Help me OC a q9550 on a GA EP45-UD3L w/ 1066 DDRII

[Suftop]

oh I did not realize that by leaving them set to auto even if it showed them as 5, 5, 5, 15 they weren't actually set to that!

I just went through and made sure all of those, as well as the voltages are manually entered to stock settings, and 2.1v for ram.

Incidentally I was getting bluescreens when trying that small FFT.

Im about to try again now with it set back to the 450/1800 @3.6

 

[EarthDog]

Thats what CPUz is for. Thats what it was showing you all along where set.

Yeah, To get 3.8Ghz you will likely need a lot more Vcore and bump a little more in MCH, VTT (FSBt) and PLL. But give it a try with Vcore alone.

 

[Suftop]

wow its working now.

Must have been a problem with one of those voltages or RAM frequencies that I had not manually entered like you said.

Temps hovering around 57-60. Anything under 75 is "safe" you say?

I'll let this run for a bit, then see if I can't get it up to 3.8ish.

Ty again, you have been an enormous help.

One more question: Once I get to the clock I want, and everything is stable, should the SSTEP and C1E be re-enabled or just left off?

 

[MattNo5ss]

I still don't get the system mem. multiplier thing, and the dif. between 2.00a, 2.00b, or 2.66a, 2.66b, etc... Resulting frequency is the same it seems between a and b, but it says to the right that it has something to do w/ FSB as well?

The number 2.00, 2.66, etc. is multiplied by your FSB to get your RAM Frequency. That is why your RAM is at 900 when you're using 2.00b and 450FSB, b/c 2*450 = 900. As I said previously...

But I don't know what the a, b, c, or d means, or how/if they are different from one another.

The reason you want a 1:1 ratio while OCing is so your RAM isn't getting OCed with your CPU. This is b/c if you do get a crash while both are OCed, it could be hard to tell which is unstable. After the you've finished the CPU OC, then you can try getting your RAM back to 1066.

Incidentally I was getting bluescreens when trying that small FFT.

Im about to try again now with it set back to the 450/1800 @3.6

I would try increasing vcore before lowering the multiplier, if you haven't already.

 

[ravaneli]

RAM multipliers were confusing to me too, at first. And those sticky guides confused me even more. They have lots of information, but are rarely clear enough.

Here is the clearest I can do:

-- effective memory clock is actually double what you see in BIOS. Manifacturers advertise EFFECTIVE clock.
-- the fist option (a,b,c or d) is the STRAP TO FSB. This is the BASE, that you will multiply with the other number. For example, if you chose d, then you strapped your memory to UNITS of 400 MHz.

Here are a few examples to get it straight.

If you STRAP it to 400, and then use multiplier of 1, your EFFECTIVE clock will be 400x1x2 = 800 mhz.
If you STRAP it to 333 and use multiplier of 2, your EFFECTIVE clock will be 333x2x2 = 1333 Mhz.

This is assuming no overclocking. Usually memory doesn't overclock well, especially cheap memory. After all the Math, you want your effective clock to be at, or below factory clock. Even if it is exactly AT factory clock, your memory is working a lot faster because of the higher FSB, and if you want to improve stability you may slightly up the voltage or/and relax timings. Hope that helps

I have the same CPU and mobo as you : ) These are champions in overclocking. You will enjoy learning with them.

 

[Suftop]

How does this look

You will notice I changed the RAM away from 1:1 after things seemed stable.

No multiplier seems to let me get it to exactly 1066. It's 1080 I guess...Does that matter?

Or should I just leave it at 1:1 clocked down? I think that would be @900 in this instance right?

I am not sure which is best. (1:1, or ram running as fast as it's capable of)

I did try to set FSB to 533 and clock things down, (to make it easier to get RAM @ 1066) but it didn't seem to want to post no matter how high I raised voltages with the FSB that high...

Thanks once again for the help. I think I am making some definite progress!

 

[MadMan007]

Yeah you're not going to get 533FSB on a quad core CPU. How you have it is fine for the memory, most people will tell you to have the memory at a higher ratio like you have it even with loose timings like 5-5-5-15 versus running it slower at tight timings.

 

[Suftop]

How does this look

You will notice I changed the RAM away from 1:1 after things seemed stable.

No multiplier seems to let me get it to exactly 1066. It's 1080 I guess...Does that matter?

Or should I just leave it at 1:1 clocked down? I think that would be @900 in this instance right?

I am not sure which is best. (1:1, or ram running as fast as it's capable of)

I did try to set FSB to 533 and clock things down, (to make it easier to get RAM @ 1066) but it didn't seem to want to post no matter how high I raised voltages with the FSB that high...

Thanks once again for the help. I think I am making some definite progress!

Thanks!

Think I should try to get the RAM even higher then...like closer to 1333 even though it's only rated @ 1066, or should I leave things alone as they are?

still wish I knew if I was supposed to set that RAM multiplier to a, b, d, or d too =\

Anyhow, I know I gotta let this prime95 run for overnight again so I suppose I will do that with it's current settings (its already running now and has been since just before my last post)

hmm, I just took a peak at speedfan. One of my cores is like 77 and all 4 are showing 73-77.
Is that too hot?

 

[EarthDog]

That ram will absolutely not do 1333Mhz....

Contrary to popular belief, overclocking your ram like that really doesnt net you much if any real world performance. Run it 1:1 and tighten her down if you can.

 

[Suftop]

tweaked a few more things.
dropped RAM back down to 900 @ 1:1
tweaked voltages some more. Seems stable at 1.25 vcore. Everything else is at stock voltages (manually set of course!)

Are these temps acceptable while prime95 is running the Small FFT torture test?

