No luck o-clocking Athlon 64 3000+ Venice on Abit AV8

[Combatter]

Hi all,

I'm really struggling to overclock my Athlon 64 3000+ (venice core) despite reading loads of articles saying how easy it is to do.

HEXUS article

I have an Abit AV8 3rd Eye motherboard (VIA K8T800Pro) with latest BIOS. I'm using an Arctic Cooling Freezer 64. I've done the following...

• Set AGP/PCI buses fixed at 66/33mhz
• Set the "LDT bus frequency" from 5x to 3x (I assume this is the same as the 'HT Frequency'?)
• Set DRAM to 266mhz but with timings as SPD for 400mhz
• Set Nothern Bridge voltage to 1.65v (default 1.55v)
• Set Southern Bridge voltage to 2.60v (default 2.50v)

At first I was having troubles just moving up to 210mhz. I eventually managed it by having just one of my 512mb RAM modules in and by setting the "LDT+PCI bus control" page of my BIOS to "Fail Safe" settings (keeping LDT bus frequency to x3). This fail safe thing did the following...

Set upstream and downstream LDT bus widths from 16bit to 8bit
Set PCI1 Post Write and PCI2 Post Write from enabled to disabled

I reached 250mhz (CPU at 2250mhz) with VCore at 1.6v (def: 1.4), DDR at 2.7v (def: 2.65v), HT at 1.3v (def:1.2v) but now my 3DMark05 score has actually dropped from 4156 to 4005.

Am I missing something really obvious here? Any help would be greatly appreciated.

Thanks in advance,

--
Ben Kennish

PS: Can anyone explain to me what "LDT bus frequency" and "HT Frequency" mean? Why is it recommended to reduce from 5x to 3x?

[Combatter]

Update:

Last night I realised that when I set the memory in the BIOS to DDR333 (for example), what it actually does (according to CPU-Z) is set the memory speed to a fraction of the CPU speed. e.g. DDR333 = 166mhz bus = 1800/11 (ish). So setting to DDR333 in BIOS sets the memory speed to CPU speed/11.

So if I select DDR333 in BIOS but push the external clock to 244mhz, this sends CPU speed to 2200mhz and the resulting DDR frequency will be 2200/11 = 200mhz (400mhz DDR). I have also set the "LDT Bus frequency" to 'x4' which gives 244x4 = 976 which is close to the default 200x5 = 1000.

I'm going to experiment further today...

[cptwhite_uk]

It is a pain in the arse I'm not experienced enough to overclock properly and managed to corrupt my windows installation somehow! I just changed everything back to auto and am now getting 2.44Ghz from my (default 2.2Ghz) Athlon 3500+. I'm happy enough with that.

[sawyen]

I have an AV8 but I'm using a 3500+ instead..

I have done a fair bit of experiments on this mobo and I can tell you this..

1. Its best not to use HT 3x, while rather just stick with 4x if you can or just drop it to 2x as it wont hurt much anyways. some AV8s seems to hate the 3x multi and always causes unwanted crashes or memory corruptions during gaming.

2. You wont need to pump NB voltages until you're thinking of breaking around the 270MHz mark. I run 9 x 290 before I only used 1.6V on NB.

3. SB only strengthens ur AGP lock and other IDE/SATA pheripherals under high HT load, I find just leaving it at default should work for any OC below 270MHz

4. HT voltage will help significantly to your HT overclock once you hit mobo ceiling, I find upping it to 1.3V concurrently with memory voltage actually helps by about a few MHz when you've already hit chipset ceiling.. 1.35V + 2.8V at 295MHz HTT

5. Try using 74MHz lock instead, the 66MHz lock (for some reason) is very prone to failures.. I found most of the ppl I know using AV8 seems to work better with 74MHz.

6. Using dividers will definitely lose performance, as lower memory frequency means lower performance. running 245MHz at 400Mhz is definitely faster than 245MHz at 333MHz divider.. you just need to know your memory's limit. Try reading some articles on memory ceilings.

[Combatter]

Its best not to use HT 3x, while rather just stick with 4x if you can or just drop it to 2x as it wont hurt much anyways. some AV8s seems to hate the 3x multi and always causes unwanted crashes or memory corruptions during gaming.

Cool. Perhaps using x4 rather than x3 was the reason for my success in the second post.

