. You wanna bench..?...go for it. But if you wanna game, rip music, etc, what's the point?

So my two cents? Try my setup earlier in this thread - it will yield a 3.53GHz and temps are pretty darn good. Your ram is 1066 - mine is 800. You experiment with my setup, ratios at 1:1, and you'll be effectively underclocking your DDR2 to 880, BUT I'm pretty positive that you'll be able to reduce latencies and increase throughput by applying 4-4-4-12 timings with only a +0.2v bump at your DRAM voltage. Your rig will run cooler and be plenty fast.

The only thing I wouldn't copy from my setup is the vcore value. For whatever reason - dumb luck maybe - my chip registers Normal Temp value in bios (any bios revision by the way) at 1.2875. Normal for you, I'm going to guess is, is 1.3500. mrfatboy has called it a super-chip. Maybe, maybe not. But I'm awfully pleased with the E6750, and with this board. So pleased, I bought a second E6750 for my "family" rig, a GA-P35-DS3R. And yeah, Normal Temp shows 1.3500 on that chip. Point is, I'm running 3.52GHz on my rig at a vcore of 1.36250v, and I'm pretty sure you'll need to go higher than that and be stable. But I don't think you'll need to go into the 1.45v range to do 3.52GHz...like I said, my two cents...

Abraços

 

[mrfatboy]

@conjugal
I was joking when I said the F5 bios would increase speed. I should have put a :roll: instead of a , lol. Just doing a comparison of Guacho's rig and mine I was surpised at the difference in 3dmark scores. Especially, when he was beating me with a lower clocked card and a 320meg at that. I guess the XFX model is just better clock for clock.

I never experimented with OC'ing between 3.2 & 3.6. I ran 3.2 at all stock and then immediately tried for 3.6. it took my sig settings to get there. I wonder what my voltages would be if I dropped it down to 3.52?

3.6 is just a nice number :laugh:


@Gaucho,
If conjugal is right and the average system can run 3.52 without the big voltage jump that 3.6 needs you should be fine espically in your heat. You won't notice the difference in games. If your vid card is as good as I think it is that is where you are going to get all of your gaming speed. Have you tried to overclock further?


@sniperdaws
and no... I have not added the ram yet :laugh:

 

[Gaucho01]

As I said, I was with a 444 x8 on E6750, 1,45V 0,3+, and 1050 MHZ and 650 MHZ on my 8800 gts 320 GB. with 163.71 drivers.
I set on high performance on the Nvidia config.
I closed my anti virus, firewall, e explorer.
This was my best score.
I usually get around 12k.
Heheheh.
Im glad I beat somebody.

And Yes I could increase my 8800 to 1100 and 700 and try again.
I´ll do it tonight.

 

[mrfatboy]

Originally posted by: Gaucho01
As I said, I was with a 444 x8 on E6750, 1,45V 0,3+, and 1050 MHZ and 650 MHZ on my 8800 gts 320 GB. with 163.71 drivers.
I set on high performance on the Nvidia config.
I closed my anti virus, firewall, e explorer.
This was my best score.
I usually get around 12k.
Heheheh.
Im glad I beat somebody.

And Yes I could increase my 8800 to 1100 and 700 and try again.
I´ll do it tonight.

well, 12308 is the score to beat now :laugh: You should be the winner though after you make your adjustments

 

[Gaucho01]

Not for tonigh.
I could not repeat my 12269 mark, It stoped at 12050.
Did not pass over 660x1080.
Dont know what happened.
Cant remember what I really did when I got 12269.
I guess is the ambient temperature.
Tonight is preaty hot.(32C)
So for now you are the Winner.

 

[mrfatboy]

:laugh: I am (for now) Your original score is still weird to me. I wonder what you did to get it. I know you can gain several point by eliminating as many background processes as you can but your score was abnormally high. When you get a chance try it again.

thanks for waking me up though. I hadn't updated my drivers in awhile so 200 marks extra was a nice surprice.

