Q9550 + DDR3 OC Help Needed

[neverwind]

Hello,

I hope there's someone that could help me with my desperate attempts at getting the MAX out of my current system.
I know it's a pretty old one, but still I really love it and I don't need to upgrade to a more modern one at this time.
I guess it's safe to say I'm pretty new to overclocking, but I've been messing with the settings for a while now and the truth is it it turned out to be quite fun

Here's my current setup:
CPU: Intel® Core™2 Quad Processor Q9550 (12M Cache, 2.83 GHz, 1333 MHz FSB)
Motherboard: Gigabyte GA-EP45T-UD3LR (Intel® P45 + ICH10R Chipset)
Memory: Kingston KHX18C9T3K2/8X

I have managed to achieve two semi-stable profiles - one with Memory @ 1800 Mhz, and one with Memory @ 1600 Mhz. Semi-stable means that they seem stable, but I haven't tested any of them long enough to be absolutely sure.

Memory 1800Mhz:
1. CPU: 3825 MHz (8.5 x 450), Core @ 1.38 V (BIOS)
2. FSB: 450, MCH Core @ 1.6 V
3. Memory: Dual Channel 1800Mhz, Timings 9-11-9-27 CR2, Motherboard Performance Level 8 @ 1.696 V

Memory 1600Mhz:
1. CPU: 4088 (8.5 x 481), Core @ 1.38 V (BIOS)
2. FSB: 481, MCH Core @ 1.6 V
3. Memory Dual Channel 1600 Mhz, Timing 9-9-9-27 CR2, Motherboard Performance Level 10 @ 1.696 V

For both profiles, at full load, CPU Temp doesn't exceed 65C (around 63C-64C, actually).
North Bridge temp is around 35C at normal operation and when gaming, goes up to around 45C (This should be because of the fact the NB is responsible for managing the PCIe2 x16 slot I guess)
I don't really know if it matters, but both profiles have the PCIe frequency set at 100Mhz.

So, I'm wondering if I could get more of my current rig?
I've certainly reached my limit, because whenever I touch anything to push it a bit more, it becomes unstable.
As far as my experience with it up until now tells me, the CPU is not an issue, since even at 4Ghz, Stock Voltage, it's stable and the the temp is kept below 65C at full load with Prime95.
I guess it's a FSB limit I'm facing, but is there any way to overcome it. I'd really like to push the memory a bit more with the CPU at 4Ghz

Also, I searched a lot, but I couldn't find a straight answer - what is the maximum considered Safe Voltage for DDR3 memory? I know for newer systems it is about 1.65V because of the memory controller being implemented in the processors, but since this is not my case, can I play a bit more with it?

I'd welcome any advice(s) someone can give me!

P.S.: There's a funny fact. When going at 480 FSB, under full load, my board started making a very warning hissing sound. I guess this is the frequency that resonates exactly with the coils and thus making them whine?
At FSB 481 however, under full load, the sound is barely recognizable so my presumption is that it's OK to leave it like that. Please tell me If I'm wrong

Finally, please excuse me for any mistakes I may have made - I'm not a native English speaker

Thanks,
Bobby

 

[ATMINSIDE]

What cooling do you have?
Sounds like it's time to do some longer Prime95 tests to determine stability

Also, your English is spot-on, very nicely written

 

[neverwind]

I'm running it on air with a Noctua NHU12P SE2

I'd really like to push it a little more and then do some serious testing. Unfortunately, I'm not sure if there's some trick or something else I don't know about or if this is the very limit of the motherboard itself...

About my English - thanks, I'm really trying

 

[ATMINSIDE]

You'll always run in to a limitation of either the motherboard, CPU, or cooling capability of your setup.
In this case it does seem to be the motherboard. 481MHz isn't bad at all for the FSB

As far as safe voltage, I wouldn't go much over 1.7V on air for daily usage. About 1.75V-1.8V for benchmarking.
Have you tried setting the memory multiplier back up to 4X while you're on 481MHz? I wouldn't worry too much about being at 1600 CL9, that's solid for gaming and general usage.

I'd go test that second list of settings on Prime95 for two hours. Note with HWMonitor what all your max temps are and see if it's stable. If so, you should be good to go

 

[neverwind]

Going with Auto DRAM voltage on both profiles, the motherboard sets it at 1.8V. I know this was normal for DDR2, but I guess it's too much for DDR3.

I've tried any possible latch and no matter what, the motherboard doesn't accept anything beyond 1800Mhz. I haven't played with the timings too much, but it doesn't make any sense to loosen them up a lot just for the extra frequency gain. Is this correct?

