Another Q9950 & UD3P Overclock

[Conners]

1020 is correct Jim, my bad for not putting it in the template. I'll correct it if I can in the post.

I don't know any thing about them either, Intel says max voltage is 1.36xx and max temp is 71.4 if thats helps you any.

Anyway here's the screenie. And thanks for your help.

[profJim]

After my post, I searched the forums for Q9550 and found a thread where they're talking about 4.5GHz, but you'd need to use extreme cooling that would cool your cpu to 1 degree above absolute zero!

It looks like you might be able to clock your cpu to 450 MHz with air cooling. From your screen shot, we can definitely improve your memory timings by lowering tRD and tRFC.

Let us know when you're ready to go off the deep end.

[Conners]

Thanks Jim, lets play with the tRD and tRFC before I go higher.

Let me know what I need.

[kick]

There is an epp for your memory at 1066/533.You can use that .
If you enable the xmp in bios.
What you have to decide is how you reach that speed.
The lower your fsb /actual cpu reference speed is the less v you should need for cpu and mch.
I suggested 2.4 B because it is the lowest multiplier (aside from 2:00,1:1).

Id like to say you could use a 2:00 multiplier either B or D but I havent seen many/any quads running with a 533 fsb/cpu ref frequency.

You have plenty of scope on your cpu multiplier to get a safe fsb for an approximate 1066 memory speed.
If you decide to use the xmp profile in the bios the mbrd should set the timings and mem voltage automatically for you.
All you will need to be concerned about is :the memory multiplier ,the cpu v and the mch voltage and which multiplier you use on the cpu(probably).
1020 to 1066,well u know the maths..46 divided by 2.4 = about a 18 mhz fsb/cpu reference frequency increase,bringing it to 443 approx.

Or you could try the 2.66D multiplier with an appropriate fsb/cpu ref frequency to get the memory to 1066.

I personally would go for the highest stable fsb/cpu reference frequency I could get,while getting the cpu to its reasonable max,taking temps/voltages etc into account.

If I had a 9550, with memory that had a 1066 profile ,id be somewhere in the 440-480 cpu/fsb ref frequency region.

As it is Im pushing it a little on air, 490 would be pretty safe 500 is for neatness for a 24/7 rig.

[profJim]

Don't use XMP profile!!
In another thread, (using Corsair DDR3 memory) XMP = enabled lead to mem voltage = 2.20 volts, which is over spec. The only advantage to enabling XMP is that you don't have to manually set the primary memory timings and voltage. Once we're stable with tight memory timings, we can test to see if enabling XMP offers any benefits.

tRD & tRFC:
tRD offers the biggest potential memory performance gains, so we'll start with tRD. tRFC tweaking will follow, once we have the best tRD setting.

From Conners most recent template, it's clear that the tRD AUTO setting is conservatively setting it to 10. I recommend jumping to 8 because of the current 425 MHz frequency. Normal procedure is to take it one step at a time [i.e. 9, test, then 8, test], but I'm willing to bet Conners mobo that 8 will work and won't send his mobo into a reboot loop. Prudent practice would be to lower tRD to 9, and then proceed with thorough stability testing. If Conner doesn't mind, it means that he will have to 2 series of stress testing to reach the 8 level. tRD = 7 might be stable, but in all probability we will have to raise it back up a bit when we get around to raising cpu frequency to the 440 MHz (or higher) range.

Conners, if you're not familiar with running Memtest86+, Prime95, OCCT, Orthos, or BurnTest, let us know. Don't worry, you don't have to run all these tests.

Math Formula for Overclocking

It doesn't exist!! There are general guidelines that are refined over time based on experience. AnandTech and others have written many excellent articles that deal with memory and cpu overclocking, and have developed formulas that deal with a specific cpu, on a specific mobo, with several brands of memory. Great for guidelines, but that's it.

A Roll of the Dice

The final overclock will depend on how "good" Conners components are. If he got a really "good" cpu/mobo/memory combination, we'll be able to clock higher than if we're working with average or below average parts.

Conners, if you're not familiar with saving and loading bios profiles, speak up. It's a great tool to use when testing overclock settings.

CPU MHz

Kick has it right, we'll probably end up using the 2.40B or 2.66D memory multiplier with a final cpu speed in the mid 400 MHz range. If we clock the cpu to the high 400 MHz range, we'll have to use one of the 2.00 multipliers.

Disclaimer

I haven't overclocked any quads, but from what I've seen on the forum and my previous experience we should do just fine. As always, feedback from others is encouraged.

In your next post, include a link to your memory's specs.
It would be helpful to run synthetic memory benchmarks and to submit the results for others to see what gains can be made. Obviously, the real bottom line is how much faster and better your system runs.
Clear as mud? OK, let's lock & load....

[Lsdmeasap]

^^ Good info there ^^

Let me know if you guys need any further help here

[Conners]

Wow guys thats more information than I can process at one time... Nice job! Ok where to start..

