Tips for improving overclock?

[Cybercat]

I got a 1700+ at 2.1GHz, not extremely stable (can't run the first Prime95 test). It's more like 2082MHz (181x11.5) but when Windows boots it adds another MHz to the FSB and makes it 182x11.5. *shrug* Anyway, my HSF isn't exactly great, but sufficient for the 2600+ speeds I have it at (40ºC idle, upper 50s load). My PSU is also crappy, as the -12V rail runs at over -13 all the time, and my VCore fluctuates while playing games, etc (that's not normal, right?). Anyway, my PSU will be replaced soon, but what else could I do to improve my OC, and what may be contributing to my unstable/disappointing OC right now? My Vcore right now is set at 1.85V btw. CPU stepping is JIUCB DUT3C. RAM VDIMM is at default 2.6V.

 

[myocardia]

Buy some ram that Athlons like, like either OCZ LL, Corsair XMS, or Kingston HyperX. You're the second person in as many days that I've seen with Geil ram, who can't get over roughly 180fsb.

 

[Cybercat]

Reeeaaally? Now that's something I haven't heard before... It's odd though, I should be able to get higher, considering I'm only running my RAM at 364MHz, when it's DDR400 at stock... I should then at least get to 200FSB, while running at 1:1.

 

[myocardia]

Originally posted by: Cybercat
Reeeaaally? Now that's something I haven't heard before... It's odd though, I should get higher, consider I'm only running my RAM at 364MHz, when it's DDR400 at stock...

Yeah, that's exactly what the other people who try to use Geil ram with an Athlon system always say!

 

[Cybercat]

lol, perhaps, but right now RAM's out of the budget (no job ), and if I'm going to upgrade it, I'm gonna get something good
like some of these. *drools at stock timings*

 

[myocardia]

Yeah, OCZ is the ram to buy if you want to save money, but still want fast ram in an Athlon system. Has anyone told you that dual-channel makes hardly any difference with Athlons? I would buy the 512MB stick that newegg sells for $105 here: link. Plus, if you buy one 512MB stick, you can add another later. Alot of nForce2 boards don't handle three sticks of ram very well.

 

[Cybercat]

I'm pretty sure it does make a difference, because while the Athlon XP doesn't initially support dual channel RAM on the CPU itself, the NB chipset will split the data and run it on both channels. Like having RAID0 on a pair of HDDs. I don't have any benchmarks or comparisons showing the difference with it enabled or disable with me, but for running those RAM intensive stuff, such as video editing and ripping DVDs, I'm pretty sure there's a difference.

And yes, it would have trouble running three DIMMs, as it's max stable speed for having three DIMMs is said to be DDR333 according to the manufacturer. With that being said, I think it would be wiser to get two 256 sticks and keep the dual channel despite the small $12 savings.

 

[myocardia]

Anandtech needs some better servers, obviously!!

 

[myocardia]

Psada did a whole slew of tests, mostly with video encoding, and there was 2.5%-7.5% gain when comparing dual-channel to single channel. You would get MUCH more than that, by having 2x512MB running in dual-channel, as you will see here: link.

edit: Plus, I mispelled Pspada's name the first try!

 

[myocardia]

Pspada did a whole slew of tests, mostly with video encoding, and there was 2.5%-7.5% gain when comparing dual-channel to single channel. You would get MUCH more than that, by having 2x512MB running in dual-channel, as you will see here: link.

 

[Jeff7181]

Originally posted by: Cybercat
I'm pretty sure it does make a difference, because while the Athlon XP doesn't initially support dual channel RAM on the CPU itself, the NB chipset will split the data and run it on both channels. Like having RAID0 on a pair of HDDs. I don't have any benchmarks or comparisons showing the difference with it enabled or disable with me, but for running those RAM intensive stuff, such as video editing and ripping DVDs, I'm pretty sure there's a difference.

And yes, it would have trouble running three DIMMs, as it's max stable speed for having three DIMMs is said to be DDR333 according to the manufacturer. With that being said, I think it would be wiser to get two 256 sticks and keep the dual channel despite the small $12 savings.