Also, I've gone through many posts over the internet saying that a safe MCH Core voltage is about 1.35V-1.40V. I've noticed that in my case this is the setting to affect the stability for the entire system the most. Intel's temperature specifications for the P45 Chipset are around 100C. I'm way below that and I was wondering if I could keep pushing it with more voltage? Do you think it's safe/semi-safe to go for more than 1.6V?

The thing is I'm using the system for some (not professional) audio recording / editing and although rarely - gaming as well.
Gaming is fine. I'm having no issues whatsoever with both profiles while playing games, but since audio processing is very memory demanding, I was hoping to find a way to squeeze a bit more from the memory.

 

[ATMINSIDE]

Correct, don't loosen timings a ton to get a slightly higher frequency.
You can see if it will benefit by doing CAS Latency/Frequency X 1000Hz. This will give you a number in nanoseconds. Lower is better.
For example, for 1866 CL9 it would be 9/1866x1000=4.82ns.
You can check if raising/lowering speed/timings would be better or worse this way.

I'm not sure about the MCH voltage, sorry.

Maybe you could run some tests with both settings lists and see which performs better for your audio work?
That'd be the best way for you to determine the better configuration for you

 

[neverwind]

You know, it started as a way to speed things up a bit and now it's more like messing around with it for fun

So, If I understand everything correctly, I may try having a lower frequency setting for the memory and go for lower timings and basically have the same (or even better) performance.
If we are talking about memory performance, it should be the same with both the profiles I have, right? I mean:
1800 on CAS 9 is 5ns
1600 on CAS 8 is 5ns

Please help me out on this one.
Going through the SPD stored profiles of my memory kit and doing some simple math I came to the conclusion I should aim at 1450Mhz on CAS 7 (4.83ns).
I know the difference is very small, but maybe this way I'll have room to push the FSB a little more at no memory performance expense, right? Cause up until now, I was purely aiming at the highest possible memory frequency...

Btw, It's already 4:00 am local time, so I'm signing out for today/tonight.
Thank you very much for the help thus far! I really appreciate it!

P.S.: I couldn't resist and tried it out! And guess what? It worked! I'm currently running the FSB at 482, which was IMPOSSIBLE previously!
Man, I can't wait to go to sleep, wake up and continue!

 

[ATMINSIDE]

It's incredibly fun and addictive, that's why I have a system just for overclocking

You've got almost the same timings listed in your original post, so the RAM would slow down a bit on the second set.
But yes, 1600 CAS 8 is going to run basically the same as 1866 CAS 9. It'll be so close you won't notice any difference.

Give the different settings each a test in your editing program, then you'll know for sure which are better.
The nanoseconds method gets you close, but different programs like speed more and some like lower latency more.

Man, that's a big time difference!!
You're very welcome, glad to help

So, 1450MHz at CAS 7 worked? Did you run any benchmarks to see if it was faster or slower for your usage?

 

[neverwind]

I've done a lot of testing and I've played around with everything I could find and here's the verdict!

1. A bit fooling around:
When using single channel memory setup (still with both DIMMs), I was actually able to achieve the following with NO ISSUES AT ALL:

CPU: 4.250Ghz, Auto Voltage set it at around 1.4V (Vdroop still present)
FSB: 500Mhz, MCH Core 1.6V
Memory: 2000Mhz on 9-11-9-27 CR1

Kinda cool on such an old motherboard, but still memory performance was nowhere near to dual channel setup...


2. The real setup:
I was actually able to get to 500FSB with dual channel setup, with 1640Mhz memory on 8-10-8-24 Timings but it was very unstable with Prime95 and I couldn't find a way to stabilize it...i guess here the CPU Voltage was an issue but I quit after going all the way up to 1.44V (no Vdroop present) with no luck.

So I started going down the FSB slope, and ended up...right where I had started...it looked stable already at 490, but it had a lot of unbearable (to my liking, at least) coil whining and hissing...

CPU: 4.08Ghz, Auto Voltage (1.32-1.36)
FSB: 481, MCH Core 1.6V
Memory: 1443Mhz on 7-9-7-21 CR2

I'll leave it for a couple of days and maybe try again later, since I just feel it can do more than that

 

[ATMINSIDE]

You'll definitely want to stick with dual channel on the RAM, but it could be good to know for benchmarking that you can make 500+ on the FSB
Did you try loosening the timings on the 1640MHz RAM to see if it would stabilize?

