Hitting 2.75GHz on a 3500+ Venice

CitizenInsomniac

Limp Gawd
Joined
May 26, 2005
Messages
134
(sorry about the long post! I like to provide a lot of information in order to paint a clear picture before all the questions come in)

I'd like to hit 2.75GHz on my Athlon64 3500+ Venice CPU. So far I've been close, but unsuccessful, so I'd appreciate any advice on what approach I should take next.

My MB is an Asus A8V-E Deluxe. I'm using a Zalman CNPS7000B-AlCu HFT with stock paste. Unfortunately, ClockGen doesn't work with my mobo so I haven't had the luxury of testing my FSB/memory limits step-by-step from Windows. Instead, I've had to do it the old-fashioned way: change the FSB in the BIOS, boot to Windows, run tests, then do it all over again.

So far I've established the following:

* 255 MHz appears to be the max FSB clock with Vcore under 1.5V. In fact, I can get stable Prime95 operation at 255 @ 1.475V, but I can't get the tests to pass at 260 @ 1.525V. Can 5 MHz really make that much of a voltage difference?

* 212 MHz appears to be the max memory clock with timings set to 2.5-3-3-10. I'm puzzled by this because the moment I raise the mem clock above this, I get errors in Memtest #5 regardless of how relaxed my timings are. In fact, I have a feeling that I can probably get away with even tighter timings at 212 MHz - but once I go past that speed, it's all over.

* I can set the memory timings as low as 2-2-2-10 when mem clock is under 200 MHz, which is great news if I want to run 2750 MHz at a 6:5 memory ratio (196 MHz).

So while it seems that this is enough to hit 2750 at 11 x 250 MHz, I can't seem to get my system to run stable at this speed. The Memtest86 tests pass, but in Prime95 I get rounding errors after as little as 5 minutes sometimes. I tried raising Vcore as high as 1.525V, but I still saw errors. It would seem unnecessary to raise Vcore any higher because I can get 2500 MHz (10 x 250) to work on a low 1.425V.

So far my stable highs are:

2550 MHz (10 x 255 MHz, with HTT at 4x255 = 1020, RAM at 2550/12 = 212.5) @ 1.47V
2640 MHz (11 x 240 MHz, with HTT at 4x240 = 960, RAM at 2640/14 = 188.5) @ 1.425V

If I had to pick one of those 2 as my "consolidation OC award", I think I'd go with the former because of the higher memory bandwidth. But I'd really like to hit 2750 because everything seems to indicate it's possible. Any advice would be much appreciated.
 
My impression is that everyone on this board hates reading. :rolleyes: Sorry, I didn't include any shiny screenshots of CPUz, but all the important details are included above, I promise. :)
 
Hmm, turns out the RAM wasn't as happy at 212 MHz as I had thought. Memtest86 passed fine, Prime95 small FFT passed fine, but when I started running large FFTs - I got errors within 5 minutes. Same errors crept up during 3dmark2001 runs. It also might've been the overclocked HTT bus (1020 MHz) that was causing all this, but who's counting? :)

So looks like 2640 (11 x 240) is the max I can hit right know. Anyone have any suggestions? Read the opening post for all the details.
 
It's all about the pics. :D

Venice2640.jpg



As you can see, the memory is clocked at 188 MHz due to the 1/14 divider, but the timings are nice and tight at 2-2-2-10. The HTT bus is slighttly underclocked at 960 MHz, but I had no choice.

Is there a way to change the memclock divider to a custom value? 1/13 would be nice in this scenario.
 
I got my 3500+ Venice to 2750 w/ 1.42v. I can't get my chip stable after 260, but at 250 it runs great. Have you dropped your mem down to 2/1 to see how far you chip will go? Looks like you got to 255 @ 1.47v. I don't think there is a way to get custom mem. dividers, might be a good excuse to upgrade your ram. Not really sure what you're asking now...

Good luck w/ the OC and let us know how it goes.
http://valid.x86-secret.com/show_oc?id=20748

edit: That cpu-z link was at 1.52v. Turned out to not be prime stable, but that might have been my memory. Was running 1:1 255mhz (2-2-2-8).
 