IIRC, that's not really how the socket 462 implimentation of dual channel memory works... which is why you don't see the performance gains. I believe the method Intel used in the 865/875 chipsets is more like your RAID 0 analogy. I wish I could find where I read about the differences... but I'm pretty sure the dual memory controllers for AMD systems don't act exactly like that. It's more like, the 2nd memory channel can prepare for another data transfer by activating rows and such, but actual data transmission can't occur simultaneously, which is why even a dual channel Athlon XP system can't transfer data to/from memory at greater than 3.2 GB/s. It only gets closer to that maximum because the two memory channels work to reduce waiting because when one controller has activated a row and it's transferring data, the other is preparing for a data transfer and as soon as the first data transfer is done, the 2nd conroller already has a row and colum activated and can start transferring data right away.

I THINK that's how it works... if somebody knows differently, please correct me.

 

[Cybercat]

IIRC, that's not really how the socket 462 implimentation of dual channel memory works... which is why you don't see the performance gains. I believe the method Intel used in the 865/875 chipsets is more like your RAID 0 analogy. I wish I could find where I read about the differences... but I'm pretty sure the dual memory controllers for AMD systems don't act exactly like that. It's more like, the 2nd memory channel can prepare for another data transfer by activating rows and such, but actual data transmission can't occur simultaneously, which is why even a dual channel Athlon XP system can't transfer data to/from memory at greater than 3.2 GB/s. It only gets closer to that maximum because the two memory channels work to reduce waiting because when one controller has activated a row and it's transferring data, the other is preparing for a data transfer and as soon as the first data transfer is done, the 2nd conroller already has a row and colum activated and can start transferring data right away.

I THINK that's how it works... if somebody knows differently, please correct me.

Well if that's the case, they lie when they say that the motherboard has a maximum bandwidth of 6.4GB/s in the manual.

 

[myocardia]

Maybe so, but who says that an Athlon XP has 6.4GB/second bandwidth. Just because one component has that much doesn't mean every component in the system has it. All I know is you would be much better off buying one 512 stick, and adding another identical stick later, especially if you are planning on doing video encoding or editing. Nothing that I know of requires more memory than those do.

 

[Jeff7181]

I found a site that does some pretty good tests. One thing to keep in mind is that the CPU's 400 Mhz effective bus can ONLY move 3.2 GB/s. So the CPU doesn't need any more bandwidth... but... since lots of other things use DMA, the 2nd memory channel can serve those other devices, and still provide the CPU with the full 3.2 GB/s that it can use.

link

 

[Cybercat]

Originally posted by: myocardia
Maybe so, but who says that an Athlon XP has 6.4GB/second bandwidth.

The user manual for my motherboard...

 

[Cybercat]

Anyway, back on subject. Besides my budget RAM, what else could be contributing to the limited overclock? Someone once told me that the HSF could be seated wrong. As far as I know, it's seated just fine, but I'd like to read your thoughts on it. Could that be a possibility?

Also, could the power supply which will run the -12V rail at over -13V and push the Vcore up pretty far when running games and benchmarks have anything major to do with OCing potential?

 

[myocardia]

Originally posted by: Cybercat

Originally posted by: myocardia
Maybe so, but who says that an Athlon XP has 6.4GB/second bandwidth.

The user manual for my motherboard...

The user manual for your motherboard gives specifications on Athlon's? Wow, all of them? E-mail it to me, why don't you?

 

[Cybercat]

I'll let you be the judge of what it's saying.

Quote from manual:
"2-2 Specifications

3. Dual DDR
-Dual DDR architecture combines two independent 64-bit memory controllers
-Dual DDR400 provides up to 6.4GB/s bandwidth for system memory"

 

[WA261]

Ram, Cooling, PSU....you need an upgrade on these items. Both of my 1700's run at 2.4 on air. BUT I use Antec PSU's, Corsair ram, and MCX462 sinks.

 

[Cybercat]

Oooo, I really like that heatsink. :heart::heart: But I'm thinking of settling with the Volcano 12+ and using the temp sensor to control its speed. That's a big@$$ fan right there. I don't really mind the noise. My comp is at least 40dB as it is.

Oh, and what kind of Corsair are you using on those 1700+ processors? Value or XMS? My friend got some Value 512MB PC2700 and it OCed surprisingly well.

 

[myocardia]

Originally posted by: Cybercat
I'll let you be the judge of what it's saying.

Quote from manual:
"2-2 Specifications

3. Dual DDR
-Dual DDR architecture combines two independent 64-bit memory controllers
-Dual DDR400 provides up to 6.4GB/s bandwidth for system memory"

Okay, that isn't even saying that your motherboard can move 6.4GB/second through the memory controller! That statement is quite true, though. PC3200 ram can move 3.2GB/s in single-channel mode. So, the ram itself can transfer up to 6.4GB/s, since you must have 2 sticks of ram to run dual-channel. But your motherboard manual makes no mention whatsoever of either your motherboard or an Athlon XP being able to transfer data that fast. That's what we call "marketing".