 

[neverwind]

I definitely think it isn't the memory this time.
I've just tried loosing the timings up to 10-12-10-30 and I'm still getting a BSOD. My guess is that the CPU is not getting enough juice...
So, what I tried just now is to leave the voltage control to auto but turn on the Line Calibration feature (Vdroop Improvement). Prime runs fine without getting a BSOD.
In BIOS, it says my CPU Core voltage is 1.444V. This is too much for daily usage, isn't it?
Also, turning on the Vdroop improvement makes the hissing sound come back really load when on load...
Any advice?

 

[ATMINSIDE]

Yes, you'll definitely want LLC to be running when pushing big numbers.
As long as you're keeping the core temp under ~65°C then you're fine up to ~1.45V for daily usage.

 

[neverwind]

Thanks...I don't think I can live with the coil whining/hissing...it's HORRIBLE! I'll try something around 500FSB to see if it will go away...
Also, unfortunately, under full load at this voltage with LLC, the temps go beyond 65...more like 67-68...I guess I'll have to start trying to manually adjust the voltage...

Btw, I've passmarked the PC @ 4250Mhz:



- looks kinda nice...4 out of 5 stars, compared to other submitted results, is not that bad for a 6 year old pc

 

[ATMINSIDE]

Yeah, definitely sounds like it's time for you to come off of auto voltage for the CPU. Try ~1.42V and see if it will run 4250MHz there.

 

[neverwind]

I was able to stabilize the CPU @ 4.25Ghz with 1.44V, LLC enabled, but it runs too hot and still my main goal is to maximize the memory performance...

Now I'm running the system as follows:

CPU: 4Ghz (8x), Auto Voltage, No LLC
FSB: 500Mhz
Memory: 2000Mhz on 9-11-9-24 CR1 1.7V, performance level 'turbo' as labeled in the bios (main setting of turbo should actually affect the static tRead value and it's 9)
MCH: 1.6V core, +50ps clock skew

At these settings, the memory latency is 61.5ns
Memory read is ~13300MB/s
Memory write is ~10600MB/s
These are actually the best results I've seen on this system thus far...

It seems OK under windows, but still haven't tested it with Prime95...to be honest, I'm kinda afraid of the disappointment that may follow
Any advice on Clock Skewing?
I understand that it delays the signals between the components on the motherboard and the clock generator but I can't really get how does it work exactly...I mean it is very unpredictable.
For example, 480FSB doesn't work with any skew setting but 500 works fine with +50...

 

[ATMINSIDE]

I'm not sure about the skew setting at all.

 

[neverwind]

Is there anyone else here who can help me out on this one?

 

[neverwind]

A bit of an update.
Since now my memory is running in synchronous/symmetric mode (FSB : DDR = 1 : 2), it seems the MCH doesn't really need the extra voltage anymore. It runs fine with it being set on auto

I still need some help with clock skews, though...

 

[PolRoger]

Sometimes I still run a similar Gigabyte P45 chipset motherboard...

On my board I find that the 400fsb strap with the 3.33D memory divider allows for good overclocking on the fsb of C2-Quads. I'm not really sure which fsb strap and memory multi(s) you have already tried to run and test??

You should download and install Gigabyte EasyTune 6 for your setup. It will show you what voltages the BIOS "auto" rules are using if run a particular voltage setting on auto. It will also allow you to do on the fly voltage adjustments in windows to help with dialing in a particular overclock.

GTL refs and clock skews can both help refine overclocked settings... Lately I've mostly left them on auto. Back in the day when 775 was current it was much easier to find out info about how to best tweak them. 775 is pretty old tech now and most have moved on to newer platforms.

 

[neverwind]

Thanks for your input, PolRoger.
I've tried all of the straps and on 500FSB the only one that works fine is 1:2 (that's 4.00D in BIOS).

Unfortunately, ET6 doesn't allow me to see or tweak any voltages on my board, just CPU multiplier and other frequencies. Frankly, the application seems to behave very strange and is kinda 'buggy'. I guess it's the board.
I see you have the extreme version of Gigabyte's P45 DDR3 series, which is the highest class, so generally it should be able to overclock a lot better than mine
Any chance to know the limit of MCH Core Voltage?

I know 775 is an old setup but still it works fine for me so I'm pretty skeptical about moving on. Plus, I really need WinXP to be supported by my platform (not just for benching, but for everyday usage) and I'm not sure I can run it on newer platforms. I'm considering a z77 based system since I know it's a 100% XP compatible, but still I don't really need to change my current one yet