Yeah, this is the part that puzzles me: I know my FSB can go up to 250 (I had great stability at 10x250), but when I try 11x250 - the system isn't stable. That's with 6:5 (aka DDR166) memclock. I could try tuning the memory down even more to 3:2 (DDR133), but running my memory at 166 MHz (133 * 250/200) seems counterproductive - I will lose a lot of memory bandwidth.

My RAM is Corsair TWINX1024-3200C2PT. This is supposed to be good memory - it's a bit disappointing that I can't push it above 210MHz. However, that should still be sufficient to hit 2750 at DDR166 where memory runs at 196MHz.

So I guess my question is: if my FSB can take 250 MHz, and my memory can take 200 MHz, and my system runs fine at 10x250 - why am I having trouble getting up to 11x250?
 
changing the multi to 11x from 10x is only effecting your cpu. 6:5 - 250mhz puts your ram at (if my math is correct) 208mhz. You should be able to get great timing at 208 w/ that ram so that is a good plan. Try 1.5 or 1.6 vcore, your chip can handle that unless your temps. get too high. Stay out of the mid-high 50C range and you'll be fine.

No point in running your ram at anything under 200mhz. You might have to loosen the timing up alittle to get past 210, but don't go under 200. You paid good money for that ram and you should make it work for you.

My suggestion would be put alittle more voltage to your chip and then run 250x11, 6:5 divider. This was my first A64 OC so I'm still learning too. :D

also noticed your chipset fan was near 9k rmp. Mine won't go over 3k and Smart Guardian is reading 40-42C on my chipset. :confused:

btw, i still OC the old fashioned way. I don't trust all these new fancy-pants softwares touching my bios settings.
 
anyone else got any suggestions? CitizenInsomniac seems to be doing everything I would do.
 
gnomepunk said:
changing the multi to 11x from 10x is only effecting your cpu. 6:5 - 250mhz puts your ram at (if my math is correct) 208mhz. You should be able to get great timing at 208 w/ that ram so that is a good plan. Try 1.5 or 1.6 vcore, your chip can handle that unless your temps. get too high. Stay out of the mid-high 50C range and you'll be fine.

I've had trouble with high temps in the past - to the point where I thought the sensors were bad (and I think that's still possible). When I had the CPU at 1.5V, I managed to get the temperature up to 63C on a particularly hot day!

I've taken steps to improve the temperature as much as possible since then. I switched to Arctic Silver 5 and I've strived to keep the voltage under 1.5V. The temps are much better now at 1.425V. I haven't seen it go above 51C and I expect that to improve as AS5 kicks in more (I've only had it on for <48 hrs).

gnomepunk said:
No point in running your ram at anything under 200mhz. You might have to loosen the timing up alittle to get past 210, but don't go under 200. You paid good money for that ram and you should make it work for you.

I agree, it is disappointing to see that RAM run at 188 MHz. The only reason I'm not kicking and screaming is because at least I'm running it at 2.0-2-2 now. What really confuses me though is that at 2640MHz the memory divider is 14 and not 13. It's weird. I think that's actually true at 2750 too - the RAM ends up running at 196 instead of 208, like one would expect. What gives?

gnomepunk said:
My suggestion would be put alittle more voltage to your chip and then run 250x11, 6:5 divider. This was my first A64 OC so I'm still learning too. :D

I'm game for giving it another shot, especially now that the CPU has been burned in a bit. However, if it doesn't work at ~1.525V with loose memory timings, I'm giving up because I truly can't think of anything else. Sigh. It's a shame when we have to settle for only 2640MHz of brute 64-bit processing power. What has this world come to? ;)

gnomepunk said:
also noticed your chipset fan was near 9k rmp. Mine won't go over 3k and Smart Guardian is reading 40-42C on my chipset. :confused:

My chipset fan is very small - no more than 1.5 inches in diameter. That's probably why the fan speed is so high.
 
Thanks for the reply, although you're right - I have already read all of that stuff. Thanks anyway though!