1020 to 1066,well u know the maths..46 divided by 2.4 = about a 18 mhz fsb/cpu reference frequency increase,bringing it to 443 approx

Kick, see I don't know the math I have never done this in my life but there's got to be formulas you guys use. But I think I can see what you did in the example above. I take it as the diffrence between the memory speed - the memory multiplier = mhz fsb/cup. Then add that sum to the current fsb/cpu to get the new fsb/cpu. Then the 443 x 4 (quad core) = 1772 fsb/cpu. So that extra fsb/cpu would make the memory speed 1066. I that correct?

From Conners most recent template, it's clear that the tRD AUTO setting is conservatively setting it to 10. I recommend jumping to 8 because of the current 425 MHz frequency. Normal procedure is to take it one step at a time [i.e. 9, test, then 8, test], but I'm willing to bet Conners mobo that 8 will work and won't send his mobo into a reboot loop. Prudent practice would be to lower tRD to 9, and then proceed with thorough stability testing. If Conner doesn't mind, it means that he will have to 2 series of stress testing to reach the 8 level. tRD = 7 might be stable, but in all probability we will have to raise it back up a bit when we get around to raising cpu frequency to the 440 MHz (or higher) range.

Jim, I believe I did have the tRD at 9 one time, but I will do it as you suggest and post the results after 9 then 8. Edit: Actually I haven't, I was thinking about the Static tRead Value. So is it the tRD Phase0-3 adjustment where you put 9? Edit2: tRD is static read so never mind.

Conners, if you're not familiar with running Memtest86+, Prime95, OCCT, Orthos, or BurnTest, let us know. Don't worry, you don't have to run all these tests.

I ran Memtest86, Prime95 and OCCT, but always at their default. The last two I am not familiar with.

Conners, if you're not familiar with saving and loading bios profiles, speak up. It's a great tool to use when testing overclock settings.

I've never actually done this but how hard can it be? Edit: I guess its the timing profiles under xmp. Edit2: Nope Bios profile F11.

In your next post, include a link to your memory's specs.
It would be helpful to run synthetic memory benchmarks and to submit the results for others to see what gains can be made. Obviously, the real bottom line is how much faster and better your system runs

Links to Ram

NewEgg Corsair Website

What do you mean by synthetic memory bench mark? Just lay it on me man.
Jim, yes the bottom line is how much faster and better my system runs but at the same time it doesn't have to break the sound barrier. Just a good 24/7 setup for gaming.
Thanks guys!

Oh and thanks for stopping in Lsdmeasp.

Added: BTW, I ran p95 for over 12 hours last night last thread (core 4?) failed after 12 hours and I noticed somting during the theat the vcore voltage hit 4.1. Yikes! Any ideas on that?

[Conners]

I ran this this morning after changing tRD to 9 and got an error on the 1st thread.

[kick]

On the memory multipliers on this board its reasonably straight forward.
Some boards show "dividers" that multiply the cpu ref freq after its been divided by another number or allow u to adjust memory speed independently in 1 mhz steps as long as you dont try and run the memory below the ref speed.
Here tho its cpu ref speed x(number)=the memory speed.
If you use 2:00 its twice the cpu ref speed ,if you use 3:20 its 3.2 times the cpu ref speed etc.
The only good thing about xmp setting is it will use the base memory timings and voltages stamped on your memory ,but there are also advanced timings on your memorys epp(enhanced performance profile).
The mbrd can see those ,but cpuz doesnt show them. To see what the are you will have to use a program such as Sisoft Sandra or everest.
Synthetic memory benchmarks are available in the two programs listed above.Another memory bandwith/latency program I use is "cacheburst32".
They measure your memorys performance(bandwith/latency) at different speeds and timings."They wouldnt be much use if you never changed the memory speed/timings kick "
They are different to running say a real program like ,I dunno crysis very gx intensive part ,or some video conversion. A lot of this stuff is covered in the stickies above^
The overclock"How I did it" by Baron Frankenstein etc.

I noticed somting during the theat the vcore voltage hit 4.1. Yikes!

Im not sure what that means 4.1 core voltage there are only 2 core voltages shown cpu and memory,well i suppose mch is a 3rd but at 4.1 v theose items are gonna smoke /fail.
Its not 1.4 is it for cpu or mch?.If its those it might be a little high but not scary.

edit :your screenshots popped up while i was typing the above.The system/case doesnt seem super cool,whats the room temp?
You do seem to have lots of headroom on the cpu volts/temperatures still.
Id suggest playing around with settings to make sure your memory will run at 1066 first while keeping the cpu as low as you want.
Keep cpu voltage in bios below 1.4,memory voltage in bios below 2.5,and mch v below 1.4 while seeing what it will do/at what speeds/temps and for how long and you shouldnt damage anything.
Ive been running my sig specs for approx 2 months at at least those speeds /voltages,I havent done any overnight testing in awhile tho.
Once you get something stableyou can adjust the myriad of other settings

[Conners]

LOL kick I must be deslexic, it should be 1.4....
Thanks for the input. Also I am rerunning P95 and with the tRD 9 and so far after almost an hour no errors. My scrennie above must have been a fluke. Then it's on to 8 tRD.