You wont need to pump NB voltages until you're thinking of breaking around the 270MHz mark....SB only strengthens ur AGP lock and other IDE/SATA pheripherals under high HT load, I find just leaving it at default should work for any OC below 270MHz

Thanks... yes setting these both to default seems to have caused no problems. The only reason I used them was that I copied the voltages from the 'Turbo' uGuru preset (like the overclocking newbie that I am )

HT voltage will help significantly to your HT overclock once you hit mobo ceiling, I find upping it to 1.3V concurrently with memory voltage actually helps by about a few MHz when you've already hit chipset ceiling.. 1.35V + 2.8V at 295MHz HTT

Are you saying that I should set both HT voltage and DDR voltages back to default (1.20v & 2.65v) until getting up to around the 290mhz mark?

Try using 74MHz lock instead, the 66MHz lock (for some reason) is very prone to failures.. I found most of the ppl I know using AV8 seems to work better with 74MHz.

OK I'll give that a try. I was reluctant to do that for fear of frying my AGP / PCI cards (1xGeforce 6800GT, 1xSoundblaster Audigy 2 ZS); am I being paranoid?

Using dividers will definitely lose performance, as lower memory frequency means lower performance. running 245MHz at 400Mhz is definitely faster than 245MHz at 333MHz divider.. you just need to know your memory's limit. Try reading some articles on memory ceilings.

CPU-Z is currently reporting that my memory frequency is at 200.5mhz (400DDR) even tho the DRAM is set to 333DDR in the BIOS. I have my theory that this is because the DRAM setting in the BIOS sets the DRAM frequency to a fraction of the CPU clock whilst at the same time assuming that this is at the default (i.e 1800mhz). The reason I came to this conclusion is that CPU-Z quotes "FSB : DRAM = CPU/11" and sure enough, if I use uGuru to overclock on the fly, the reported RAM frequency goes up as the CPU speed does.

See my CPU-Z screenshot

[sawyen]

leaving at 1.2V should be fine to about 275mhz (at least mine did).. once u hit mobo ceiling or memory ceiling.. whichever comes 1st.. u will have to up them together.. when u up DDR voltage, HT has to be upped as well... I found out thats the most stable way of handling Abit's AV8.

Dividers are just well.. 'dividers'... when u place them at 333MHz, it just means it will run at 166MHz when you're at default 200MHz HTT.. but when u up HTT, the memory clock goes up as well; under your specified divider.

[kevpuk]

Dividers are the key.....especially with 'value' RAM as it will allow you to get the CPU frequency (most important !) and scales the RAM up accordingly.
Personally, and it depends on your target, I like to keep all voltages at stock as I feel this is 'easier' for the components and does not subject them to 'extra' stress......once you start on the voltage trail, the risk of damage increases, although some very impressive results can be had......

[Combatter]

What troubles me is that, regarding voltages, I can't even get up to 2187mhz (243mhz external clock) without upping the CPU core voltage to around 1.575v. If I set it any lower, I get graphics anomolies in Half Life 2

[kevpuk]

'spose that could just be an 'unlucky' chip......I have had mine up to the 2.4GHz mark with no voltage changes, but preferred a 'lower stress' on the system and backed down to 2.3GHz....it does seem like a bit of a lottery whether one gets a 'good un' or not......although none of the above probably helps you...anyone else ?

[taqie]

yerp...perhaps it just a bad chip...care to tell what is your memory...perhaps its just your memory...if you dont want the hassle of calculating what your ram speed is....here is the tool to calculate it
http://forums.extremeoverclocking.co...d.php?t=149717

basically actual ram speed = HTT(divider/200)....

[chriswood_7]

Ive had mine stable at 2.3GHz but when it was hot in my room my temps went right up! Personally i dont think id go back with a via chipset as the nforce based boards seem to be much better for overclocking performance. Id try your LDT frequency at both 3x and 4x and see how far you get. Use superpi on a 1million run to check stability and also use 3dmark2001Se a few times

Chris

[Combatter]

Well it's almost certainly not my RAM causing the problems. At 2187mhz (what I'm running atm but have to increase the core voltage to get), the memory is at 199.2Mhz (according to CPU-Z) and I've run memtest86 for 6 hours solid with no problems.

Oh and my RAM is 2 modules of Crucial 576MB CAS-3 stuff running with the SPD timings for DDR400 - 3.0-3-3-8 (with the BIOS set to DDR333 because of the increased external clock)

[arthurleung]

Try push it down to 266 divider and see if it helps?