 

[honolululu]

Can I play too? 8)
3DMark Score 13026 3DMarks

SM 2.0 Score 5700 Marks

SM 3.0 Score 5916 Marks

CPU Score 3291 Marks

That's at 3.8 GHz and 610/1000. That's was awhile ago. Now I've gone mister conservative. Running at stock with manual but normal voltages.

ConjugalVisit I rolled back to F2 to see if I could get this overclocked EIST action to drop down below 1V on 6X multi. It seems you do have a magic CPU, because it never dropped below ~1.12V. But now that I'm back to F5 I'm still getting the same voltage drops of ~0.2V with voltage at manual, but normal, EIST and CIE enabled also in Windows like you said. And this also works at 3.2 GHz, but it only drops to about 1.18V

A cool thing I noticed is with CPU at stock and EIST and CIE enabled; when the multi drops it seems CoreTemp will report a new lower VID, which accounting for Vdrop was right where my voltage was at in the reduced power state.

 

[mrfatboy]

nice score That GTX spanks our GTS's


@conjugal

Did you ever find out what your max OC was with your super CPU?

 

[Gaucho01]

Im downloading the last Drivers 169.04, still beta.
I'll try again and see what happens.
But our score was pretty close to 8800 GTX.

 

[mrfatboy]

I just downloaded 169.04 beta. I didn't know it was there

Even Faster. I dig it


3DMark Score 12760 3DMarks

SM 2.0 Score 5673 Marks

SM 3.0 Score 5696 Marks

CPU Score 3233 Marks

 

[biltong]

Well, I eventually got stable at 3.552GHz (FSB=444, 2.4 Mem Mult=1066MHz), but I need a Vcore of 1.48125 which is a bit higher than everyone else seems to need for 3.6GHz

I tried upping the G(MCH) at lower Vcores, but that didn't make any difference for me. Then I put the G(MCH) back to +0.0 and instead upped the FSB voltage, but again that didn't make any difference. So I've ended up with the VCore at 1.48125, G(MCH) +0.0, FSB +0.0 and RAM +0.3 (thought I might get away with the RAM set to +0.2, but the RAM isn't stable).

Seems to be a big Vdroop under load though......With the Vcore set to 1.48125, cpu-z reports 1.424V (under load, running Orthos blend). Is that normal ?
EDIT: From some of the other posts I've just read, seems it is...

 

[Conjugal Visit]

@biltong - have you thought about running your 1066 ram as follows:

Mem at 1:1 ratio

VDimm at +0.2v or +0.3v - whatever works for you

Timings @ 4-4-4-12

With that CPU's multiplier at x8 and that speed of DDR2, you are effectively running it underclocked at 444 (444x2=888) which I would think would give you the opportunity to tighten up your timings and run 1:1. I think you'd have a pretty zippy rig using that config.

 

[biltong]

@Conjugal Visit - I was originally running with the timings set to 4-4-4-12 and the mem ratio at 1:1 exactly as you suggest.

Then I figured if I set the FSB to 444 with a mem multiplier of 1:1.4 (x2.4) I could run the RAM at its rated 1066MHz with 5-5-5-15 timings

To be honest, I wasn't really sure what would give the best performance, but after doing a few crude tests, it seemed to give better performance at the higher clock rate with looser timings rather than the lower clock rate (888MHz) with the tighter timings.

Maybe I should do some more tests.....Whats a good tool to test the memory performance ? I was thinking of just doing a few loops of memtest86+ and seeing how long it takes with each setting ? Or is there a better way of testing the memory performance ?

 

[mrfatboy]

I always wanted to know this question as well. Search here at Anandtech or google for "Sandra memory test". I think it will do the job.

Let us know the results

 

[honolululu]

Originally posted by: biltong
Whats a good tool to test the memory performance ?

I'd recommend either Everest Home 2.20.405 which does Read, Write, and Latency
or
the Lite version of SANDRA which only does Read (one or the other I believe), and Latency in the free version.

They're both great programs. The SANDRA one is very handy for saving the results of past settings and you can compare a bunch at once.