The linked Clockgen version is the one I tried. It doesn't work completely on my board. It only allows Vcore changes, but no FSB changes. The FSB overclocking section is grayed out. I tried following the tip on the Clockgen website, but it didn't help either.
 
Short update:
I tried hitting 11x250 yesterday again. Memory timings set to 2.5-3-3, I ran 20 cycles of Memtest86 test #5 - no problems there.
I set Vcore to 1.525V and ran Prime95 large FFT test. It ran fine for 6 hours and than halted with an error. Doh.
I raised Vcore to 1.5375V this morning and left it to run while I'm at work. We'll see how it does.

Isn't it strange than raising my FSB from 240 to 250 requires an entire 0.1V boost? I'd expect the voltage requirement to rise proportionally to the FSB speed, but it doesn't. Is that normal?
 
CitizenInsomniac said:
Isn't it strange than raising my FSB from 240 to 250 requires an entire 0.1V boost? I'd expect the voltage requirement to rise proportionally to the FSB speed, but it doesn't. Is that normal?

When my chip hit the wall, it hit it fast considering the voltage. I can boot at 2.8, but no matter how much voltage I add, it won't go any farther. I don't think the voltage is linear to the FSB. Going from 200 to 240 at stock voltage is pretty impressive in my book. I can boot at 250 w/ stock vcore, but needs a little more for prime stable.
 
CitizenInsomniac said:
I'd expect the voltage requirement to rise proportionally to the FSB speed, but it doesn't. Is that normal?
yes, that is normal. more voltage has deminishing returns ;)

and i'm going to re-emphasize, as more data really is handy. i know a lot of people have overclocked revE cpu's, but haven't posted to help out, and it's making me sad :(
second, post in here to help others make an informed processor choice.
 
have you tried altering the voltage on the ram? Maybe the increased ram speed is too much for the stock voltage on it.

And who cares about prime? ;) are your games/benches stable?

If your system reboots itself during one of these, it could be overheating/old bios/problem with video card/bad ram. In short, anything to do with the "loop error." Google that or look it up on microsoft.com (I think they have an article or two on it).

Is your AGP/PCI bus locked? If they are getting increased as you increase the "FSB" then it could result in instability.
 
(cf)Eclipse said:
yes, that is normal. more voltage has deminishing returns ;)

and i'm going to re-emphasize, as more data really is handy. i know a lot of people have overclocked revE cpu's, but haven't posted to help out, and it's making me sad :(
second, post in here to help others make an informed processor choice.

Thanks Eclipse, I will add data to the other thread.

BTW, I was reading one of your threads and found a link to this thread over at XtremeSystems about the different ways to set vcore voltage. I've been wondering about that because my Asus A8V-E Dlx allows me to set +100 and +200 mV gain on top of the "hammer" voltage. So I'm wondering if setting 1.3V+200mV will be any different than 1.4V+100mV. I'll try that when I get home.
 
Success!
I managed to run Prime95 large FFT test for 24hrs at 2750 MHz (11x250) at 1.5375V!

Well, it's good to be there, now the only problem is: how do I get my temps down? :) After 24hrs of P95 my CPU temp rose to 57C. Not catastrophic, but my concern is that the upcoming summer months will only make that worse. Yesterday the outside high temperature was only 63F. During the summer months the temps will go up into the 90s (F) - and that will surely add 5-6C to the system temperature. Not sure I want my CPU to melt in July. :)

I already have a good HTF (Zalman 7000AlCu) and good paste (AS5). AS5 was applied following instructions, the heatsink is making good contact - everything checks out.
 
Looks like I spoke too soon. I tried running Prime95 small FFT test at 11x250 and it failed after about 16 hrs. Damn. So close!

Well, I'll up the voltage to 1.55V and try again, but that's where I draw the line because this time around my CPU temps hit 59C. I don't want to go higher than that, particularly when I can achieve 100% stability at 2640MHz at only 1.4375V. It just doesn't seem worth it.
 
one thing you might want to try is put a cpu duct in to feed your zalmon cooler air. Should lower it a few degress, and that seems like all you need
 
I think I have found a new sweet spot. Instead of pushing my system to 2.75GHz (11x250), I've decided to try 2.7GHz (11x245). And oddly enough, the 5 MHz FSB spread makes a world of difference for stability.