Also, anyone know what Temp2 is? I've tried Google searches about it but am getting conflicting reports.

 

[ravaneli]

1:1 is better. You will be more stable and you won't lose any performance.

Memory faster that DDR2 800 doesn't bring performance benefits, as it was said already. It is only for benchmarks : )

I saw a gaming test with all kinds of memory, in Tomshardware. Identical system, they were just swapping memory. There was NO difference in frames per second. None, not even 1 frame.

 

[EarthDog]

I dont like Speedfan Suftop. I would get Realtemp or Coretemp like we mentioned earlier just to confirm its accuracy. But honestly, I would just ditch Speedfan and go with one of the two I mentioned.

@ Rav - there are some games that show more than 1 fps difference. It just depends on if its memory bandwidth dependent. Even with that in mind, I still wouldnt go above 1:1 unless Im benching.

 

[Boulard83]

1:1 is better. You will be more stable and you won't lose any performance.

Memory faster that DDR2 800 doesn't bring performance benefits, as it was said already. It is only for benchmarks : )

I saw a gaming test with all kinds of memory, in Tomshardware. Identical system, they were just swapping memory. There was NO difference in frames per second. None, not even 1 frame.

+1 !

This was with my computer @ 425x8.5:
I made some test and in pure bandwith, going from 870mhz(4-4-4-12) to 1047mhz(5-5-5-15) took my mem bandwith from something like 82gb/s to 88gb/s. 20% mem OC for ... 5% bandwith. Didnt tested with 3dmark06 or gaming but i dont think this can yeild real FPS gain ... maybe 1fps ...

And i needed to push my mem volts to 2.04v to run this speed.

Currently running 900mhz with 1.96v and 4-4-4-12 timing.

 

[qq]

I used to run my 4 sticks OCZ PC6400 RAM at 887mhz 4-4-4-12 @ 2.2v, but recently I switched to 1064mhz 5-5-5-15 @ 2.1v. In everest free edition, my memory read went from 8574mb/sec to 9082mb/sec so I thought I'm going to gain some points in games and 3DMark. So I tried to run 3DMark05 and did a CS:S stress test, I gain 10 points in 3DMark05 and 0.4fps in CS:S stress test woohoo I did a test run on GTA4, and the result is 0.2 fps more so high memory bandwidth does help in a way... I gain more fps by setting transaction booster to enable +3. I just RMA my old evga 9800gt back to newegg for a refund to get a biostar gts250 512mb $94.99 after $30 mail in rebate. I'm pretty sure that will give me even more fps boost for a $15 difference.

 

[Boulard83]

I used to run my 4 sticks OCZ PC6400 RAM at 887mhz 4-4-4-12 @ 2.2v, but recently I switched to 1064mhz 5-5-5-15 @ 2.1v. In everest free edition, my memory read went from 8574mb/sec to 9082mb/sec so I thought I'm going to gain some points in games and 3DMark. So I tried to run 3DMark05 and did a CS:S stress test, I gain 10 points in 3DMark05 and 0.4fps in CS:S stress test woohoo I did a test run on GTA4, and the result is 0.2 fps more so high memory bandwidth does help in a way... I gain more fps by setting transaction booster to enable +3. I just RMA my old evga 9800gt back to newegg for a refund to get a biostar gts250 512mb $94.99 after $30 mail in rebate. I'm pretty sure that will give me even more fps boost for a $15 difference.

Nice testing.

 

[Suftop]

Thanks guys, useful advice.

Question re: memory, and this is CPU section I know but I'd feel silly creating another thread in a dif section for the same overclock...

If I have it (RAM) clocked down to 800, running 1:1, should I change the timings to 4,4,4,12 instead of 5,5,5,15?

I tried, and in order for it to be stable I had to bump the Ram voltage to 2.18, and slightly increase the ich and mch. (2 notches each I think)

Is this a good thing to do? Right now its set like this and it ran prime on blend for the past 10 hrs or so.

As for the speedfan thing, thanks. I will download this other one you are talking about tonight.

Ty!

 

[EarthDog]

You wont notice a difference so 5-5-5-15 will be fine. What does the SPD tab on CPUz say about your ram at 400Mhz?

 

[Suftop]

I used to run my 4 sticks OCZ PC6400 RAM at 887mhz 4-4-4-12 @ 2.2v, but recently I switched to 1064mhz 5-5-5-15 @ 2.1v. In everest free edition, my memory read went from 8574mb/sec to 9082mb/sec so I thought I'm going to gain some points in games and 3DMark. So I tried to run 3DMark05 and did a CS:S stress test, I gain 10 points in 3DMark05 and 0.4fps in CS:S stress test woohoo I did a test run on GTA4, and the result is 0.2 fps more so high memory bandwidth does help in a way... I gain more fps by setting transaction booster to enable +3. I just RMA my old evga 9800gt back to newegg for a refund to get a biostar gts250 512mb $94.99 after $30 mail in rebate. I'm pretty sure that will give me even more fps boost for a $15 difference.

Hmm

I made my last post before seeing this...very interesting.

I suppose this is gonna be one of those things that are a matter of opinion huh.

I just want to me sure I have no bottlenecking going on.

 

[EarthDog]

It depends on teh game(s) like I said. Some RTS games that rely heavily on the CPU show noteable improvements.

Most though will not.

 

[Suftop]

You wont notice a difference so 5-5-5-15 will be fine. What does the SPD tab on CPUz say about your ram at 400Mhz?

well the memory tab shows the 4,4,4,12, but on the SPD tab it's got me confused...says 5,5,5,15 in that far right column?

Also, what is this 52 clocks tRFC? That sounds awful high...any way to tighten that one?