Here's a sample pic with SANDRA
Screenshot

EDIT: sorry typo

 

[Gaucho01]

We have a new BIOS, F6. Im just downloading.

 

[mrfatboy]

What is the link? It's not on the U.S. website. What does it say about the fixes? PS2 combatibility? :laugh:


Actually, all joking aside, I was wondering yesterday if the PS2 combatibility problem was related to the "cycle reboot" problem. Maybe the bios' were having problems recognizing the keyboard and not POSTing and rebooting itself. Just a guess.

My motherboard still doesn't like to POST somethings after a reboot. I have to wait five minutes and try again. When it does POST again i'm back down to stock FSB. Maybe the gigabyte failsafe bios system is borking.

 

[Gaucho01]

http://www.gigabyte.com.tw/Sup...oductName=GA-P35C-DS3R

 

[mrfatboy]

hmmmm. strange. They don't have the F5 bios on the Taiwanese website like the U.S. webstite. They go straight to F6 from F4. I wonder if it's a typo?


Gigabyte at it's finest



@gaucho, anything different? Any new bios options?

 

[mrfatboy]

FYI, I just download both F5 and F6. I ran microsofts file compare program on them

If i did everything correctly, they are basically the same file but with 10 bytes different.

I looked up the the hex values in an ASCII chart. lol, oldschool style

It looks just like a header name to me. It spells P35Cds3r compared to the something similiar but incorrect. Somebody probably typo'd and they fixed it.

I believe f5 & f6 are basically the same Anybody else care to do some super sleuthing

 

[Conjugal Visit]

The plot thickens...if you open @bios, and instruct it to download the latest bios from the Taiwan server, it comes back with a version labeled F5. I just did this. I'm thinking this F6 is the F5. I 'll load it up and see if there are any new options that go beyond F5. Then I'll prolly unload it and go back to good ol' F2 for the undervolt at idle. I'll post here.

Regards,

 

[Conjugal Visit]

I don't see anything new from the F5 version I'd loaded in the past, as far as options go. But I no longer have the F5 to load up. Interesting thing, though, and maybe I didn't notice before, but with this F6, when I went to the PC Health page and looked at voltages, VDIMM was showing 2.12 - I think it used to show 2.14 (I apply +0.2v to DDR2.) VCore shown was not the voltage I set in MIT page - this was 1.33 and I set it at 1.3625....maybe I just never noticed these things before.

 

[mrfatboy]

Not only does this board overvolt the vdimm it does flucutate a bit. I can't comment your Vcore voltage observations.

 

[biltong]

For some light relief, I finally got round to running a few memory benchmarks to see what impact memory speed and memory timings had on performance. I ran these benchmarks with the following memory settings:

1. 888MHz, 5-5-5-15 (FSB444 x2.0)
2. 888MHz, 4-4-4-12 (FSB444 x2.0)
3. 1066MHz 5-5-5-15 (FSB444 x2.4)

Apart from the memory multiplier and memory timings, I left all other settings the same. Same voltage settings, same FSB etc.

The first test I did was a single timed loop of memtest86+

1. 888MHz, 5-5-5-15 (FSB444 x2.0) --- 20 min 33 sec
2. 888MHz, 4-4-4-12 (FSB444 x2.0) --- 19 min 07 sec
3. 1066MHz 5-5-5-15 (FSB444 x2.4) -- 18 min 45 sec

Then I used SandraLite and ran the Memory bandwidth and Memory Latency benchmarks:

SandraLite Memory bandwidth benchmark
SandraLite Latency benchmark

For the latency test, the actual benchmark figures were:

1. 888MHz, 5-5-5-15 (FSB444 x2.0) - 80nS, 94.3 (speed factor)
2. 888MHz, 4-4-4-12 (FSB444 x2.0) - 75nS, 89.0 (speed factor)
3. 1066MHz 5-5-5-15 (FSB444 x2.4) - 72nS, 85.40 (speed factor)

So seems running at 1066MHz with looser timings is better than running at 888MHz with tighter timings.......