I ran Prime95 large FFT for 32 hours without errors at 11x245 MHz @ 1.475V, 5:6 memory (192 MHz) @ 2.0-2-2-10 timings. I will run another Prime95 torture test, this time with small FFT, for 24hrs and if that passes - I think I have found my happy OC place.

It's also great seeing AS5 kick in. It's been about 1 week now since I installed it. Asus Probe has a nice temperature logging feature which lets me see the temperature history - and I could see the temperature drop from 56C to 52C over the course of 36 hours. Even if it was just the ambient temperature that did that - it's still cool to see the tempeture drop and not rise over a long period of max CPU work.
 
CitizenInsomniac said:
2550 MHz (10 x 255 MHz, with HTT at 4x255 = 1020, RAM at 2550/12 = 212.5) @ 1.47V

CitizenInsomniac said:
I think I have found a new sweet spot. Instead of pushing my system to 2.75GHz (11x250), I've decided to try 2.7GHz (11x245). And oddly enough, the 5 MHz FSB spread makes a world of difference for stability.

Something you may have overlooked -- the HTT typically doesn't like being over / at / too close to 1000 MHz.

I'm actually surprised you were stable at 4x255. Anyway, try dropping the multiplier down to 3x, then try your 11x250 again and see how that goes.

I apologize if you'd already considered / tried this -- it's just something that a lot of people tend to overlook.

Good luck!
 
Read the thread and did not see anything about NorthBridge cooling?

The fact that you hit a wall makes me wonder if a better cooling solution on the NB might help?

Also can you give the NB a little more voltage?
:)
 
jamori said:
Something you may have overlooked -- the HTT typically doesn't like being over / at / too close to 1000 MHz.

Surely you only meant to say "over", right? HTT is supposed to run at 1000 MHz, so 4x250 would've been right on the money.

jamori said:
I'm actually surprised you were stable at 4x255. Anyway, try dropping the multiplier down to 3x, then try your 11x250 again and see how that goes.

Yeah, eventually 10x255 (4x HTT) hit an error in Prime95 too and I didn't feel like raising the voltage any more so I gave up on that one. I've tried so many combinations in the last few weeks that it's become difficult to keep track of them all. Fortunately, I've been writing down most settings, perf numbers, stability info, etc. I might try 10x255 again, with slightly raised voltage. I can try lowering HTT multi to 3x, but that seriously underclocks the HTT bus. 1020MHz shouldn't be that big of a deal though, I'm pretty sure the HTT can take it.

A good question is: what is better, 10x255 (RAM @ 208 MHz 2.5-3-3-10) or 11x245 (RAM @ 192 MHz 2.0-2-2-10)? Here's how the two measure up in Sandra:

10x255 / 11x245
Dhrystones 11773 / 12438
Mem Integer 6285 / 5762
Mem Float 6223 / 5780
Combo 6810 / 6816

As you can see, CPU is faster in 11x245, but the memory in that setting gets its ass kicked by 10x255. The overall combo score (Cache & Memory benchmark) turns out to be nearly the same. I have read somewhere that Sandra's memory benchmark isn't all that helpful in real-world scenarios, though. I've read that tighter timings end up giving better performance even at slower mem clock speeds.
 
LTD speed doesnt matter at all. at 200mhz HTT your LTD is 2000mhz. Anything over 2000 and most board go crasy, make sure your dropping the LTD multiplier, i dropped mine from x5 to x2 cause like i said, it really makes ni differnce how low it is.

Also, have to tried changing your chipset voltage? Does your motherboard allow it? I was held back on my DFI board and it turned out that my chip set was not being supplied enought juice.

You should be able to push that chip a little further I would hope.

My 3000+ is doing 295x9 @ 1.52V just to give you a ball park. My xp-120 and panaflo med speed are letting it only get to 46C max load.
 
Back
Top