And your vcore (processor voltage) going up can't be possible for it failing Prime95, nor can it be possible for a raised vcore to contribute to a failed overclock. Raising your vcore is what is required to overclock a processor more than 5-10%, depending on the processor. I think it can't be anything other than your ram, or your power supply that needs to be upgraded. Either buy or borrow one or the other, so you can find out for sure which it is.

 

[Cybercat]

Obviously the Vcore must be raised in order to OC, but I was under the impression that voltage levels should remain within a certain margin and to be stable like that, and not fluctuate to higher levels. Going up is alright I'm sure, the more the merrier. The only concern is the temps, but that's one of the things that needs to be taken care of anyway. I've thought for a long time though that the PSU was one of the main contributors, and I'm basically just looking for evidence as to what specifically makes it cause the system to be unstable. I suppose raising the Vcore, even if it is on its own, can only help stability, but oh well. I already know the PSU sucks, who cares about evidence. That's getting replaced soon for certain, although not with an Antec PSU, but with a decent one nonetheless. It does come with dual fans (wow), which helps because PSUs I've worked with (mostly crap) that have had only one fan were hot like little ****s. That didn't help system temps either.

Everyone, thanks for your help, I greatly appreciate it! The boards I come from aren't nearly as knowledgeable as you people, and I hope to become a regular around here and learn as much as possible along the way.

 

[myocardia]

Okay, one last thing that I forgot to say in my last post. The more expensive Thermaltake's, like the Volcano 12+, are about 90% as good as the $50 ones. That's pretty good performance for the huge savings. If you think your temps are inhibiting your OCing, just run it with the side of the case removed (open, anyway). If you want to make sure that it isn't heat-related, set a large fan blowing into the case. Unless you have some awesome case flow, your temps will drop considerably.

 

[Jeff7181]

You could use Memtest86 (google it) to test the memory and see if that's the problem. Prime95 will stress the CPU pretty good, but I've also noticed RAM errors show up with Prime95 as well. If you want to make sure it's not the motherboard chipset holding you back, set the FSB back down to stock speed and raise the multiplier to get near the same clock speed. If it runs fine there, it may be that you got a motherboard that sucks for overclocking. If your RAM turns out to be the problem, try setting the timing to something like 2.5-3-3-7. I've read that a lot of times, the nForce2 chipset doesn't like low latencies on certain RAM... my PalmTreePC RAM being one of them... (my friend bought the same RAM at the same time as me and he's running it at 210 Mhz @ 2-2-2-5 on an Intel 845 chipset. Mine won't even run 166 Mhz @ CAS 2)

 

[Cybercat]

Originally posted by: myocardia
Okay, one last thing that I forgot to say in my last post. The more expensive Thermaltake's, like the Volcano 12+, are about 90% as good as the $50 ones. That's pretty good performance for the huge savings. If you think your temps are inhibiting your OCing, just run it with the side of the case removed (open, anyway). If you want to make sure that it isn't heat-related, set a large fan blowing into the case. Unless you have some awesome case flow, your temps will drop considerably.

To tell you the truth I don't think heat is a main limiting factor with my OC, which is why I'm getting a new PSU first. Simple reason is because I've done that very thing: take a house fan and put it on high blowing directly into the side of my case.

You could use Memtest86 (google it) to test the memory and see if that's the problem. Prime95 will stress the CPU pretty good, but I've also noticed RAM errors show up with Prime95 as well. If you want to make sure it's not the motherboard chipset holding you back, set the FSB back down to stock speed and raise the multiplier to get near the same clock speed. If it runs fine there, it may be that you got a motherboard that sucks for overclocking. If your RAM turns out to be the problem, try setting the timing to something like 2.5-3-3-7. I've read that a lot of times, the nForce2 chipset doesn't like low latencies on certain RAM... my PalmTreePC RAM being one of them... (my friend bought the same RAM at the same time as me and he's running it at 210 Mhz @ 2-2-2-5 on an Intel 845 chipset. Mine won't even run 166 Mhz @ CAS 2)

I used Memtest86 as soon as I got my RAM to make sure it wasn't erroneous. I've also tried lowering my timings way down (2.5-4-4-8). I haven't tried lowering the FSB and raising the multi though, so I'll do that. Thx